void BlockMatching::setImages(cv::Mat &imageLeft, cv::Mat &imageRight)
{
this->imageLeft = imageLeft.clone();
this->imageRight = imageRight.clone();
}
void HumanModel::faceOff(const cv::Mat& image, const std::vector<vec2f>& points)
{
face_texture = image;
// model face has been triangulated when loading, if has polygons.
// model use OpenGL coordinates, in which -Z forward, Y up.
std::vector<vec3f> verts;
std::vector<vec2f> face_texcoords;
#pragma warning (disable: 4305)
// vertex and corresponding texture coordinates, grabed by hand in Blender.
// Note that Blender use OpenGL coodinate system, texture origin is at bottom left.
verts.push_back(vec3f(0.03344, 1.65360, 0.14763)); face_texcoords.push_back(vec2f(0.631, 0.571));
verts.push_back(vec3f(-0.03354, 1.65374, 0.14678)); face_texcoords.push_back(vec2f(0.377, 0.571));
verts.push_back(vec3f(-0.0000, 1.65530, 0.15450)); face_texcoords.push_back(vec2f(0.504, 0.592));
verts.push_back(vec3f(0.04550, 1.65460, 0.14315)); face_texcoords.push_back(vec2f(0.667, 0.574));
verts.push_back(vec3f(0.01932, 1.65027, 0.14915)); face_texcoords.push_back(vec2f(0.589, 0.568));
verts.push_back(vec3f(-0.01932, 1.65027, 0.14915)); face_texcoords.push_back(vec2f(0.411, 0.568));
verts.push_back(vec3f(-0.04550, 1.65460, 0.14315)); face_texcoords.push_back(vec2f(0.333, 0.574));
verts.push_back(vec3f(0.0182, 1.6172, 0.1539)); face_texcoords.push_back(vec2f(0.574, 0.442));
verts.push_back(vec3f(-0.0182, 1.6172, 0.1539)); face_texcoords.push_back(vec2f(0.426, 0.442));
verts.push_back(vec3f(0.00000, 1.6145, 0.1730)); face_texcoords.push_back(vec2f(0.503, 0.447));
verts.push_back(vec3f(0.00907, 1.6037, 0.15788)); face_texcoords.push_back(vec2f(0.532, 0.405));
verts.push_back(vec3f(-0.00907, 1.6037, 0.15788)); face_texcoords.push_back(vec2f(0.468, 0.405));
verts.push_back(vec3f(0.0232, 1.5770, 0.15163)); face_texcoords.push_back(vec2f(0.575, 0.305));
verts.push_back(vec3f(-0.0232, 1.5770, 0.15163)); face_texcoords.push_back(vec2f(0.425, 0.305));
verts.push_back(vec3f(0.00000, 1.57779, 0.15502)); face_texcoords.push_back(vec2f(0.508, 0.225));
verts.push_back(vec3f(0.00000, 1.58820, 0.16580)); face_texcoords.push_back(vec2f(0.500, 0.356));
verts.push_back(vec3f(0.07360, 1.65420, 0.0936)); face_texcoords.push_back(vec2f(0.889, 0.594));
verts.push_back(vec3f(-0.07360, 1.65420, 0.0936)); face_texcoords.push_back(vec2f(0.111, 0.594));
verts.push_back(vec3f(0.0555, 1.5669, 0.1042)); face_texcoords.push_back(vec2f(0.824, 0.245));
verts.push_back(vec3f(-0.0555, 1.5669, 0.1042)); face_texcoords.push_back(vec2f(0.176, 0.245));
verts.push_back(vec3f(0.0403, 1.6178, 0.1487)); face_texcoords.push_back(vec2f(0.656, 0.434));
verts.push_back(vec3f(-0.0403, 1.6178, 0.1487)); face_texcoords.push_back(vec2f(0.344, 0.434));
verts.push_back(vec3f(0.0727, 1.6180, 0.0912)); face_texcoords.push_back(vec2f(0.916, 0.460));
verts.push_back(vec3f(-0.0727, 1.6180, 0.0912)); face_texcoords.push_back(vec2f(0.094, 0.460));
verts.push_back(vec3f(0.0674, 1.6900, 0.1101)); face_texcoords.push_back(vec2f(0.804, 0.721));
verts.push_back(vec3f(-0.0674, 1.6900, 0.1101)); face_texcoords.push_back(vec2f(0.196, 0.721));
verts.push_back(vec3f(0.0325, 1.67235, 0.15495)); face_texcoords.push_back(vec2f(0.622, 0.648));
verts.push_back(vec3f(-0.0325, 1.67235, 0.15495)); face_texcoords.push_back(vec2f(0.378, 0.648));
verts.push_back(vec3f(0.0000, 1.6740, 0.15900)); face_texcoords.push_back(vec2f(0.500, 0.656));
verts.push_back(vec3f(0.02380, 1.5391, 0.1280)); face_texcoords.push_back(vec2f(0.625, 0.112));
verts.push_back(vec3f(0.0000, 1.5340, 0.13860)); face_texcoords.push_back(vec2f(0.500, 0.089));
verts.push_back(vec3f(-0.02380, 1.5391, 0.1280)); face_texcoords.push_back(vec2f(0.375, 0.112));
verts.push_back(vec3f(0.0436, 1.7156, 0.1360)); face_texcoords.push_back(vec2f(0.662, 0.796));
verts.push_back(vec3f(0.0000, 1.7248, 0.1462)); face_texcoords.push_back(vec2f(0.500, 0.819));
verts.push_back(vec3f(-0.0436, 1.7156, 0.1360)); face_texcoords.push_back(vec2f(0.338, 0.796));
#pragma warning (default: 4305)
assert(points.size() == POINT_MAX);
assert(verts.size() == POINT_MAX);
// NOTE: image or cv::Mat is top down on Y axis, while OpenGL use bottom up Y axis.
float scale_x = (verts[LEYE_C].x - verts[REYE_C].x)/(points[LEYE_C].x - points[REYE_C].x);
float scale_y = (verts[EYEBROW_M].y - verts[NOSTRIL_R].y)/(points[EYEBROW_M].y - points[NOSTRIL_R].y);
std::cout << "face area scaling: (" << scale_x << "," << scale_y << ")\n";
float ratio = scale_y / scale_x;
// looking for a material named "face"
auto is_face_material = [](const material_t& material) -> bool { return material.name == "face"; };
auto it = std::find_if(materials.begin(), materials.end(), is_face_material);
assert(it != materials.end());
size_t face_material_id = std::distance(materials.begin(), it);
std::cout << "face material id =" << _ << face_material_id << std::endl;
#ifdef DEBUG
Mat texture = cv::imread("Model_obj/face.png");
assert(!texture.empty());
const Size texture_size = texture.size();
#else
const Size texture_size(1024, 1024);
#endif
Rect rect_src(Point(0, 0), texture_size);
const std::vector<vec3i> delaunay = getSubdivTriangle(rect_src, face_texcoords);
#ifdef DEBUG
const Size image_size = image.size();
Rect rect_dst(Point(0, 0), image_size);
Mat image2 = image.clone();
std::vector<vec2f> points_tl;
points_tl.reserve(points.size());
for(const vec2f& point : points)
{
vec2f point_tl(point.x, 1.0f - point.y);
Point pt(cvRound(point_tl.x * image_size.width), cvRound(point_tl.y * image_size.height));
circle(image2, pt, 3, CV_RGB(0, 255, 00), 1, LINE_8);
points_tl.push_back(point_tl);
}
std::vector<vec2f> face_texcoords_tl;
face_texcoords_tl.reserve(face_texcoords.size());
for(const vec2f& texcoord : face_texcoords)
face_texcoords_tl.push_back(vec2f(texcoord.u, 1.0f - texcoord.v));
cv::imshow("texture", drawTriangle(texture, face_texcoords_tl, delaunay));
cv::imshow("image", drawTriangle(image2, points_tl, delaunay));
imwrite("Output_obj/texture2.png", texture);
imwrite("Output_obj/image.png", image2);
cv::waitKey();
#endif // DEBUG
const float H = 165.937f; // Cutoff height, above which is the head, below the body.
// const vec2f head_bl(-10.09333f, 167.03551f), head_tr(10.09333, 187.22217); // face projects to XOZ plane
// const vec2f head_bl(-0.36657f, -0.50000f), head_tr(0.36657f, 0.50000f);
const vec2f head_bl(-0.08885, 1.52904/*, 0.09143*/), head_tr(0.08885, 1.77111/*, 0.09143*/);
const vec2f head_size = head_tr - head_bl;
int sum = 0;
std::vector<vec2f> verts_xy(POINT_MAX);
for(int i = 0; i < POINT_MAX; ++i)
verts_xy[i] = vec2f(verts[i].x, verts[i].y) - head_bl; // relative to the head bottom left. I'm running out of names!
for(std::pair<std::string, mesh_t>& string_mesh : meshes)
{
mesh_t& mesh = string_mesh.second;
std::vector<vec3f>& vertices = mesh.vertices;
std::vector<vec2f>& texcoords= mesh.texcoords;
const size_t vertex_count = vertices.size();
// head height remains unchanged, scale head width according to the ratio.
for(vec3f& vertex : vertices)
{
// if(vertex.y > H)
// vertex.x = (vertex.x - verts[EYES_C].x) * ratio + verts[EYES_C].x;
}
if(mesh.material_id != face_material_id)
continue;
// map position from source delaunay mesh to destination delaunay mesh
// I was expecting that cv::remap can fulfill this job.
for(size_t j = 0; j < vertex_count; ++j)
{
// handle vertex positions and texture coordinates
vec3f& vertex = vertices[j];
vec2f& texcoord = texcoords[j];
vec2f floor(std::floor(texcoord.s), std::floor(texcoord.t));
vec2f clamp = texcoord - floor;
for(const vec3i& tri : delaunay)
{
vec2f st;
Triangle2 src(face_texcoords[tri.i], face_texcoords[tri.j], face_texcoords[tri.k]);
if(src.contains(clamp, st))
{
Triangle2 dst(points[tri.i], points[tri.j], points[tri.k]);
texcoord = floor + dst.getPosition(st); // flip vertical
break;
}
}
/*
float model_x_m = verts[NOSE_C].x, image_x_m = points[NOSE_C].x;
int edge[] = { EDGE_CHIN_R, EDGE_LIP_R, EDGE_NOSE_R, EDGE_EYES_R, FOREHEAD_R, EDGE_FRONT_R};
for(int i = 0; i < sizeof(edge)/sizeof(edge[0]) - 1; ++i)
{
if(verts[i].y <= vertex.y && vertex.y < verts[i+1].y)
vertex.x = model_x_m + (vertex.x - image_x_m) / scale_x;
}
*/
vec2f st;
bool out_of_area = true;
vec2f point = vec2f(vertex.x, vertex.y) - head_bl;
for(const vec3i& tri : delaunay)
{
Triangle2 src(verts_xy[tri.i], verts_xy[tri.j], verts_xy[tri.k]);
if(src.contains(point, st))
{
// src.A + s * (src.B - src.A) + t * (src.C - src.A);
Triangle2 dst(face_texcoords[tri.i], face_texcoords[tri.j], face_texcoords[tri.k]);
vec2f texcoord_map = dst.getPosition(st);
vec2f vertex_map = texcoord_map * head_size;
// vertex.x = head_bl.x + vertex_map.width;
// vertex.y = head_bl.y + vertex_map.height;
out_of_area = false;
break;
}
}
// handle border vertex, we assume nose the center of face when projected on XOY plane.
if(out_of_area) // choose the nearest one
{
auto less_than = [&point](const vec2f& lhs, const vec2f& rhs) -> bool
{
vec2f vl = lhs - point, vr = rhs - point;
return (vl.x*vl.x + vl.y*vl.y) < (vr.x*vr.x + vr.y*vr.y);
};
auto it = std::min_element(verts_xy.begin(), verts_xy.end(), less_than);
size_t index = std::distance(verts_xy.begin(), it);
vec2f neareast_vertex = *it;
assert(index != NOSE_C);
float factor = distance(verts_xy[index], verts_xy[NOSE_C]) / distance(points[index], points[NOSE_C]);
//assert(0.01f < factor && factor < 100.0f);
vec2f tmp = factor * (point - verts_xy[index]);
point = points[index] + tmp;
// vertex.x = head_bl.x + point.x;
// vertex.y = head_bl.y + point.y;
}
}
sum += vertices.size();
}
std::cout<< "vertex count:" << _ << sum << std::endl;
}
//! Fitting line or Curve via RANSAC has constraint in some conditions.
// \minDataNum: the minimum number of data required to fit the model.
// \closeDataNum: the number of close data values required to assert that a model fits well to data.
// \iterNum: the number of iterations performed by the algorithm.
// \thValue: a threshold value for determining when a datum fits a model.
void FittingCurve_RANSAC(const std::vector<cv::Point2d> &pointSet,
const int &termNum, const int &minDataNum,
const int &iterNum, const double &thValue,
const int &closeDataNum, cv::Mat &coefs, const cv::Mat &img)
{
double startTime = (double)cv::getTickCount();
std::vector<cv::Point2d> bestPointSet; //data points from which this model has been estimated
double bestError = INFINITY; //the error of this model relative to the data
double errorSum, minErrorSum; //Prevent the sampling number smaller than closeDataNum
int errorFlag = 0;
cv::Mat maybeModel = cv::Mat::zeros(termNum, 1, CV_64F);
cv::Mat img2 = img.clone();
for(int iter = 0; iter < iterNum; iter++)
{
std::vector<cv::Point2d> tempPointSet = pointSet;
errorSum = 0;
//! Generate a set of maybe_inliers
std::vector<cv::Point2d> maybeInliersPointSet;
uint64 seed = iter + 1;
cv::RNG rng(seed);
for(int i = 0; i < minDataNum; i++)
{
int n = rng.uniform(0, (int)tempPointSet.size()-1);
maybeInliersPointSet.push_back(tempPointSet.at(n));
tempPointSet.erase(tempPointSet.begin()+n);
if (tempPointSet.size() == 0)
break;
}//end for
//! Fitting maybe_inliers to model by Least Squares Method
FittingCurve_LS(maybeInliersPointSet, termNum, maybeModel);
bestPointSet = maybeInliersPointSet; //Random Sampling Model
#if 0
for(int i=0; i<bestPointSet.size(); i++)
{
cv::circle(img2, bestPointSet.at(i), 1, CV_RGB(125, 125, 0));
}
std::vector<cv::Point2d> sampledPoints;
IPMDrawCurve(maybeModel, img2, sampledPoints, CV_RGB(100, 0, 0) );
//cv::imshow("RANSAC", img2);
//cv::waitKey();
#endif
//! Different from the linear modeling, the non-linear asks for computation of the distance between point and a polynomial curve.
//! deviration of distance function d{l(x)}/d{x} = 0
//! It is: 4*a2*x^3 + 6*a1*a2*x^2 + 2*(1+2*a2*(a0-y0)+a1^2)*x + 2*(a1*(-y0+a0)-x0) = 0
//! Solve this cubic funtion (polynomial of degree three)
double a0, a1, a2;
if (termNum == 2) {
//!Linear RANSAC
a0 = maybeModel.at<double>(0,0);
a1 = maybeModel.at<double>(1,0);
for (int i = 0; i < tempPointSet.size(); i++) {
int x = cvRound(tempPointSet.at(i).x);
int y = cvRound(tempPointSet.at(i).y);
double error = std::pow(std::abs(a1*x - y + a0)/sqrt(a1*a1+1), 2);
errorSum += error;
if (error < thValue) {
bestPointSet.push_back(tempPointSet.at(i));
}
}//end for
}
else if (termNum == 3) {
//!NonLinear RANSAC
a0 = maybeModel.at<double>(0,0);
a1 = maybeModel.at<double>(1,0);
a2 = maybeModel.at<double>(2,0);
for (int i = 0; i < tempPointSet.size(); i++) {
int x0 = cvRound(tempPointSet.at(i).x);
int y0 = cvRound(tempPointSet.at(i).y);
double error = std::pow(std::abs(y0 - (a0+(a1+a2*x0)*x0)), 2);
errorSum += error;
//cout << "Outliers Error: " << error << endl;
if (error < thValue) {
bestPointSet.push_back(tempPointSet.at(i));
}
}//end for
}
else {
std::cerr << "Only fitting line of degree 1 or 2." << std::endl;
}
//! First sampling as initialization
if(iter == 0) minErrorSum = errorSum;
// cout << "errorSum: " << errorSum << " minError: " << minErrorSum << endl;
//! Prevent the situation that all maybePoint number smaller than the close data number
if(errorSum < minErrorSum)
{
minErrorSum = errorSum;
if(errorFlag == 0)
coefs = maybeModel; //! Only works when all sampling doesn't have a good model.
}
//! this implies that we may have found a good model;
//! now test how good it is
if((int)bestPointSet.size() > closeDataNum)
{
double thisError = 0;
//cout << "Error Update!" << endl;
cv::Mat thisModel = cv::Mat::zeros(termNum, 1, CV_64F);
FittingCurve_LS(bestPointSet, termNum, thisModel);
if (termNum == 2) {
a0 = thisModel.at<double>(0,0);
a1 = thisModel.at<double>(1,0);
for(int i = 0; i < bestPointSet.size(); i++)
{
int x = cvRound(bestPointSet.at(i).x);
int y = cvRound(bestPointSet.at(i).y);
thisError += pow((a1*x - y + a0)/sqrt(a1*a1+1), 2);
}//end for
}
else if(termNum == 3) {
a0 = thisModel.at<double>(0,0);
a1 = thisModel.at<double>(1,0);
a2 = thisModel.at<double>(2,0);
for(int i = 0; i < bestPointSet.size(); i++)
{
int x0 = cvRound(bestPointSet.at(i).x);
int y0 = cvRound(bestPointSet.at(i).y);
thisError += std::pow(std::abs(y0 - (a0+(a1+a2*x0)*x0)), 2);
}//end for
}
else {
std::cerr << "Only fitting line of degree 1 or 2." << std::endl;
}
//! The line above contains the bug.
//! This_error should be replaced by a score that is either the size of the consensus set,
//! or the robust error norm computed on ALL samples (not just the consensus set).
// cout << "thisError: " << thisError << " bestError: " << bestError << endl;
if(thisError < bestError)
{
//! we have found a model which is better than any of the previous ones;
//! keep it until a better one is found)
//! best_consensus_set := consensus_set;
coefs = thisModel; //bestModel := thisModel
bestError = thisError;//bestError := thisError
errorFlag = 1;
}//end if
}//end if
// std::vector<cv::Point2d> sampledPoints2;
// IPMDrawCurve(coefs, img2, sampledPoints2, CV_RGB(0, 255, 0) );
// cv::imshow("RANSAC", img2);
}//end for iteration
double costTime = ((double)cv::getTickCount() - startTime)/cv::getTickFrequency();
printf("RANSAC Line Fitting Needs %.2f msec about %.2f Hz\n", costTime*1000, 1/costTime);
}//end FittingCurve_RANSAC
// ----------------------------------------------------------------------------------
void QualityMatcher::matchImagesAsync(cv::Mat imageSrc, cv::Mat imageDst, cv::Mat priorH, MatchingResultCallback cb)
{
if (m_matchingThread)
{
m_matchingThread->join();
}
// extract one channel for matching -> better have YUV, but green channel is god enough
cv::Mat rgbSrc[4];
cv::Mat rgbDst[4];
cv::split(imageSrc, rgbSrc);
cv::split(imageDst, rgbDst);
m_matchingThread.reset(new std::thread(std::bind(&QualityMatcher::doTheMagic, this, rgbSrc[1], rgbDst[1], priorH.clone(), cb)));
}
void LayoutTest::testFeatureCollector(const cv::Mat & src) const {
rdf::Timer dt;
// parse xml
PageXmlParser parser;
parser.read(mConfig.xmlPath());
// test loading of label lookup
LabelManager lm = LabelManager::read(mConfig.featureCachePath());
qInfo().noquote() << lm.toString();
// compute super pixels
SuperPixel sp(src);
if (!sp.compute())
qCritical() << "could not compute super pixels!";
// feed the label lookup
SuperPixelLabeler spl(sp.getMserBlobs(), Rect(src));
spl.setLabelManager(lm);
// set the ground truth
if (parser.page())
spl.setRootRegion(parser.page()->rootRegion());
if (!spl.compute())
qCritical() << "could not compute SuperPixel labeling!";
SuperPixelFeature spf(src, spl.set());
if (!spf.compute())
qCritical() << "could not compute SuperPixel features!";
FeatureCollectionManager fcm(spf.features(), spf.set());
fcm.write(spl.config()->featureFilePath());
FeatureCollectionManager testFcm = FeatureCollectionManager::read(spl.config()->featureFilePath());
for (int idx = 0; idx < testFcm.collection().size(); idx++) {
if (testFcm.collection()[idx].label() != fcm.collection()[idx].label())
qWarning() << "wrong labels!" << testFcm.collection()[idx].label() << "vs" << fcm.collection()[idx].label();
else
qInfo() << testFcm.collection()[idx].label() << "is fine...";
}
// drawing
cv::Mat rImg = src.clone();
// save super pixel image
//rImg = superPixel.drawSuperPixels(rImg);
//rImg = tabStops.draw(rImg);
rImg = spl.draw(rImg);
rImg = spf.draw(rImg);
QString dstPath = rdf::Utils::instance().createFilePath(mConfig.outputPath(), "-textlines");
rdf::Image::save(rImg, dstPath);
qDebug() << "debug image saved: " << dstPath;
qDebug() << "image path: " << mConfig.imagePath();
}
//===========================================================================
void Patch::Init(int t, double a, double b, cv::Mat &W)
{
assert((W.type() == CV_32F)); _t=t; _a=a; _b=b; _W=W.clone(); return;
}
bool Stabilizer::forward_backward_track(const cv::Mat& frame)
{
cv::Mat previous_frame_gray;
cv::cvtColor(prevFrame, previous_frame_gray, cv::COLOR_BGR2GRAY);
cv::goodFeaturesToTrack(previous_frame_gray, previousFeatures, 500, 0.01, 5);
size_t n = previousFeatures.size();
CV_Assert(n);
// Compute optical flow in selected points.
std::vector<cv::Point2f> currentFeatures;
std::vector<uchar> state;
std::vector<float> error;
cv::calcOpticalFlowPyrLK(prevFrame, frame, previousFeatures, currentFeatures, state, error);
float median_error = median<float>(error);
std::vector<cv::Point2f> good_points;
std::vector<cv::Point2f> curr_points;
for (size_t i = 0; i < n; ++i)
{
if (state[i] && (error[i] <= median_error))
{
good_points.push_back(previousFeatures[i]);
curr_points.push_back(currentFeatures[i]);
}
}
size_t s = good_points.size();
CV_Assert(s == curr_points.size());
//Compute backward optical flow
std::vector<cv::Point2f> backwardPoints;
std::vector<uchar> backState;
std::vector<float> backError;
cv::calcOpticalFlowPyrLK(frame, prevFrame, curr_points, backwardPoints, backState, backError);
float median_back_error = median<float>(backError);
CV_Assert(s == backwardPoints.size());
std::vector<float> diff(s);
for (size_t i = 0; i < s; ++i)
{
diff[i] = cv::norm(good_points[i] - backwardPoints[i]);
// diff[i] = (good_points[i].x - backwardPoints[i].x) * (good_points[i].x - backwardPoints[i].x) + (good_points[i].y - backwardPoints[i].y) * (good_points[i].y - backwardPoints[i].y);
}
for (int i = s - 1; i >= 0; --i)
{
if (!backState[i] || (backError[i] <= median_back_error) || (diff[i] > 400))
{
good_points.erase(good_points.begin() + i);
curr_points.erase(curr_points.begin() + i);
}
}
s = good_points.size();
// Find points shift.
std::vector<float> shifts_x(s);
std::vector<float> shifts_y(s);
for (size_t i = 0; i < s; ++i)
{
shifts_x[i] = curr_points[i].x - good_points[i].x;
shifts_y[i] = curr_points[i].y - good_points[i].y;
}
// Find median shift.
cv::Point2f median_shift(median<float>(shifts_x), median<float>(shifts_y));
xshift.push_back(median_shift.x);
yshift.push_back(median_shift.y);
prevFrame = frame.clone();
return true;
}
bool place::reshowPlacement(const std::string &scanName,
const std::string &zerosFile,
const std::string &doorName,
const place::DoorDetector &d,
const std::string &preDone) {
const std::string buildName = scanName.substr(scanName.find("_") - 3, 3);
const std::string scanNumber = scanName.substr(scanName.find(".") - 3, 3);
const std::string placementName =
buildName + "_placement_" + scanNumber + ".dat";
std::ifstream in(preDone + placementName, std::ios::in | std::ios::binary);
if (!in.is_open())
return false;
if (!FLAGS_reshow)
return true;
if (!FLAGS_quietMode)
std::cout << placementName << std::endl;
std::vector<cv::Mat> rotatedScans, toTrim;
std::vector<Eigen::Vector2i> zeroZero;
place::loadInScans(scanName, zerosFile, toTrim, zeroZero);
place::trimScans(toTrim, rotatedScans, zeroZero);
std::vector<std::vector<place::Door>> doors = loadInDoors(doorName, zeroZero);
int num;
in.read(reinterpret_cast<char *>(&num), sizeof(num));
std::vector<place::posInfo> scores(num);
for (auto &s : scores)
in.read(reinterpret_cast<char *>(&s), sizeof(place::posInfo));
int cutOffNum = place::getCutoffIndex(
placementName, scores, [](const place::posInfo &s) { return s.score; });
cutOffNum = FLAGS_top > 0 ? FLAGS_top : cutOffNum;
num = std::min(num, cutOffNum);
cvNamedWindow("Preview", CV_WINDOW_NORMAL);
if (!FLAGS_quietMode)
std::cout << "Showing minima: " << num << std::endl;
for (int k = 0; k < std::min(num, (int)scores.size());) {
auto ¤tScore = scores[k];
const cv::Mat &bestScan = rotatedScans[currentScore.rotation];
const int xOffset = currentScore.x - zeroZero[currentScore.rotation][0];
const int yOffset = currentScore.y - zeroZero[currentScore.rotation][1];
cv::Mat_<cv::Vec3b> output = fpColor.clone();
auto &res = d.getResponse(0);
for (int i = 0; i < res.outerSize(); ++i)
for (Eigen::SparseMatrix<char>::InnerIterator it(res, i); it; ++it)
if (it.value() > 1)
output(it.row(), it.col()) = cv::Vec3b(0, 255, 0);
for (int j = 0; j < bestScan.rows; ++j) {
if (j + yOffset < 0 || j + yOffset >= fpColor.rows)
continue;
const uchar *src = bestScan.ptr<uchar>(j);
for (int i = 0; i < bestScan.cols; ++i) {
if (i + xOffset < 0 || i + xOffset >= fpColor.cols)
continue;
if (src[i] != 255) {
output(j + yOffset, i + xOffset) = cv::Vec3b(0, 0, 255 - src[i]);
}
}
}
for (int j = -10; j < 10; ++j)
for (int i = -10; i < 10; ++i)
output(j + currentScore.y, i + currentScore.x) = cv::Vec3b(255, 0, 0);
for (auto &d : doors[currentScore.rotation]) {
auto color = randomColor();
for (double x = 0; x < d.w; ++x) {
Eigen::Vector3i index =
(d.corner + x * d.xAxis + Eigen::Vector3d(xOffset, yOffset, 0))
.unaryExpr([](auto v) { return std::round(v); })
.cast<int>();
for (int k = -2; k <= 2; ++k) {
for (int l = -2; l <= 2; ++l) {
output(index[1] + k, index[0] + l) = color;
}
}
}
}
if (!FLAGS_quietMode) {
std::cout << ¤tScore << std::endl;
std::cout << "% of scan unexplained: "
<< currentScore.scanFP / currentScore.scanPixels
<< " Index: " << k << std::endl
<< std::endl;
}
const int keyCode = cv::rectshow(output);
if (keyCode == 27) {
cv::imwrite(preDone + buildName + "_ss_" + scanNumber + ".png", output);
break;
} else if (keyCode == 8)
k = k > 0 ? k - 1 : k;
else
++k;
}
return true;
}
void OpenCV_Function::filter(){
g_srcImage=cv::imread("girl-t1.jpg");
//均值滤波（邻域平均滤波）
cv::blur(g_srcImage,g_resultImage,cv::Size(2,2)); //值越大越模糊
cv::imshow( "blur", g_resultImage );
//高斯滤波
cv::GaussianBlur( g_srcImage, g_resultImage, cv::Size( 99, 99 ), 0, 0 ); //值越大越模糊,且值只能是正数和奇数
cv::imshow("GaussianBlur", g_resultImage );
//方框滤波
cv::boxFilter(g_srcImage,g_resultImage,-1,cv::Size(5,5));//
cv::imshow("boxFilter", g_resultImage );
//中值滤波
cv::medianBlur(g_srcImage,g_resultImage,9);//这个参数必须是大于1的奇数
cv::imshow("medianBlur", g_resultImage );
//bilateralFilter 双边滤波器
cv::bilateralFilter( g_srcImage, g_resultImage, 25, 25*2, 25/2 );
cv::imshow("bilateralFilter", g_resultImage );
//erode函数，使用像素邻域内的局部极小运算符来腐蚀一张图片
cv::Mat element = cv::getStructuringElement(cv::MORPH_ELLIPSE, cv::Size(3,3));
cv::erode(g_srcImage,g_resultImage,element,cv::Point(-1,-1),1);
cv::imshow("erode", g_resultImage );
//dilate函数，使用像素邻域内的局部极大运算符来膨胀一张图片
cv::dilate(g_srcImage,g_resultImage,element);
cv::imshow("dilate", g_resultImage );
//开运算（Opening Operation），其实就是先腐蚀后膨胀的过程
//dst=open(src,element)=dilate(erode(src,element));
//闭运算(Closing Operation), 其实就是先膨胀后腐蚀的过程
//dst=close(src,element)=erode(dilate(src,element));
//形态学梯度（Morphological Gradient）为膨胀图与腐蚀图之差
//dst=morph_grad(src,element)=dilate(src,element)-erode(src,element);
//顶帽运算（Top Hat）又常常被译为”礼帽“运算。为原图像与上文刚刚介绍的“开运算“的结果图之差
//dst=src-open(src,element);
//黑帽（Black Hat）运算为”闭运算“的结果图与原图像之差。
//dst=close(src,element)-src;
cv::morphologyEx(g_srcImage,g_resultImage, cv::MORPH_OPEN, element);
cv::imshow( "morphologyEx", g_resultImage );
//最简单的canny用法，拿到原图后直接用。
//这个函数阈值1和阈值2两者的小者用于边缘连接，而大者用来控制强边缘的初始段，推荐的高低阈值比在2:1到3:1之间。
cv::Mat cannyMat=g_srcImage.clone();
cv::Canny(cannyMat,cannyMat,3,9);
cv::imshow( "Canny", cannyMat );
//----------------------------------------------------------------------------------
// 二、高阶的canny用法，转成灰度图，降噪，用canny，最后将得到的边缘作为掩码，拷贝原图到效果图上，得到彩色的边缘图
//----------------------------------------------------------------------------------
cv::Mat dst,edge,gray;
dst.create( g_srcImage.size(), g_srcImage.type() ); // 【1】创建与src同类型和大小的矩阵(dst)
cv::cvtColor(g_srcImage,gray,CV_BGR2GRAY);// 【2】将原图像转换为灰度图像
cv::blur( gray, edge, cv::Size(3,3) ); // 【3】先用使用 3x3内核来降噪
cv::Canny(edge,edge,3,9);
dst = cv::Scalar::all(0); //【5】将dst内的所有元素设置为0
g_srcImage.copyTo( dst, edge); //【6】使用Canny算子输出的边缘图g_cannyDetectedEdges作为掩码，来将原图g_srcImage拷到目标图g_dstImage中
cv::imshow( "Canny2", dst );
//----------------------------------------------------------------------------------
// 调用Sobel函数的实例代码
//----------------------------------------------------------------------------------
cv::Mat grad_x, grad_y;
cv::Mat abs_grad_x, abs_grad_y;
//【3】求 X方向梯度
cv::Sobel( g_srcImage, grad_x, CV_16S, 1, 0, 3, 1, 1, cv::BORDER_DEFAULT );
cv::convertScaleAbs( grad_x, abs_grad_x );
cv::imshow("【效果图】 X方向Sobel", abs_grad_x);
//【4】求Y方向梯度
cv::Sobel( g_srcImage, grad_y, CV_16S, 0, 1, 3, 1, 1, cv::BORDER_DEFAULT );
cv::convertScaleAbs( grad_y, abs_grad_y );
cv::imshow("【效果图】Y方向Sobel", abs_grad_y);
//【5】合并梯度(近似)
addWeighted( abs_grad_x, 0.5, abs_grad_y, 0.5, 0, dst );
cv::imshow("【效果图】整体方向Sobel", dst);
//----------------------------------------------------------------------------------
// Laplacian 算子是n维欧几里德空间中的一个二阶微分算子，定义为梯度grad（）的散度div（）。因此如果f是二阶可微的实函数，则f的拉普拉斯算子定义为：
//----------------------------------------------------------------------------------
cv::Laplacian( g_srcImage, dst, CV_16S, 3, 1, 0, cv::BORDER_DEFAULT );
cv::convertScaleAbs( dst, abs_grad_y );
cv::imshow("Laplacian", abs_grad_y);
//----------------------------------------------------------------------------------
// scharr一般我就直接称它为滤波器，而不是算子。上文我们已经讲到，它在OpenCV中主要是配合Sobel算子的运算而存在的,
//----------------------------------------------------------------------------------
//【3】求 X方向梯度
cv::Scharr( g_srcImage, grad_x, CV_16S, 1, 0, 1, 0, cv::BORDER_DEFAULT );
cv:: convertScaleAbs( grad_x, abs_grad_x );
cv::imshow("【效果图】 X方向Scharr", abs_grad_x);
//【4】求Y方向梯度
cv::Scharr( g_srcImage, grad_y, CV_16S, 0, 1, 1, 0, cv::BORDER_DEFAULT );
cv::convertScaleAbs( grad_y, abs_grad_y );
cv::imshow("【效果图】Y方向Scharr", abs_grad_y);
cv::imshow( WINDOW_NAME1, g_srcImage );
}
scoredContour goal_pipeline(cv::Mat input, bool suppress_output=false) {
std::vector< std::vector<cv::Point> > contours;
cv::Mat contourOut = input.clone();
cv::findContours(contourOut, contours, CV_RETR_LIST, CV_CHAIN_APPROX_NONE);
std::vector<scoredContour> finalscores;
if(!suppress_output) { std::cout << "Found " << contours.size() << " contours." << std::endl; }
unsigned int ctr = 0;
for(std::vector< std::vector<cv::Point> >::iterator i = contours.begin();
i != contours.end(); ++i) {
double area = cv::contourArea(*i);
double perimeter = cv::arcLength(*i, true);
cv::Rect bounds = cv::boundingRect(*i);
const double cvarea_target = (80.0/240.0);
const double asratio_target = (20.0/12.0);
const double area_threshold = 1000;
/* Area Thresholding Test
* Only accept contours of a certain total size.
*/
if(area < area_threshold) {
continue;
}
if(!suppress_output) {
std::cout << std::endl;
std::cout << "Contour " << ctr << ": " << std::endl;
ctr++;
std::cout << "Area: " << area << std::endl;
std::cout << "Perimeter: " << perimeter << std::endl;
}
/* Coverage Area Test
* Compare particle area vs. Bounding Rectangle area.
* score = 1 / abs((1/3)- (particle_area / boundrect_area))
* Score decreases linearly as coverage area tends away from 1/3. */
double cvarea_score = 0;
double coverage_area = area / bounds.area();
cvarea_score = scoreDistanceFromTarget(cvarea_target, coverage_area);
/* Aspect Ratio Test
* Computes aspect ratio of detected objects.
*/
double tmp = bounds.width;
double aspect_ratio = tmp / bounds.height;
double ar_score = scoreDistanceFromTarget(asratio_target, aspect_ratio);
/* Image Moment Test
* Computes image moments and compares it to known values.
*/
cv::Moments m = cv::moments(*i);
double moment_score = scoreDistanceFromTarget(0.28, m.nu02);
/* Image Orientation Test
* Computes angles off-axis or contours.
*/
// theta = (1/2)atan2(mu11, mu20-mu02) radians
// theta ranges from -90 degrees to +90 degrees.
double theta = (atan2(m.mu11,m.mu20-m.mu02) * 90) / pi;
double angle_score = (90 - fabs(theta))+10;
if(!suppress_output) {
std::cout << "nu-02: " << m.nu02 << std::endl;
std::cout << "CVArea: " << coverage_area << std::endl;
std::cout << "AsRatio: " << aspect_ratio << std::endl;
std::cout << "Orientation: " << theta << std::endl;
}
double total_score = (moment_score + cvarea_score + ar_score + angle_score) / 4;
if(!suppress_output) {
std::cout << "CVArea Score: " << cvarea_score << std::endl;
std::cout << "AsRatio Score: " << ar_score << std::endl;
std::cout << "Moment Score: " << moment_score << std::endl;
std::cout << "Angle Score: " << angle_score << std::endl;
std::cout << "Total Score: " << total_score << std::endl;
}
finalscores.push_back(std::make_pair(total_score, std::move(*i)));
}
if(finalscores.size() > 0) {
std::sort(finalscores.begin(), finalscores.end(), &scoresort);
return finalscores.back();
} else {
return std::make_pair(0.0, std::vector<cv::Point>());
}
}
bool pixkit::labeling::twoPass(const cv::Mat &src,cv::Mat &dst,const int offset){
//////////////////////////////////////////////////////////////////////////
///// EXCEPTION
if(src.type()!=CV_8UC1){
CV_Error(CV_StsBadArg,"[xxx] src's type should be CV_8UC1.");
}
//標籤為從1開始
int LableNumber = 1;
bool **Table = NULL;
bool *assitT = NULL;
bool *checkTable = NULL;
int *refTable = NULL;
int **result = NULL;
int *resultT = NULL;
int C[5],min,temp;
int W,A,Q,E,S;
dst = src.clone();
//Mat convert to [][]
result = new int *[src.rows];
resultT = new int [src.rows*src.cols];
memset(resultT,0,src.rows*src.cols);
for(int i=0;i<src.rows;i++,resultT+=src.cols)
result[i] = resultT;
for(int i=0;i<src.rows;i++)
{
int temp = i*src.cols;
for(int j=0;j<src.cols;j++)
{
result[i][j] = (int)src.data[temp+j];
}
}
//正規化用標籤
std::vector<int> ObjectIndex;
//第一次掃描
for(int i=0;i<src.rows;i++)
{
for(int j=0;j<src.cols;j++)
{
C[0] = result[i][j];
if (C[0] <128)
continue;
/*如果 C[0] > 128 代表有物件*/
min = src.rows*src.cols;
if(j-1 <0)
C[1] = 0;
else
C[1] = result[i][j-1];
if (i-1<0 || j-1 <0)
C[2] = 0;
else
C[2] = result[i-1][j-1];
if(i-1<0)
C[3] = 0;
else
C[3] = result[i-1][j];
if (i-1<0 || j+1 >=src.cols)
C[4] = 0;
else
C[4] = result[i-1][j+1];
if(C[1] ==0 && C[2] ==0 && C[3] ==0 && C[4] ==0)
{
C[0] = LableNumber;
LableNumber++;
}
else
{
for(int k=1;k<=4;k++)
{
if(C[k]<min && C[k] != 0)
min = C[k];
}
C[0] = min;
}
result[i][j] = C[0];
}
}
//LableNumber != 1 代表有前景物件存在
//LableNumber == 1 代表無前景物件
//使用動態新增Table記憶體
if(LableNumber != 1)
{
Table = new bool *[LableNumber];
assitT = new bool [LableNumber*LableNumber];
memset(assitT,0,LableNumber*LableNumber);
refTable = new int [LableNumber];
checkTable = new bool [LableNumber];
for(int i=0;i<LableNumber;i++,assitT+=LableNumber)
{
Table[i] = assitT;
refTable[i] = 0;
checkTable[i] = 0;
}
}
//第二次掃描
for(int i=0;i<src.rows;i++)
{
for(int j=0;j<src.cols;j++)
{
S = result[i][j];
if(S == 0)
continue;
if(i-1<0)
W = 0;
else
W = result[i-1][j];
if(i-1<0 || j-1<0)
Q = 0;
else
Q = result[i-1][j-1];
if(j-1<0)
A = 0;
else
A = result[i][j-1];
if (i-1<0 || j+1 >= src.cols)
E = 0;
else
E = result[i-1][j+1];
/*填Table值*/
Table[S][S] = 1;
if(W != 0)
{
Table[S][W] = 1;
Table[W][S] = 1;
}
if(A!=0)
{
Table[S][A] = 1;
Table[A][S] = 1;
}
if(Q!=0)
{
Table[S][Q] = 1;
Table[Q][S] = 1;
}
if (E!=0)
{
Table[S][E] = 1;
Table[E][S] = 1;
}
}
}
//Equivalent Table
//因為標籤值是從1開始，所以i,j從1開始
for(int i=1;i<LableNumber;i++)
{
for (int j=1;j<LableNumber;j++)
{
if(Table[i][j] ==1)
{
for(int ii = 1;ii<LableNumber;ii++)
if(Table[j][ii] == 1)
{
Table[i][ii] = 1;
Table[ii][i] = 1;
}
}
}
}
//建立ref對照表
for(int i=1;i<LableNumber;i++)
{
for (int j=1;j<LableNumber;j++)
{
if(Table[i][j] != 0)
{
refTable[i] = j;
checkTable[j] = true;
break;
}
}
}
//LableNumber != 1 代表有前景物件存在
for(int i=1;i<LableNumber;i++)
{
if(checkTable[i] == true)
{
ObjectIndex.push_back(i);
}
}
//利用對照表去Refine並同時將標籤正規化
int labelValue = 0,labelIndex = 0;
for(int i=0;i<src.rows;i++)
{
for(int j=0;j<src.cols;j++)
{
temp = result[i][j];
if(temp != 0)
{
labelValue = refTable[temp];
for(int k=0;k<ObjectIndex.size();k++)
{
if(ObjectIndex[k] == labelValue)
{
result[i][j] = k+offset;
break;
}
}
}
}
}
//[][] convert to Mat
for(int i=0;i<src.rows;i++)
{
int temp = i*src.cols;
for(int j=0;j<src.cols;j++)
{
dst.data[temp+j] = result[i][j];
}
}
//LableNumber != 1 代表有前景物件存在
//刪除Table記憶體
if(LableNumber != 1)
{
delete []Table[0];
delete []Table;
delete []result[0];
delete []result;
delete []refTable;
delete []checkTable;
}
return true;
}
void SolutionViewer::LoadImage(cv::Mat& img)
{
input_img = img.clone();
image_loaded_flag = true;
}
WeightedGaussian::WeightedGaussian(double weight, const cv::Mat &mean, const cv::Mat &covariance)
: _Weight(weight), _Mean(mean.clone()), _Covariance(covariance.clone())
{
}
void FSolver::normalizePoints(const cv::Mat &P, cv::Mat &Q) const
{
// turn P into homogeneous coords first
if(P.rows == 3) // P is 3xN
{
if(P.type() == CV_64F)
Q = P.clone();
else
P.convertTo(Q, CV_64F);
cv::Mat aux = Q.row(0);
cv::divide(aux, Q.row(2), aux);
aux = Q.row(1);
cv::divide(aux, Q.row(2), aux);
}
else if(P.rows == 2) // P is 2xN
{
const int N = P.cols;
Q.create(3, N, CV_64F);
Q.row(2) = 1;
if(P.type() == CV_64F)
{
Q.rowRange(0,2) = P * 1;
}
else
{
cv::Mat aux = Q.rowRange(0,2);
P.convertTo(aux, CV_64F);
}
}
else if(P.cols == 3) // P is Nx3
{
if(P.type() == CV_64F)
Q = P.t();
else
{
P.convertTo(Q, CV_64F);
Q = Q.t();
}
cv::Mat aux = Q.row(0);
cv::divide(aux, Q.row(2), aux);
aux = Q.row(1);
cv::divide(aux, Q.row(2), aux);
}
else if(P.cols == 2) // P is Nx2
{
const int N = P.rows;
Q.create(N, 3, CV_64F);
Q.col(2) = 1;
cv::Mat aux;
if(P.type() == CV_64F)
{
aux = Q.rowRange(0,2);
P.rowRange(0,2).copyTo(aux);
}
else
{
aux = Q.colRange(0,2);
P.convertTo(aux, CV_64F);
}
Q = Q.t();
}
}
bool cv::RGBDOdometry( cv::Mat& Rt, const Mat& initRt,
const cv::Mat& image0, const cv::Mat& _depth0, const cv::Mat& validMask0,
const cv::Mat& image1, const cv::Mat& _depth1, const cv::Mat& validMask1,
const cv::Mat& cameraMatrix, float minDepth, float maxDepth, float maxDepthDiff,
const std::vector<int>& iterCounts, const std::vector<float>& minGradientMagnitudes,
int transformType )
{
const int sobelSize = 3;
const double sobelScale = 1./8;
Mat depth0 = _depth0.clone(),
depth1 = _depth1.clone();
// check RGB-D input data
CV_Assert( !image0.empty() );
CV_Assert( image0.type() == CV_8UC1 );
CV_Assert( depth0.type() == CV_32FC1 && depth0.size() == image0.size() );
CV_Assert( image1.size() == image0.size() );
CV_Assert( image1.type() == CV_8UC1 );
CV_Assert( depth1.type() == CV_32FC1 && depth1.size() == image0.size() );
// check masks
CV_Assert( validMask0.empty() || (validMask0.type() == CV_8UC1 && validMask0.size() == image0.size()) );
CV_Assert( validMask1.empty() || (validMask1.type() == CV_8UC1 && validMask1.size() == image0.size()) );
// check camera params
CV_Assert( cameraMatrix.type() == CV_32FC1 && cameraMatrix.size() == Size(3,3) );
// other checks
CV_Assert( iterCounts.empty() || minGradientMagnitudes.empty() ||
minGradientMagnitudes.size() == iterCounts.size() );
CV_Assert( initRt.empty() || (initRt.type()==CV_64FC1 && initRt.size()==Size(4,4) ) );
vector<int> defaultIterCounts;
vector<float> defaultMinGradMagnitudes;
vector<int> const* iterCountsPtr = &iterCounts;
vector<float> const* minGradientMagnitudesPtr = &minGradientMagnitudes;
if( iterCounts.empty() || minGradientMagnitudes.empty() )
{
defaultIterCounts.resize(4);
defaultIterCounts[0] = 7;
defaultIterCounts[1] = 7;
defaultIterCounts[2] = 7;
defaultIterCounts[3] = 10;
defaultMinGradMagnitudes.resize(4);
defaultMinGradMagnitudes[0] = 12;
defaultMinGradMagnitudes[1] = 5;
defaultMinGradMagnitudes[2] = 3;
defaultMinGradMagnitudes[3] = 1;
iterCountsPtr = &defaultIterCounts;
minGradientMagnitudesPtr = &defaultMinGradMagnitudes;
}
preprocessDepth( depth0, depth1, validMask0, validMask1, minDepth, maxDepth );
vector<Mat> pyramidImage0, pyramidDepth0,
pyramidImage1, pyramidDepth1, pyramid_dI_dx1, pyramid_dI_dy1, pyramidTexturedMask1,
pyramidCameraMatrix;
buildPyramids( image0, image1, depth0, depth1, cameraMatrix, sobelSize, sobelScale, *minGradientMagnitudesPtr,
pyramidImage0, pyramidDepth0, pyramidImage1, pyramidDepth1,
pyramid_dI_dx1, pyramid_dI_dy1, pyramidTexturedMask1, pyramidCameraMatrix );
Mat resultRt = initRt.empty() ? Mat::eye(4,4,CV_64FC1) : initRt.clone();
Mat currRt, ksi;
for( int level = (int)iterCountsPtr->size() - 1; level >= 0; level-- )
{
const Mat& levelCameraMatrix = pyramidCameraMatrix[level];
const Mat& levelImage0 = pyramidImage0[level];
const Mat& levelDepth0 = pyramidDepth0[level];
Mat levelCloud0;
cvtDepth2Cloud( pyramidDepth0[level], levelCloud0, levelCameraMatrix );
const Mat& levelImage1 = pyramidImage1[level];
const Mat& levelDepth1 = pyramidDepth1[level];
const Mat& level_dI_dx1 = pyramid_dI_dx1[level];
const Mat& level_dI_dy1 = pyramid_dI_dy1[level];
CV_Assert( level_dI_dx1.type() == CV_16S );
CV_Assert( level_dI_dy1.type() == CV_16S );
const double fx = levelCameraMatrix.at<double>(0,0);
const double fy = levelCameraMatrix.at<double>(1,1);
const double determinantThreshold = 1e-6;
Mat corresps( levelImage0.size(), levelImage0.type() );
// Run transformation search on current level iteratively.
for( int iter = 0; iter < (*iterCountsPtr)[level]; iter ++ )
{
int correspsCount = computeCorresp( levelCameraMatrix, levelCameraMatrix.inv(), resultRt.inv(DECOMP_SVD),
levelDepth0, levelDepth1, pyramidTexturedMask1[level], maxDepthDiff,
corresps );
if( correspsCount == 0 )
break;
bool solutionExist = computeKsi( transformType,
levelImage0, levelCloud0,
levelImage1, level_dI_dx1, level_dI_dy1,
corresps, correspsCount,
fx, fy, sobelScale, determinantThreshold,
ksi );
if( !solutionExist )
break;
computeProjectiveMatrix( ksi, currRt );
resultRt = currRt * resultRt;
#if SHOW_DEBUG_IMAGES
std::cout << "currRt " << currRt << std::endl;
Mat warpedImage0;
const Mat distCoeff(1,5,CV_32FC1,Scalar(0));
warpImage<uchar>( levelImage0, levelDepth0, resultRt, levelCameraMatrix, distCoeff, warpedImage0 );
imshow( "im0", levelImage0 );
imshow( "wim0", warpedImage0 );
imshow( "im1", levelImage1 );
waitKey();
#endif
}
}
Rt = resultRt;
return !Rt.empty();
}
/**
* @author JIA Pei, YAO Wei
* @version 2010-05-20
* @brief Additive ASM ND Profiles Fitting, for static images, so that we record the whole fitting process
* @param iImg Input - image to be fitted
* @param oImages Output - the fitting process
* @param dim Input - profile dimension, 1, 2, 4 or 8
* @param epoch Input - the iteration epoch
* @param pyramidlevel Input - pyramid level, 1, 2, 3 or 4 at most
* @note Refer to "AAM Revisited, page 34, figure 13", particularly, those steps.
*/
float VO_FittingASMNDProfiles::VO_ASMNDProfileFitting( const cv::Mat& iImg,
std::vector<cv::Mat>& oImages,
unsigned int epoch,
unsigned int pyramidlevel,
unsigned int dim,
bool record)
{
double t = (double)cv::getTickCount();
this->m_iNbOfPyramidLevels = pyramidlevel;
this->SetProcessingImage(iImg, this->m_VOASMNDProfile);
this->m_iIteration = 0;
if(record)
{
cv::Mat temp = iImg.clone();
VO_Fitting2DSM::VO_DrawMesh(this->m_VOFittingShape, this->m_VOASMNDProfile, temp);
oImages.push_back(temp);
}
// Get m_MatModelAlignedShapeParam and m_fScale, m_vRotateAngles, m_MatCenterOfGravity
this->m_VOASMNDProfile->VO_CalcAllParams4AnyShapeWithConstrain( this->m_VOFittingShape,
this->m_MatModelAlignedShapeParam,
this->m_fScale,
this->m_vRotateAngles,
this->m_MatCenterOfGravity);
this->m_VOFittingShape.ConstrainShapeInImage(this->m_ImageProcessing);
if(record)
{
cv::Mat temp1 = iImg.clone();
VO_Fitting2DSM::VO_DrawMesh(this->m_VOFittingShape, this->m_VOASMNDProfile, temp1);
oImages.push_back(temp1);
}
// Explained by YAO Wei, 2008-2-9.
// Scale this->m_VOFittingShape, so face width is a constant StdFaceWidth.
//this->m_fScale2 = this->m_VOASMNDProfile->m_VOReferenceShape.GetWidth() / this->m_VOFittingShape.GetWidth();
this->m_fScale2 = this->m_VOASMNDProfile->m_VOReferenceShape.GetCentralizedShapeSize() / this->m_VOFittingShape.GetCentralizedShapeSize();
this->m_VOFittingShape *= this->m_fScale2;
int w = (int)(iImg.cols*this->m_fScale2);
int h = (int)(iImg.rows*this->m_fScale2);
cv::Mat SearchImage = cv::Mat(cv::Size( w, h ), this->m_ImageProcessing.type(), this->m_ImageProcessing.channels() );
float PyrScale = pow(2.0f, (float) (this->m_iNbOfPyramidLevels-1.0f) );
this->m_VOFittingShape /= PyrScale;
const int nQualifyingDisplacements = (int)(this->m_VOASMNDProfile->m_iNbOfPoints * VO_Fitting2DSM::pClose);
// for each level in the image pyramid
for (int iLev = this->m_iNbOfPyramidLevels-1; iLev >= 0; iLev--)
{
// Set image roi, instead of cvCreateImage a new image to speed up
cv::Mat siROI = SearchImage(cv::Rect(0, 0, (int)(w/PyrScale), (int)(h/PyrScale) ) );
cv::resize(this->m_ImageProcessing, siROI, siROI.size());
if(record)
{
cv::Mat temp2 = cv::Mat(iImg.size(), iImg.type(), iImg.channels());
cv::Mat temp2ROI = temp2(cv::Range (0, (int)(iImg.rows/PyrScale) ), cv::Range (0, (int)(iImg.cols/PyrScale) ) );
cv::resize(iImg, temp2ROI, temp2ROI.size() );
oImages.push_back(temp2);
VO_Fitting2DSM::VO_DrawMesh(this->m_VOFittingShape / this->m_fScale2, this->m_VOASMNDProfile, temp2);
}
this->m_VOEstimatedShape = this->m_VOFittingShape;
this->PyramidFit( this->m_VOEstimatedShape,
SearchImage,
oImages,
iLev,
VO_Fitting2DSM::pClose,
epoch,
dim,
record);
this->m_VOFittingShape = this->m_VOEstimatedShape;
if (iLev != 0)
{
PyrScale /= 2.0f;
this->m_VOFittingShape *= 2.0f;
}
}
// Explained by YAO Wei, 2008-02-09.
// this->m_fScale2 back to original size
this->m_VOFittingShape /= this->m_fScale2;
t = ((double)cv::getTickCount() - t )/ (cv::getTickFrequency()*1000.);
printf("MRASM fitting time cost: %.2f millisec\n", t);
return t;
}
//\fn void ExtrinsicParam::computeRtMatrix(double pan, double tilt, cv::Mat image);
///\brief This function computes the rotation matrix and the translation vector of the extrinsic parameters.
///\param pan Value of the camera panoramique when the image has been taken.
///\param tilt Value of the camera tilt when the image has been taken.
///\param image The image from which the rotation matrix and the translation vector should be computed.
void ExtrinsicParam::computeRtMatrix(double pan, double tilt, cv::Mat image)
{
// Save the value of the pan and the tilt in order to modify the extrinsic parameters when the camera position change
this->pan = pan;
this->tilt = tilt;
cv::Mat img = image.clone();
cv::Mat initImg = image.clone();
int cnt = 0;
double fx = K(0,0), fy = K(1,1), cx = K(0,2), cy = K(1,2);
double u, v, x, y;
double dist_to_point;
Eigen::MatrixXd Pr, Pi;
Pr.resize(4,6);
Pr.setZero();
Pi.resize(3,6);
Pi.setZero();
pp.cnt = 0;
// Create a window of the scene
cv::namedWindow("Extrinsic Image",CV_WINDOW_AUTOSIZE);
cvSetMouseCallback("Extrinsic Image",On_Mouse,0);
cv::waitKey(10);
cv::imshow("Extrinsic Image", img);
cv::waitKey(10);
// We need 6 points to compute the translation vector and the rotation matrix
while (pp.cnt <= 6)
{
if (cnt > pp.cnt) // Case where we do a right click in order to remove the last point inserted
{
switch (cnt)
{
case 1:
{
img = image.clone();
cnt = pp.cnt;
}
break;
case 2:
{
img = image.clone();
cv::circle(img,pp.p1[0],3,cv::Scalar(255,0,0));
cv::putText(img,"(0,0,0)",pp.p1[0],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cnt = pp.cnt;
}
break;
case 3:
{
img = image.clone();
cv::circle(img,pp.p1[0],3,cv::Scalar(255,0,0));
cv::putText(img,"(0,0,0)",pp.p1[0],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cv::circle(img,pp.p1[1],3,cv::Scalar(255,0,0));
cv::putText(img,"(1.5,0,0)",pp.p1[1],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cnt = pp.cnt;
}
break;
case 4:
{
img = image.clone();
cv::circle(img,pp.p1[0],3,cv::Scalar(255,0,0));
cv::putText(img,"(0,0,0)",pp.p1[0],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cv::circle(img,pp.p1[1],3,cv::Scalar(255,0,0));
cv::putText(img,"(1.5,0,0)",pp.p1[1],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cv::circle(img,pp.p1[2],3,cv::Scalar(255,0,0));
cv::putText(img,"(0,1.5,0)",pp.p1[2],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cnt = pp.cnt;
}
break;
case 5:
{
img = image.clone();
cv::circle(img,pp.p1[0],3,cv::Scalar(255,0,0));
cv::putText(img,"(0,0,0)",pp.p1[0],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cv::circle(img,pp.p1[1],3,cv::Scalar(255,0,0));
cv::putText(img,"(1.5,0,0)",pp.p1[1],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cv::circle(img,pp.p1[2],3,cv::Scalar(255,0,0));
cv::putText(img,"(0,1.5,0)",pp.p1[2],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cv::circle(img,pp.p1[3],3,cv::Scalar(255,0,0));
cv::putText(img,"(0,0,1)",pp.p1[3],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cnt = pp.cnt;
}
break;
case 6:
{
img = image.clone();
cv::circle(img,pp.p1[0],3,cv::Scalar(255,0,0));
cv::putText(img,"(0,0,0)",pp.p1[0],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cv::circle(img,pp.p1[1],3,cv::Scalar(255,0,0));
cv::putText(img,"(1.5,0,0)",pp.p1[1],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cv::circle(img,pp.p1[2],3,cv::Scalar(255,0,0));
cv::putText(img,"(0,1.5,0)",pp.p1[2],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cv::circle(img,pp.p1[3],3,cv::Scalar(255,0,0));
cv::putText(img,"(0,0,1)",pp.p1[3],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cv::circle(img,pp.p1[4],3,cv::Scalar(255,0,0));
cv::putText(img,"(1.5,0,1)",pp.p1[4],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
cnt = pp.cnt;
}
break;
default:
break;
}
cv::imshow("Extrinsic Image", img);
cv::waitKey(10);
}
if (pp.clicked) // Case where we do a left click in order to insert a point
{
cv::circle(img,pp.p1[pp.cnt-1],3,cv::Scalar(255,0,0));
if (pp.cnt == 1) // First point to insert (0,0,0) (on the mocap basis)
{
cv::putText(img,"(0,0,0)",pp.p1[pp.cnt-1],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
// Get the image coordinates
u = pp.p1[0].x;
v = pp.p1[0].y;
Eigen::Vector3d undist;
undist(0) = u;
undist(1) = v;
undist(2) = 1.;
// Get the distance between the camera and the 3d point (the scale s)
if (cam == 1)
{
dist_to_point = DIST_TO_000_CAM1;
}
else if (cam == 2)
{
dist_to_point= DIST_TO_000_CAM2;
}
Pi(0,0) = u*dist_to_point;
Pi(1,0) = v*dist_to_point;
Pi(2,0) = dist_to_point;
Pr(3,0) = 1.;
} else if (pp.cnt == 2) // Second point to insert (500mm,0,0) (on the mocap basis)
{
cv::putText(img,"(1.5,0,0)",pp.p1[pp.cnt-1],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
u = pp.p1[1].x;
v = pp.p1[1].y;
Eigen::Vector3d undist;
undist(0) = u;
undist(1) = v;
undist(2) = 1.;
if (cam == 1)
{
dist_to_point = DIST_TO_0500_CAM1;
}
if (cam == 2)
{
dist_to_point = DIST_TO_0500_CAM2;
}
x = (u)*(dist_to_point);
y = (v)*(dist_to_point);
Pi(0,1) = x;
Pi(1,1) = y;
Pi(2,1) = dist_to_point;
Pr(0,1) = 1500.;
Pr(3,1) = 1.;
} else if (pp.cnt == 3) // Third point to insert (0,500mm,0) (on the mocap basis)
{
cv::putText(img,"(0,1.5,0)",pp.p1[pp.cnt-1],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
u = pp.p1[2].x;
v = pp.p1[2].y;
Eigen::Vector3d undist;
undist(0) = x;
undist(1) = y;
undist(2) = 1.;
if (cam == 1)
{
dist_to_point = DIST_TO_0050_CAM1;
}
if (cam == 2)
{
dist_to_point = DIST_TO_0050_CAM2;
}
x = (u)*(dist_to_point);
y = (v)*(dist_to_point);
Pi(0,2) = x;
Pi(1,2) = y;
Pi(2,2) = dist_to_point;
Pr(1,2) = 1500.;
Pr(3,2) = 1.;
} else if (pp.cnt == 4) // Fourth point to insert (0,0,1000mm) (on the mocap basis)
{
cv::putText(img,"(0,0,1)",pp.p1[pp.cnt-1],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
u = pp.p1[3].x;
v = pp.p1[3].y;
Eigen::Vector3d undist;
undist(0) = u;
undist(1) = v;
undist(2) = 1.;
if (cam == 1)
{
dist_to_point = DIST_TO_001_CAM1;
}
if (cam == 2)
{
dist_to_point = DIST_TO_001_CAM2;
}
x = (u)*(dist_to_point);
y = (v)*(dist_to_point);
Pi(0,3) = x;
Pi(1,3) = y;
Pi(2,3) = dist_to_point;
Pr(2,3) = 1000.;
Pr(3,3) = 1.;
}else if (pp.cnt == 5) // Fifth point to insert (1500mm,0,1000mm) (on the mocap basis)
{
cv::putText(img,"(1.5,0,1)",pp.p1[pp.cnt-1],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
u = pp.p1[4].x;
v = pp.p1[4].y;
Eigen::Vector3d undist;
undist(0) = u;
undist(1) = v;
undist(2) = 1.;
if (cam == 1)
{
dist_to_point = 4856.439;
}
if (cam == 2)
{
dist_to_point = 6333.64;
}
x = (u)*(dist_to_point);
y = (v)*(dist_to_point);
Pi(0,4) = x;
Pi(1,4) = y;
Pi(2,4) = dist_to_point;
Pr(0,4) = 1500.;
Pr(2,4) = 1000.;
Pr(3,4) = 1.;
}else if (pp.cnt == 6) // sixth point to insert (0,1500mm,1000mm) (on the mocap basis)
{
cv::putText(img,"(0,1.5,1)",pp.p1[pp.cnt-1],CV_FONT_HERSHEY_PLAIN|CV_FONT_ITALIC,1,cv::Scalar(255,0,0));
u = pp.p1[5].x;
v = pp.p1[5].y;
Eigen::Vector3d undist;
undist(0) = u;
undist(1) = v;
undist(2) = 1.;
if (cam == 1)
{
dist_to_point = 5142.713;
}
if (cam == 2)
{
dist_to_point = 4389.19;
}
x = (u)*(dist_to_point);
y = (v)*(dist_to_point);
Pi(0,5) = x;
Pi(1,5) = y;
Pi(2,5) = dist_to_point;
Pr(1,5) = 1500.;
Pr(2,5) = 1000.;
Pr(3,5) = 1.;
}
cnt = pp.cnt;
pp.clicked = false;
}
// Keep showing the image and the modification on it
cv::imshow("Extrinsic Image", img);
cv::waitKey(10);
}
// Compute the rotation matrix and the translation vector
Eigen::MatrixXd Prinv, Rt;
pseudoInverse(Pr,Prinv);
Rt.noalias() = (K.inverse())*Pi*(Prinv);
rotation(0,0) = Rt(0,0);rotation(0,1) = Rt(0,1);rotation(0,2) = Rt(0,2);
rotation(1,0) = Rt(1,0);rotation(1,1) = Rt(1,1);rotation(1,2) = Rt(1,2);
rotation(2,0) = Rt(2,0);rotation(2,1) = Rt(2,1);rotation(2,2) = Rt(2,2);
translation(0) = Rt(0,3);
translation(1) = Rt(1,3);
translation(2) = Rt(2,3);
// Save the values in files
if (cam == 1)
{
if (!fprintMatrix(rotation, "rotation_cam1.txt"))
std::cout << "Error writing in rotation_cam1.txt" << std::endl;
if (!fprintMatrix(translation, "translation_cam1.txt"))
std::cout << "Error writing in rotation_cam1.txt" << std::endl;
if(!fprintPT(pan,tilt, "PT_cam1.txt"))
std::cout << "Error writing file PT_cam1.txt" << std::endl;
} else if (cam == 2)
{
if (!fprintMatrix(rotation, "rotation_cam2.txt"))
std::cout << "Error writing in rotation_cam2.txt" << std::endl;
if (!fprintMatrix(translation, "translation_cam2.txt"))
std::cout << "Error writing in rotation_cam2.txt" << std::endl;
if(!fprintPT(pan,tilt, "PT_cam2.txt"))
std::cout << "Error writing file PT_cam2.txt" << std::endl;
}
initial_rotation.noalias() = rotation;
initial_translation.noalias() = translation;
rotation_computed = true;
// Destroy the window
cv::destroyWindow("Extrinsic Image");
}
bool PatternDetector::findPattern(const cv::Mat& image, PatternTrackingInfo& info)
{
// Convert input image to gray
getGray(image, m_grayImg);
// Extract feature points from input gray image
extractFeatures(m_grayImg, m_queryKeypoints, m_queryDescriptors);
// Get matches with current pattern
getMatches(m_queryDescriptors, m_matches);
#if _DEBUG
cv::showAndSave("Raw matches", getMatchesImage(image, m_pattern.frame, m_queryKeypoints, m_pattern.keypoints, m_matches, 100));
#endif
#if _DEBUG
cv::Mat tmp = image.clone();
#endif
// Find homography transformation and detect good matches
bool homographyFound = refineMatchesWithHomography(
m_queryKeypoints,
m_pattern.keypoints,
homographyReprojectionThreshold,
m_matches,
m_roughHomography);
if (homographyFound)
{
#if _DEBUG
cv::showAndSave("Refined matches using RANSAC", getMatchesImage(image, m_pattern.frame, m_queryKeypoints, m_pattern.keypoints, m_matches, 100));
#endif
// If homography refinement enabled improve found transformation
if (enableHomographyRefinement)
{
// Warp image using found homography
cv::warpPerspective(m_grayImg, m_warpedImg, m_roughHomography, m_pattern.size, cv::WARP_INVERSE_MAP | cv::INTER_CUBIC);
#if _DEBUG
cv::showAndSave("Warped image",m_warpedImg);
#endif
// Get refined matches:
std::vector<cv::KeyPoint> warpedKeypoints;
std::vector<cv::DMatch> refinedMatches;
// Detect features on warped image
extractFeatures(m_warpedImg, warpedKeypoints, m_queryDescriptors);
// Match with pattern
getMatches(m_queryDescriptors, refinedMatches);
// Estimate new refinement homography
homographyFound = refineMatchesWithHomography(
warpedKeypoints,
m_pattern.keypoints,
homographyReprojectionThreshold,
refinedMatches,
m_refinedHomography);
#if _DEBUG
cv::showAndSave("MatchesWithRefinedPose", getMatchesImage(m_warpedImg, m_pattern.grayImg, warpedKeypoints, m_pattern.keypoints, refinedMatches, 100));
#endif
// Get a result homography as result of matrix product of refined and rough homographies:
info.homography = m_roughHomography * m_refinedHomography;
// Transform contour with rough homography
#if _DEBUG
cv::perspectiveTransform(m_pattern.points2d, info.points2d, m_roughHomography);
info.draw2dContour(tmp, CV_RGB(0,200,0));
#endif
// Transform contour with precise homography
cv::perspectiveTransform(m_pattern.points2d, info.points2d, info.homography);
#if _DEBUG
info.draw2dContour(tmp, CV_RGB(200,0,0));
#endif
}
else
{
info.homography = m_roughHomography;
// Transform contour with rough homography
cv::perspectiveTransform(m_pattern.points2d, info.points2d, m_roughHomography);
#if _DEBUG
info.draw2dContour(tmp, CV_RGB(0,200,0));
#endif
}
}
#if _DEBUG
if (1)
{
cv::showAndSave("Final matches", getMatchesImage(tmp, m_pattern.frame, m_queryKeypoints, m_pattern.keypoints, m_matches, 100));
}
std::cout << "Features:" << std::setw(4) << m_queryKeypoints.size() << " Matches: " << std::setw(4) << m_matches.size() << std::endl;
#endif
return homographyFound;
}
void reset_output() const {
#ifdef REAK_HAS_OPENCV
world_map_output = world_map_image.clone();
#endif
};
bool Stabilizer::init( const cv::Mat& frame)
{
prevFrame = frame.clone();
return true;
}
bool Stabilizer::track(const cv::Mat& frame)
{
cv::Mat previous_frame_gray;
cv::cvtColor(prevFrame, previous_frame_gray, cv::COLOR_BGR2GRAY);
size_t n = previousFeatures.size();
CV_Assert(n);
if (flagUpdateFeatures) {
cv::goodFeaturesToTrack(previous_frame_gray, previousFeatures, 500, 0.01, 5);
flagUpdateFeatures = false;
}
// Compute optical flow in selected points.
std::vector<cv::Point2f> currentFeatures;
std::vector<uchar> state;
std::vector<float> error;
cv::calcOpticalFlowPyrLK(prevFrame, frame, previousFeatures, currentFeatures, state, error);
float median_error = median<float>(error);
std::vector<cv::Point2f> good_points;
std::vector<cv::Point2f> curr_points;
for (size_t i = 0; i < n; ++i)
{
if (state[i] && (error[i] <= median_error))
{
good_points.push_back(previousFeatures[i]);
curr_points.push_back(currentFeatures[i]);
}
}
size_t s = good_points.size();
CV_Assert(s == curr_points.size());
// Find points shift.
std::vector<float> shifts_x(s);
std::vector<float> shifts_y(s);
for (size_t i = 0; i < s; ++i)
{
shifts_x[i] = curr_points[i].x - good_points[i].x;
shifts_y[i] = curr_points[i].y - good_points[i].y;
}
std::sort(shifts_x.begin(), shifts_x.end());
std::sort(shifts_y.begin(), shifts_y.end());
// Find median shift.
cv::Point2f median_shift(shifts_x[s / 2], shifts_y[s / 2]);
xshift.push_back(median_shift.x);
yshift.push_back(median_shift.y);
if (s < 100) {
previousFeatures.clear();
previousFeatures = currentFeatures;
flagUpdateFeatures = true;
}
prevFrame = frame.clone();
return true;
}
// SETTERS #############################################################
void Sakbot::setImage( cv::Mat current )
{
m_image = current.clone();
}
void drawImageWithLock(cv::Mat &disp)
{
disp = color.clone();
}
bool ddbtc::compress(cv::Mat &src,cv::Mat &dst,short BlockSize){
if(BlockSize!=8&&BlockSize!=16){ // current version supports only 8 and 16
return false;
}
if(src.type()!=CV_8U){ // should be grayscale image
return false;
}
// = = = = = pre-defined data = = = = = //
int &imgWidth=src.cols,&imgHeight=src.rows;
short CM_Size=BlockSize; // class matrix size
const short DM_Size=3; // diffused weighting size
const short CM8[8][8]={ {42, 47, 46, 45, 16, 13, 11, 2},
{61, 57, 53, 8, 27, 22, 9, 50},
{63, 58, 0, 15, 26, 31, 40, 30},
{10, 4, 17, 21, 3, 44, 18, 6},
{14, 24, 25, 7, 5, 48, 52, 39},
{20, 28, 23, 32, 38, 51, 54, 60},
{19, 33, 36, 37, 49, 43, 56, 55},
{12, 62, 29, 35, 1, 59, 41, 34}};
const float DW8[3][3]={ {0.271630, 1.000000, 0.271630},
{1.000000, 0.000000, 1.000000},
{0.271630, 1.000000, 0.271630}};
const short CM16[16][16]={ {6, 7 ,20 ,10 ,53 ,55 ,66 , 87 , 137, 142, 143, 144, 172, 122, 175, 164},
{3, 9 ,23 ,50 ,60 ,51 ,65 , 74, 130, 145, 138, 148, 179, 180, 214, 221},
{0, 14 ,24 ,37 ,67 ,79 ,96 , 116, 39, 149, 162, 198, 12, 146, 224, 1},
{15,26 ,43 ,28 ,71 ,54 ,128, 112, 78, 159, 177, 201, 208, 223, 225, 242},
{22,4 ,48 ,32 ,94 ,98 ,80 , 135, 157, 173, 113, 182, 222, 226, 227, 16},
{40,85 ,72 ,83 ,104 ,117 ,163, 133, 168, 184, 200, 219, 244, 237, 183, 21},
{47,120 ,101 ,105 ,123 ,132 ,170, 176, 190, 202, 220, 230, 245, 235, 17, 41},
{76,73 ,127 ,109 ,97 ,134 ,178, 181, 206, 196, 229, 231, 246, 19, 42, 49},
{103,99 ,131 ,147 ,169 ,171 ,166, 203, 218, 232, 243, 248, 247, 33, 52, 68},
{108,107, 140 ,102 ,185 ,167 ,204, 217, 233, 106, 249, 255, 44, 45, 70, 69},
{110,141, 88 ,75 ,192 ,205 ,195, 234, 241, 250, 254, 38, 46, 77, 5, 100},
{111,158, 160 ,174 ,119 ,215 ,207, 240, 251, 252, 253, 61, 62, 93, 84, 125},
{151,136, 189 ,199 ,197 ,216 ,236, 239, 25, 31, 56, 82, 92, 95, 124, 114},
{156,188, 191 ,209 ,213 ,228 ,238, 29, 36, 59, 64, 91, 118, 139, 115, 155},
{187,194, 165 ,212 ,2 ,13 ,30 , 35, 58, 63, 90, 86, 152, 129, 154, 161},
{193,210, 211 ,8 ,11 ,27 ,34 , 57, 18, 89, 81, 121, 126, 153, 150, 186}};
const float DW16[3][3]={{0.305032, 1.000000, 0.305032},
{1.000000, 0.000000, 1.000000},
{0.305032, 1.000000, 0.305032}};
// = = = = = give cm and dw data = = = = = //
std::vector<std::vector<short>> CM(CM_Size,std::vector<short>(CM_Size,0));
std::vector<std::vector<float>> DM(DM_Size,std::vector<float>(DM_Size,0));
if(BlockSize==8){
for(int i=0;i<CM_Size;i++){
for(int j=0;j<CM_Size;j++){
CM[i][j]=CM8[i][j];
}
}
for(int i=0;i<DM_Size;i++){
for(int j=0;j<DM_Size;j++){
DM[i][j]=DW8[i][j];
}
}
}else if(BlockSize==16){
for(int i=0;i<CM_Size;i++){
for(int j=0;j<CM_Size;j++){
CM[i][j]=CM16[i][j];
}
}
for(int i=0;i<DM_Size;i++){
for(int j=0;j<DM_Size;j++){
DM[i][j]=DW16[i][j];
}
}
}
// = = = = = create Temp space = = = = = //
std::vector<std::vector<float>> Tempmap(imgHeight,std::vector<float>(imgWidth,0));
for(int i=0;i<imgHeight;i++){
for(int j=0;j<imgWidth;j++){
Tempmap[i][j]=src.data[i*src.cols+j];
}
}
// = = = = = get processing positions = = = = = //
std::vector<std::vector<short>> ProPo(CM_Size*CM_Size,std::vector<short>(2,0));
for(int m=0;m<CM_Size;m++){
for(int n=0;n<CM_Size;n++){
ProPo[CM[m][n]][0]=m;
ProPo[CM[m][n]][1]=n;
}
}
// = = = = = get bitmap = = = = = //
std::vector<std::vector<char>> DoMap(imgHeight,std::vector<char>(imgWidth,false)); // it is originally the bool, not the current char, because that the process of vector<bool> is "very" slow.
//////////////////////////////////////////////////////////////////////////
// initialization
cv::Mat tdst; // temp dst
tdst=src.clone();
// = = = = = process = = = = = //
for(int i=0;i<imgHeight;i+=CM_Size){
for(int j=0;j<imgWidth;j+=CM_Size){
//////////////////////////////////////////////////////////////////////////
// calculate the local mean, maxv and minv or original image's block
float mean=0.;
uchar maxv=tdst.data[i*tdst.cols+j],
minv=tdst.data[i*tdst.cols+j];
short count_mean=0;
for(int m=0;m<CM_Size;m++){
for(int n=0;n<CM_Size;n++){
if(i+m>=0&&i+m<imgHeight&&j+n>=0&&j+n<imgWidth){
count_mean++;
mean+=(float)tdst.data[(i+m)*tdst.cols+(j+n)];
if(tdst.data[(i+m)*tdst.cols+(j+n)]>maxv){
maxv=tdst.data[(i+m)*tdst.cols+(j+n)];
}
if(tdst.data[(i+m)*tdst.cols+(j+n)]<minv){
minv=tdst.data[(i+m)*tdst.cols+(j+n)];
}
}
}
}
mean/=(float)count_mean;
//////////////////////////////////////////////////////////////////////////
// = = = = = diffusion = = = = = //
short memberIndex=0;
while(memberIndex!=CM_Size*CM_Size){
// to decide whether the prospective coordinate is out of scope or not
int ni=i+ProPo[memberIndex][0],nj=j+ProPo[memberIndex][1];
if(ni>=0&&ni<imgHeight&&nj>=0&&nj<imgWidth){
// = = = = = dot diffusion = = = = = //
// get error // for maintain the dot diffusion structure, here have to take DifMap into account
float error;
if(Tempmap[ni][nj]<mean){ // y = max or min (determined by the bitmap)
tdst.data[ni*tdst.cols+nj]=minv;
}else{
tdst.data[ni*tdst.cols+nj]=maxv;
}
error=Tempmap[ni][nj]-(float)tdst.data[ni*tdst.cols+nj]; // e = v - y
DoMap[ni][nj]=true;
// get fm
double fm=0;
short hDM_Size=DM_Size/2;
for(int m=-hDM_Size;m<=hDM_Size;m++){
for(int n=-hDM_Size;n<=hDM_Size;n++){
if(ni+m>=0&&ni+m<imgHeight&&nj+n>=0&&nj+n<imgWidth){
if(DoMap[ni+m][nj+n]==false){
fm+=DM[m+hDM_Size][n+hDM_Size];
}
}
}
}
// diffusing
for(int m=-hDM_Size;m<=hDM_Size;m++){
for(int n=-hDM_Size;n<=hDM_Size;n++){
if(ni+m>=0&&ni+m<imgHeight&&nj+n>=0&&nj+n<imgWidth){
if(DoMap[ni+m][nj+n]==false){
Tempmap[ni+m][nj+n]+=error *DM[m+hDM_Size][n+hDM_Size]/fm; // v = x + e*weight
}
}
}
}
}
memberIndex++;
}
}
}
dst = tdst.clone();
return true;
}
void LayoutTest::testLayout(const cv::Mat & src) const {
// TODOS
// - line spacing needs smoothing -> graphcut
// - DBScan is very sensitive to the line spacing
// Workflow:
// - implement noise/text etc classification on SuperPixel level
// - smooth labels using graphcut
// - perform everything else without noise pixels
// Training:
// - open mode (whole image only contains e.g. machine printed)
// - baseline mode -> overlap with superpixel
cv::Mat img = src.clone();
//cv::resize(src, img, cv::Size(), 0.25, 0.25, CV_INTER_AREA);
Timer dt;
// find super pixels
rdf::SuperPixel superPixel(img);
if (!superPixel.compute())
qWarning() << "could not compute super pixel!";
QVector<QSharedPointer<Pixel> > sp = superPixel.getSuperPixels();
// find local orientation per pixel
rdf::LocalOrientation lo(sp);
if (!lo.compute())
qWarning() << "could not compute local orientation";
// smooth estimation
rdf::GraphCutOrientation pse(sp);
if (!pse.compute())
qWarning() << "could not compute set orientation";
// pixel labeling
QSharedPointer<SuperPixelModel> model = SuperPixelModel::read(mConfig.classifierPath());
//FeatureCollectionManager fcm = FeatureCollectionManager::read(mConfig.featureCachePath());
//// train classifier
//SuperPixelTrainer spt(fcm);
//if (!spt.compute())
// qCritical() << "could not train data...";
//auto model = spt.model();
SuperPixelClassifier spc(src, sp);
spc.setModel(model);
if (!spc.compute())
qWarning() << "could not classify SuperPixels";
//// find tab stops
//rdf::TabStopAnalysis tabStops(sp);
//if (!tabStops.compute())
// qWarning() << "could not compute text block segmentation!";
//// find text lines
//rdf::TextLineSegmentation textLines(sp);
//textLines.addLines(tabStops.tabStopLines(30)); // TODO: fix parameter
//if (!textLines.compute())
// qWarning() << "could not compute text block segmentation!";
qInfo() << "algorithm computation time" << dt;
// drawing
//cv::Mat rImg(img.rows, img.cols, CV_8UC1, cv::Scalar::all(150));
cv::Mat rImg = img.clone();
//// draw edges
//rImg = textBlocks.draw(rImg);
//rImg = lo.draw(rImg, "1012", 256);
//rImg = lo.draw(rImg, "507", 128);
//rImg = lo.draw(rImg, "507", 64);
//// save super pixel image
//rImg = superPixel.drawSuperPixels(rImg);
//rImg = tabStops.draw(rImg);
//rImg = textLines.draw(rImg);
rImg = spc.draw(rImg);
QString maskPath = rdf::Utils::instance().createFilePath(mConfig.outputPath(), "-classified");
rdf::Image::save(rImg, maskPath);
qDebug() << "debug image added" << maskPath;
//// write XML -----------------------------------
//QString loadXmlPath = rdf::PageXmlParser::imagePathToXmlPath(mConfig.imagePath());
//rdf::PageXmlParser parser;
//parser.read(loadXmlPath);
//auto pe = parser.page();
//pe->setCreator(QString("CVL"));
//pe->setImageSize(QSize(img.rows, img.cols));
//pe->setImageFileName(QFileInfo(mConfig.imagePath()).fileName());
//// start writing content
//auto ps = PixelSet::fromEdges(PixelSet::connect(sp, Rect(0, 0, img.cols, img.rows)));
//if (!ps.empty()) {
// QSharedPointer<Region> textRegion = QSharedPointer<Region>(new Region());
// textRegion->setType(Region::type_text_region);
// textRegion->setPolygon(ps[0]->convexHull());
//
// for (auto tl : textLines.textLines()) {
// textRegion->addUniqueChild(tl);
// }
// pe->rootRegion()->addUniqueChild(textRegion);
//}
//parser.write(mConfig.xmlPath(), pe);
//qDebug() << "results written to" << mConfig.xmlPath();
}
bool ShapeDiscriptor::discribeImage(cv::Mat &image)
{
std::vector< std::vector<cv::Point> > contours;
cv::Mat timage = image.clone();
printf("rows : %d\n", image.rows);
findContours(timage, contours, CV_RETR_TREE, CV_CHAIN_APPROX_SIMPLE);
for(int contourIdx = 0; contourIdx < contours.size(); contourIdx++)
{
cv::Moments moms = moments(cv::Mat(contours[contourIdx]));
if(doFilterArea)
{
double area = moms.m00;
if (area < minArea || area > maxArea)
//drawContours(image, contours, contourIdx, cv::Scalar(0,0,0), CV_FILLED);
continue;
}
if(doFilterCircularity)
{
double area = moms.m00;
double perimeter = cv::arcLength(cv::Mat(contours[contourIdx]), true);
double ratio = 4 * CV_PI * area / (perimeter * perimeter);
if (ratio < minCircularity)
//drawContours(image, contours, contourIdx, cv::Scalar(0,0,0), CV_FILLED);
continue;
}
if(doFilterInertia)
{
double denominator = sqrt(pow(2 * moms.mu11, 2) + pow(moms.mu20 - moms.mu02, 2));
const double eps = 1e-2;
double ratio;
if (denominator > eps)
{
double cosmin = (moms.mu20 - moms.mu02) / denominator;
double sinmin = 2 * moms.mu11 / denominator;
double cosmax = -cosmin;
double sinmax = -sinmin;
double imin = 0.5 * (moms.mu20 + moms.mu02) - 0.5 * (moms.mu20 - moms.mu02) * cosmin - moms.mu11 * sinmin;
double imax = 0.5 * (moms.mu20 + moms.mu02) - 0.5 * (moms.mu20 - moms.mu02) * cosmax - moms.mu11 * sinmax;
ratio = imin / imax;
}
else
{
ratio = 1;
}
//p.inErtia = ratio;
if (ratio < minInertia)
//drawContours(image, contours, contourIdx, cv::Scalar(0,0,0), CV_FILLED);
continue;
}
if(doFilterConvexity)
{
cv::vector < cv::Point > hull;
convexHull(cv::Mat(contours[contourIdx]), hull);
double area = cv::contourArea(cv::Mat(contours[contourIdx]));
double hullArea = cv::contourArea(cv::Mat(hull));
double ratio = area / hullArea;
//p.convexity = ratio;
if (ratio < minConvexity)
//drawContours(image, contours, contourIdx, cv::Scalar(0,0,0), CV_FILLED);
continue;
}
timage.release();
return true;
}
timage.release();
return false;
}
void ObjectDetectorProvider::processRequest(
const re_vision::SearchFor::Request &req,
const cv::Mat &image,
re_vision::SearchFor::Response &res)
{
ROS_DEBUG("Processing request...");
cv::Mat _image = image.clone(); // ###
ros::WallTime t_begin = ros::WallTime::now();
vector<string> valid_objects;
vector<re_msgs::DetectedObject *> pointers;
getValidObjects(req.Objects, valid_objects, res.Detections, pointers);
detectObjects(valid_objects, image, req.MaxPointsPerObject, pointers);
rectifyDetections(res.Detections, image.cols, image.rows,
req.MaxPointsPerObject);
#if 0
if(!res.Detections.empty() && !res.Detections[0].points2d.empty())
{
vector<cv::Point3f> points3d;
for(unsigned int i = 0; i < res.Detections[0].points3d.size(); ++i)
{
points3d.push_back(cv::Point3f(
-res.Detections[0].points3d[i].y,
-res.Detections[0].points3d[i].z,
res.Detections[0].points3d[i].x));
}
cv::Mat cP(points3d);
cv::Mat R = cv::Mat::eye(3,3,CV_64F);
cv::Mat t = cv::Mat::zeros(3,1,CV_64F);
vector<cv::Point2f> cam_pixels;
cv::projectPoints(cP, R, t, m_camera.GetIntrinsicParameters(),
cv::Mat::zeros(4,1,CV_32F), cam_pixels);
CvScalar color = cvScalar(255, 255, 255);
for(unsigned int i = 0; i < cam_pixels.size(); ++i)
{
cv::circle(_image, cvPoint(cam_pixels[i].x, cam_pixels[i].y),
2, color, 1);
}
cv::imwrite("_debug.png", _image);
}
#endif
ros::WallTime t_end = ros::WallTime::now();
{
int counter = 0;
vector<re_msgs::DetectedObject>::const_iterator dit;
for(dit = res.Detections.begin(); dit != res.Detections.end(); ++dit)
{
if(!dit->points3d.empty()) counter++;
}
ROS_INFO("%d objects detected", counter);
}
ros::WallDuration d = t_end - t_begin;
ROS_DEBUG("Processing ok. Elapsed time: %f", d.toSec());
}
vavImage::vavImage(const cv::Mat& im): m_pTextureDraw(0), m_pShaderResView(0)
{
m_Image = im.clone();
}
cv::Mat Local_Scharr(cv::Mat input_img,int sub_images,bool using_histogram_percentile,int histogram_percentile,bool using_threshold,int threshold, bool dx_checked,bool dy_checked,bool Otsu,double x_weight, double y_weight)
{
cv::Mat output_img;
if((input_img.channels()==1) && (!input_img.empty()))
{
int colsize = input_img.cols;
int rowsize = input_img.rows;
cv::Mat sub_image,sub_image2,edge_img;
cv::Mat seg_dx,seg_dy;
cv::Mat abs_seg_x,abs_seg_y;
edge_img = input_img.clone();
if(!(dx_checked & dy_checked))
{
edge_img.convertTo(edge_img,CV_32F);
sub_image.create(rowsize/sub_images,colsize/sub_images,CV_32F);
sub_image2.create(rowsize/sub_images,colsize/sub_images,CV_32F);
}
else
{
sub_image.create(rowsize/sub_images,colsize/sub_images,CV_8U);
sub_image2.create(rowsize/sub_images,colsize/sub_images,CV_8U);
}
int c,d,e,f;
for(int a = 0;a<sub_images;a++)
{
if(a == 0)
{
c=0;
d = floor(1.0*rowsize/sub_images)-1;
}else
{
c = d+1;
d = d+floor(1.0*rowsize/sub_images);
}
for(int b = 0;b<sub_images;b++)
{
if(b == 0)
{
e=0;
f = floor(1.0*colsize/sub_images)-1;
}else
{
e = f+1;
f = f+floor(1.0*colsize/sub_images);
}
sub_image = edge_img(cv::Rect(e,c,sub_image.cols,sub_image.rows));
sub_image.copyTo(sub_image2);
if(dx_checked && dy_checked)
{
cv::Sobel(sub_image2,seg_dx,CV_32F,1,0,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
cv::Sobel(sub_image2,seg_dy,CV_32F,0,1,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
cv::convertScaleAbs(seg_dx,abs_seg_x);
cv::convertScaleAbs(seg_dy,abs_seg_y);
cv::addWeighted(abs_seg_x,x_weight,abs_seg_y,y_weight,0,sub_image);//not perfect, can adjust dx and dy
}
else if(dx_checked)
{
cv::Sobel(sub_image2,sub_image,CV_32F,1,0,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
}
else if(dy_checked)
{
cv::Sobel(sub_image2,sub_image,CV_32F,0,1,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
}
}
}
if(((rowsize-1)-d)>0)//adds missing rows
{
sub_image.create((rowsize-1)-d,colsize/sub_images,CV_8U);
sub_image2.create((rowsize-1)-d,colsize/sub_images,CV_8U);
c = d+1;
for(int b = 0;b<sub_images;b++)
{
if(b == 0)
{
e=0;
f = floor(1.0*colsize/sub_images)-1;
}else
{
e = f+1;
f = f+floor(1.0*colsize/sub_images);
}
sub_image = edge_img(cv::Rect(e,c,sub_image.cols,sub_image.rows));
sub_image.copyTo(sub_image2);
if(dx_checked && dy_checked)
{
cv::Sobel(sub_image2,seg_dx,CV_32F,1,0,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
cv::Sobel(sub_image2,seg_dy,CV_32F,0,1,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
cv::convertScaleAbs(seg_dx,abs_seg_x);
cv::convertScaleAbs(seg_dy,abs_seg_y);
cv::addWeighted(abs_seg_x,x_weight,abs_seg_y,y_weight,0,sub_image);//not perfect, can adjust dx and dy
}
else if(dx_checked)
{
cv::Sobel(sub_image2,sub_image,CV_32F,1,0,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
}
else if(dy_checked)
{
cv::Sobel(sub_image2,sub_image,CV_32F,0,1,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
}
}
}
if(((colsize-1)-d)>0)//adds missing coloumns
{
sub_image.create(rowsize/sub_images,(colsize-1)-f,CV_8U);
sub_image2.create(rowsize/sub_images,(colsize-1)-f,CV_8U);
e = f+1;
for(int a = 0;a<sub_images;a++)
{
if(a == 0)
{
c=0;
d = floor(1.0*rowsize/sub_images)-1;
}else
{
c = d+1;
d = d+floor(1.0*rowsize/sub_images);
}
sub_image = edge_img(cv::Rect(e,c,sub_image.cols,sub_image.rows));
sub_image.copyTo(sub_image2);
if(dx_checked && dy_checked)
{
cv::Sobel(sub_image2,seg_dx,CV_32F,1,0,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
cv::Sobel(sub_image2,seg_dy,CV_32F,0,1,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
cv::convertScaleAbs(seg_dx,abs_seg_x);
cv::convertScaleAbs(seg_dy,abs_seg_y);
cv::addWeighted(abs_seg_x,x_weight,abs_seg_y,y_weight,0,sub_image);//not perfect, can adjust dx and dy
}
else if(dx_checked)
{
cv::Sobel(sub_image2,sub_image,CV_32F,1,0,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
}
else if(dy_checked)
{
cv::Sobel(sub_image2,sub_image,CV_32F,0,1,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
}
}
}
if((((colsize-1)-d)>0) && (((rowsize-1)-d)>0))
{
sub_image.create((rowsize-1)-d,(colsize-1)-f,CV_8U);
sub_image2.create((rowsize-1)-d,colsize-1-f,CV_8U);
c = d+1;
e = f+1;
sub_image = edge_img(cv::Rect(e,c,sub_image.cols,sub_image.rows));
sub_image.copyTo(sub_image2);
if(dx_checked && dy_checked)
{
cv::Sobel(sub_image2,seg_dx,CV_32F,1,0,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
cv::Sobel(sub_image2,seg_dy,CV_32F,0,1,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
cv::convertScaleAbs(seg_dx,abs_seg_x);
cv::convertScaleAbs(seg_dy,abs_seg_y);
cv::addWeighted(abs_seg_x,x_weight,abs_seg_y,y_weight,0,sub_image);//not perfect, can adjust dx and dy
}
else if(dx_checked)
{
cv::Sobel(sub_image2,sub_image,CV_32F,1,0,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
}
else if(dy_checked)
{
cv::Sobel(sub_image2,sub_image,CV_32F,0,1,CV_SCHARR,1,0,cv::BORDER_REPLICATE);
}
}
if(!(dx_checked & dy_checked))
{
edge_img.convertTo(edge_img,CV_8U);
}
if(Otsu)
{
output_img = Histogram_seg(histogram_percentile,edge_img,input_img);
}
else if(using_histogram_percentile)
{
output_img = Histogram_seg(histogram_percentile,edge_img);
}
else if(using_threshold)
{
int num_pix = edge_img.rows*edge_img.cols;
edge_img.copyTo(output_img);
output_img.create(edge_img.rows,edge_img.cols,edge_img.type());
uchar* data = output_img.ptr<uchar>(0);
uchar* old_data = edge_img.ptr<uchar>(0);
for(int i = 0; i<num_pix;i++)
{
if(old_data[i]>threshold)
{
data[i]=255;
}
else
{
data[i] = 0;
}
}
}
}
return output_img;
}
// Calculates the Gaussian Blur and stores the result on the height map given
cv::Mat GaussianBlur(cv::Mat img,int gaussian_kernel_size,float segma)
{
if(gaussian_kernel_size%2 == 0)
{
cout << "Use odd size for gaussian kernel ";
return img;
}
if(img.cols!=img.rows){
cout << "Use Image with same width and height";
return img;
}
if(img.cols%2 != 0){
cout << "Use Image with size 2 power n";
return img;
}
// init the 1D gaussian Kernel using open CV
cv::Mat gussian_kernal=cv::getGaussianKernel(gaussian_kernel_size,segma);
print_var("g_ker",gussian_kernal);
//init the variables
cv::Mat blur_img_x=img.clone();
cv::Mat final_blur_img=img.clone();
int img_size = img.rows;
int kernal_radius=int((gaussian_kernel_size-1)/2);
int start_pixel_x=kernal_radius;
int start_pixel_y=kernal_radius;
int end_pixel_x= img_size - kernal_radius;
int end_pixel_y= img_size - kernal_radius;
double temp_sum_res=0;
int temp_kernal_shift=0;
int temp_pixcel_val=0;
double temp_gussian_val=0;
//apply 1D filter on x direction of the image
for (int y=start_pixel_y;y<end_pixel_y;y++)
{
for(int x=start_pixel_x;x<end_pixel_x;x++)
{
temp_sum_res=0;
temp_kernal_shift=0;
for (int kernal_shift=x-kernal_radius;kernal_shift< x+kernal_radius+1;kernal_shift++)
{
temp_pixcel_val=img.at<uchar>(kernal_shift,y);
temp_gussian_val=gussian_kernal.at<double>(temp_kernal_shift,0);
temp_sum_res+=temp_pixcel_val*temp_gussian_val;
temp_kernal_shift++;
}
blur_img_x.at<uchar>(x,y)=temp_sum_res;
}
}
//apply 1D filter on the the result blurred_x image on Y direction
for (int y=start_pixel_y;y<end_pixel_y;y++)
{
for(int x=start_pixel_x;x<end_pixel_x;x++)
{
temp_sum_res=0;
temp_kernal_shift=0;
for (int kernal_shift=y-kernal_radius;kernal_shift< y+kernal_radius+1;kernal_shift++)
{
temp_pixcel_val=blur_img_x.at<uchar>(x,kernal_shift);
temp_gussian_val=gussian_kernal.at<double>(temp_kernal_shift,0);
temp_sum_res+=temp_pixcel_val*temp_gussian_val;
temp_kernal_shift++;
}
final_blur_img.at<uchar>(x,y)=temp_sum_res;
}
}
return final_blur_img;
}
