void Labeler::convert2Example(const Instance* pInstance, Example& exam) {
exam.clear();
const vector<string> &label1s = pInstance->label1s;
const vector<string> &label2s = pInstance->label2s;
int curInstSize = label1s.size();
for (int i = 0; i < curInstSize; ++i) {
string orcale1 = label1s[i];
string orcale2 = label2s[i];
if (orcale2.length() > 2)
orcale2 = orcale2.substr(2);
int numLabel2s = m_label2Alphabet.size();
vector<int> curlabel2s;
for (int j = 0; j < numLabel2s; ++j) {
string str = m_label2Alphabet.from_id(j);
if (str.compare(orcale2) == 0)
curlabel2s.push_back(1);
else
curlabel2s.push_back(0);
}
exam.m_label2s.push_back(curlabel2s);
Feature feat;
extractFeature(feat, pInstance, i);
exam.m_features.push_back(feat);
}
}
void Labeler::convert2Example(const Instance* pInstance, Example& exam) {
exam.clear();
const string &label = pInstance->label;
const vector<vector<string> > &words = pInstance->words;
const vector<vector<vector<string> > > &chars = pInstance->chars;
int numLabels = m_labelAlphabet.size();
for (int j = 0; j < numLabels; ++j) {
string str = m_labelAlphabet.from_id(j);
if (str.compare(label) == 0)
exam.m_labels.push_back(1);
else
exam.m_labels.push_back(0);
}
int curInstSize = words.size();
for (int i = 0; i < curInstSize; ++i) {
Feature feat;
extractFeature(feat, pInstance, i);
exam.m_features.push_back(feat);
}
if (m_linearfeat > 0) {
vector<string> linear_features;
extractLinearFeatures(linear_features, pInstance);
for (int i = 0; i < linear_features.size(); i++) {
int curFeatId = m_featAlphabet.from_string(linear_features[i]);
if (curFeatId >= 0)
exam.m_linears.push_back(curFeatId);
}
}
}
int PlateJudge::plateJudge(const Mat &inMat, int &result) {
Mat features;
extractFeature(inMat, features);
float response = svm_->predict(features);
result = (int)response;
return 0;
}
int HaarFeature::extractOneTypeFeatures45(int is_extract_feature, HaarFeatureInfoT &info, float *pt_feature)
{
int type = info.type;
int L_r = feature_sizes[type].x;
int R_r = feature_sizes[type].y;
double inv_r = feature_inv_ratio[type] * 2;
CB_PointT image_size = {template_w, template_h};
info.pos1.y = FEATURE_MARGIN;
while (info.pos1.y < template_h - 1)
{
info.pos1.x = FEATURE_MARGIN;
while (info.pos1.x < template_w - 1)
{
info.size.y = 2;
while (info.pos1.x + R_r * info.size.y < template_w)
{
info.size.x = 2;
while (info.pos1.x - L_r * info.size.x > 0)
{
CB_SlantT slant = {info.pos1.x, info.pos1.y, L_r * info.size.x, R_r * info.size.y};
CB_RectangleT rect;
int result = slantToRect(slant, rect, image_size);
if (result == 0)
{
info.size.x += HAAR_SCALE_STEP_X;
continue;
}
if (is_extract_feature == 0)
{
info.inv_area = INV_AREA_R / (info.size.x * info.size.y * inv_r);
pt_feature_infos[feature_count] = info;
}
else if (is_extract_feature > 0)
{
*(pt_feature + feature_count) = extractFeature(pt_feature_infos[feature_count]);
}
feature_count++;
info.size.x += HAAR_SCALE_STEP_X;
}
info.size.y += HAAR_SCALE_STEP_Y;
}
info.pos1.x += HAAR_SHIFT_STEP_X;
}
info.pos1.y += HAAR_SHIFT_STEP_Y;
}
return feature_count;
}
//! 设置车牌图像的置信度
//! 返回值，0代表是车牌，其他值代表不是
int PlateJudge::plateSetScore(CPlate& plate) {
Mat features;
extractFeature(plate.getPlateMat(), features);
float score = svm_->predict(features, noArray(), cv::ml::StatModel::Flags::RAW_OUTPUT);
// score值代表离margin的距离，小于0代表是车牌，大于0代表不是车牌
// 当小于0时，值越小代表是车牌的概率越大
plate.setPlateScore(score);
if (score < 0)
return 0;
else
return -1;
}
void classifierSVM(Mat& trainData, Mat& labelsTrain, vector<string> trainImgName, vector<string> testImgName, string kernelFunction, Mat& resultTrain, Mat& resultTest)
{
Mat trainFeatureData(0,1764, CV_32FC1);
Mat testFeatureData(0,1764, CV_32FC1);
string predictTrainFeatureVector="./result/PredictTrain-Features-LBP.txt";
string predictTestFeatureVector="./result/PredictTest-Features-LBP.txt";
stringstream predictTrainDataTxt;
stringstream predictTestDataTxt;
SVM::Params params;
params.svmType = SVM::C_SVC;//C_SVC用于n类分类问题
if( kernelFunction == "LINEAR" )
params.kernelType = SVM::LINEAR;
if( kernelFunction == "RBF" )
params.kernelType = SVM::RBF;
Ptr<SVM> svm = SVM::create(params);
svm->train( trainData , ROW_SAMPLE , labelsTrain );
/*----------------Predict traindata------------------*/
// Mat resultTrain; // 输出分类结果
trainFeatureData=extractFeature(trainImgName);
WriteData(predictTrainFeatureVector, trainFeatureData);
predictTrainDataTxt<<"./result/PredictTrain-LBP-SVM-"<<kernelFunction<<".txt";
svm->predict(trainFeatureData, resultTrain);
WriteData(predictTrainDataTxt.str(), resultTrain);
/*----------------Predict testdata-------------------*/
// Mat resultTest; // 输出分类结果
testFeatureData=extractFeature(testImgName);
WriteData(predictTestFeatureVector, testFeatureData);
predictTestDataTxt<<"./result/PredictTest-LBP-SVM-"<<kernelFunction<<".txt";
svm->predict(testFeatureData, resultTest);
WriteData(predictTestDataTxt.str(), resultTest);
}
void GeoJson::extractLayer(const rapidjson::Value& _in, Layer& _out, const MapTile& _tile) {
const auto& featureIter = _in.FindMember("features");
if (featureIter == _in.MemberEnd()) {
logMsg("ERROR: GeoJSON missing 'features' member\n");
return;
}
const auto& features = featureIter->value;
for (auto featureJson = features.Begin(); featureJson != features.End(); ++featureJson) {
_out.features.emplace_back();
extractFeature(*featureJson, _out.features.back(), _tile);
}
}
void ImageProcess::run()
{
switch(processStep)
{
case 1:
//提取特征
extractFeature();
break;
case 2:
//匹配特征
matchFeature();
break;
case 3:
//对齐影像
alignImage();
break;
}
}
ControlWidget::ControlWidget(MainWindow* mainWin) : QDockWidget("Control Widget", (QWidget*)mainWin) {
this->mainWin = mainWin;
// set up the UI
ui.setupUi(this);
ui.checkBoxRoadTypeHighway->setChecked(true);
ui.checkBoxRoadTypeBoulevard->setChecked(true);
ui.checkBoxRoadTypeAvenue->setChecked(true);
ui.checkBoxRoadTypeLocalStreet->setChecked(true);
// register the event handlers
connect(ui.checkBoxRoadTypeHighway, SIGNAL(stateChanged(int)), this, SLOT(showRoad(int)));
connect(ui.checkBoxRoadTypeBoulevard, SIGNAL(stateChanged(int)), this, SLOT(showRoad(int)));
connect(ui.checkBoxRoadTypeAvenue, SIGNAL(stateChanged(int)), this, SLOT(showRoad(int)));
connect(ui.checkBoxRoadTypeLocalStreet, SIGNAL(stateChanged(int)), this, SLOT(showRoad(int)));
connect(ui.pushButtonExtractFeature, SIGNAL(clicked()), this, SLOT(extractFeature()));
hide();
}
void QueryExecutor::execute(HitCollectorPtr& pCollector,
ScorerPtr& pScorer,
const QueryTracerPtr& pTracer)
{
//TODO: allocate from mempool
MatchedDocSet machedDocs(m_pPool, DEFAULT_MATCHED_DOC_BUFFER_SIZE, size());
machedDocs.setTracer(pTracer);
QueryFeature queryFeature;
queryFeature.reserve(size());
extractFeature(queryFeature);
pScorer->beginQuery(m_pFeatureProvider, queryFeature);
while (advance(machedDocs) > 0)
{
FX_TRACE("Matched doc count: [%u]", (uint32_t)machedDocs.size());
pScorer->score(machedDocs);
pCollector->collect(machedDocs);
machedDocs.reset();
}
pScorer->endQuery();
}
int HaarFeature::extractOneTypeFeatures(int is_extract_feature, HaarFeatureInfoT &info, float *pt_feature)
{
int type = info.type;
int x_r = feature_sizes[type].x;
int y_r = feature_sizes[type].y;
double inv_r = feature_inv_ratio[type];
info.pos1.y = FEATURE_MARGIN;
while (info.pos1.y < template_h - 1)
{
info.pos1.x = FEATURE_MARGIN;
while (info.pos1.x < template_w - 1)
{
info.size.y = 2;
while (info.pos1.y + y_r * info.size.y - 1 < template_h - FEATURE_MARGIN)
{
info.size.x = 2;
while (info.pos1.x + x_r * info.size.x - 1 < template_w - FEATURE_MARGIN)
{
if (is_extract_feature == 0)
{
info.inv_area = INV_AREA_R / (info.size.x * info.size.y * inv_r);
pt_feature_infos[feature_count] = info;
}
else if (is_extract_feature > 0)
{
*(pt_feature + feature_count) = extractFeature(pt_feature_infos[feature_count]);
}
feature_count++;
info.size.x += HAAR_SCALE_STEP_X;
}
info.size.y += HAAR_SCALE_STEP_Y;
}
info.pos1.x += HAAR_SHIFT_STEP_X;
}
info.pos1.y += HAAR_SHIFT_STEP_Y;
}
return feature_count;
}
