void BOW::calc(cv::Mat & img, Feature & feat){
Mat rimg;
img.copyTo(rimg);
// if (img.rows > 100 && img.cols > 100)
// rimg = Mat(img, Range(img.rows/5, img.rows*4/5), Range(img.cols/5, img.cols*4/5));
// else
// rimg = img;
if (rimg.cols > 300 || rimg.rows > 300)
{
double fr = min(300.0/rimg.rows, 300.0/rimg.cols);
resize(rimg, rimg, Size(), fr, fr, INTER_AREA);
}
ImgBGReplace(rimg, rimg);
//GaussianBlur(img, img, Size(5, 5), 1.5, 1.5, BORDER_REPLICATE);
vector<KeyPoint> kps;
m_fd->detect(rimg, kps);
Mat res;
m_ext->compute(rimg, kps, res);
if (kps.size()) {
feat.clear();
for (int i = 0; i < m_ext->descriptorSize(); ++i)
feat.push_back((double) res.at<uchar>(0, i) / max(kps.size(), 1U));
}
else {
feat.clear();
for (int i = 0; i < m_ext->descriptorSize(); ++i)
feat.push_back(0);
}
}
void Labeler::extractFeature(Feature& feat, const Instance* pInstance, int idx) {
feat.clear();
const vector<string>& words = pInstance->words;
int sentsize = words.size();
string curWord = idx >= 0 && idx < sentsize ? normalize_to_lowerwithdigit(words[idx]) : nullkey;
// words
int unknownId = m_wordAlphabet.from_string(unknownkey);
int curWordId = m_wordAlphabet.from_string(curWord);
if (curWordId >= 0)
feat.words.push_back(curWordId);
else
feat.words.push_back(unknownId);
const vector<string>& linear_features = pInstance->sparsefeatures[idx];
for (int i = 0; i < linear_features.size(); i++) {
int curFeatId = m_featAlphabet.from_string(linear_features[i]);
if (curFeatId >= 0)
feat.linear_features.push_back(curFeatId);
}
}
void Labeler::extractFeature(Feature& feat, const Instance* pInstance, int idx) {
feat.clear();
const vector<string>& words = pInstance->words;
int sentsize = words.size();
string curWord = idx >= 0 && idx < sentsize ? normalize_to_lowerwithdigit(words[idx]) : nullkey;
// word features
int unknownId = m_wordAlphabet.from_string(unknownkey);
int curWordId = m_wordAlphabet.from_string(curWord);
if (curWordId >= 0)
feat.words.push_back(curWordId);
else
feat.words.push_back(unknownId);
// tag features
const vector<vector<string> > &tagfeatures = pInstance->tagfeatures;
int tagNum = tagfeatures[idx].size();
for (int i = 0; i < tagNum; i++) {
unknownId = m_tagAlphabets[i].from_string(unknownkey);
int curTagId = m_tagAlphabets[i].from_string(tagfeatures[idx][i]);
if (curTagId >= 0)
feat.tags.push_back(curTagId);
else
feat.tags.push_back(unknownId);
}
// char features
unknownId = m_charAlphabet.from_string(unknownkey);
const vector<vector<string> > &charfeatures = pInstance->charfeatures;
const vector<string>& cur_chars = charfeatures[idx];
int cur_char_size = cur_chars.size();
// actually we support a max window of m_options.charcontext = 2
for (int i = 0; i < cur_char_size; i++) {
string curChar = cur_chars[i];
int curCharId = m_charAlphabet.from_string(curChar);
if (curCharId >= 0)
feat.chars.push_back(curCharId);
else
feat.chars.push_back(unknownId);
}
int nullkeyId = m_charAlphabet.from_string(nullkey);
if (feat.chars.empty()) {
feat.chars.push_back(nullkeyId);
}
const vector<string>& linear_features = pInstance->sparsefeatures[idx];
for (int i = 0; i < linear_features.size(); i++) {
int curFeatId = m_featAlphabet.from_string(linear_features[i]);
if (curFeatId >= 0)
feat.linear_features.push_back(curFeatId);
}
}
void ColorMean::calc(Mat& im, Feature& res)
{
Mat ras;
resize(im, ras, Size(3, 3), 0, 0, INTER_AREA);
res.clear();
for (int i = 0; i < 3; ++i)
for (int j = 0; j < 3; ++j)
{
Vec3b r = ras.at<Vec3b>(i, j);
for (int k = 0; k < 3; ++k)
{
double t = r[k];
t /= 256;
res.push_back(t);
}
}
}
int FeatureSet::LoadFeature(const char* const fileName, Feature& feature)
{
feature.clear();
FileBuffer buffer(fileName);
int rtn = 0;
int len;
rtn = buffer.GetNextData(len);
CHECK_RTN(rtn);
for (int i = 0; i < len; ++i)
{
int key;
double value;
rtn = buffer.GetNextData(key);
CHECK_RTN(rtn);
rtn = buffer.GetNextData(value);
CHECK_RTN(rtn);
feature[key] = value;
}
return 0;
}
void Labeler::extractFeature(Feature& feat, const Instance* pInstance, int idx) {
feat.clear();
const vector<vector<string> >& words = pInstance->words;
const vector<vector<vector<string> > > &chars = pInstance->chars;
static vector<int> curChars;
int sentsize = words.size();
if (idx < 0 || idx >= sentsize)
return;
int wordnumber = words[idx].size();
int unknownWordId = m_wordAlphabet.from_string(unknownkey);
int unknownCharId = m_charAlphabet.from_string(unknownkey);
for (int i = 0; i < wordnumber; i++) {
string curWord = normalize_to_lowerwithdigit(words[idx][i]);
int curWordId = m_wordAlphabet.from_string(curWord);
if (curWordId >= 0)
feat.words.push_back(curWordId);
else
feat.words.push_back(unknownWordId);
int wordlength = chars[idx][i].size();
curChars.clear();
for (int j = 0; j < wordlength; j++) {
string curChar = chars[idx][i][j];
int curCharId = m_charAlphabet.from_string(curChar);
if (curCharId >= 0)
curChars.push_back(curCharId);
else
curChars.push_back(unknownCharId);
}
feat.chars.push_back(curChars);
}
}
