void ImageRecognizer::queryBoWFeature(std::vector<std::string>* pReturnArray,
const std::string& imagePath) {
pReturnArray->clear();
boost::shared_ptr<DBAdapter> dbAdapter(new HbaseAdapter);
cv::Mat image = cv::imread(imagePath, 1);
LocalFeature localFeature;
LocalFeatureExtractor localFeatureExtractor;
localFeatureExtractor.extractFeature(image, localFeature.keypoint,
localFeature.descriptor);
BoWHistogram histogram;
ANNVocabulary::instance()->quantizeFeature(localFeature, &histogram);
const std::vector<int64_t>& visualwordIdArray =
histogram.visualwordIdArray();
const std::vector<double>& frequencyArray = histogram.frequencyArray();
std::cout << "histogram size: " << histogram.size() << std::endl;
// TODO: modify the query pipeline
boost::unordered_map<int64_t, double> candidateScoreMap;
Ticker ticker;
ticker.start();
for (size_t i = 0; i < visualwordIdArray.size(); ++i) {
std::vector<Posting> postingArray;
InvertedIndexClient invertedIndexClient;
invertedIndexClient.loadPostingList(&postingArray,
visualwordIdArray[i]);
for (size_t j = 0; j < postingArray.size(); ++j) {
candidateScoreMap[postingArray[j].imageId] += postingArray[j].score
* frequencyArray[i];
}
}
ticker.stop();
ticker.start();
std::vector<RankItem<int64_t, double> > candidateRankList;
candidateRankList.reserve(candidateScoreMap.size());
for (boost::unordered_map<int64_t, double>::iterator iter =
candidateScoreMap.begin(); iter != candidateScoreMap.end();
++iter) {
RankItem<int64_t, double> item(iter->first, iter->second);
candidateRankList.push_back(item);
}
size_t candidateSize = (size_t) GlobalConfig::CANDIDATE_COUNT;
if (candidateRankList.size() > candidateSize) {
std::nth_element(candidateRankList.begin(),
candidateRankList.begin() + candidateSize,
candidateRankList.end());
}
ticker.stop();
ticker.start();
std::vector<double> queryVector;
histogram.flatten(&queryVector, ANNVocabulary::instance()->size());
std::vector<RankItem<int64_t, double> > rankList;
for (size_t i = 0; i < candidateRankList.size() && i < candidateSize; ++i) {
int64_t otherImageId = candidateRankList[i].index;
BoWHistogram otherHistogram;
otherHistogram.load(otherImageId);
double score = otherHistogram.innerProduct(queryVector);
RankItem<int64_t, double> item(otherImageId, -1.0 * score);
rankList.push_back(item);
}
std::sort(rankList.begin(), rankList.end());
ticker.stop();
if (GlobalConfig::USE_VERIFICATION) {
for (int i = 0;
i < GlobalConfig::VERIFICATION_COUNT
&& i < (int) rankList.size(); ++i) {
RobustMatcher rmatcher;
LocalFeature otherLocalFeature;
otherLocalFeature.load(rankList[i].index);
std::vector<cv::DMatch> matches;
cv::Mat homography = rmatcher.match(localFeature, otherLocalFeature,
&matches);
if (matches.size() >= 7) {
std::string path;
dbAdapter->loadCell(&path, GlobalConfig::IMAGE_TABLE,
rankList[i].index, GlobalConfig::IMAGE_HASH_COLUMN);
pReturnArray->push_back(path);
}
}
} else {
for (size_t i = 0; i < rankList.size(); ++i) {
std::string path;
dbAdapter->loadCell(&path, GlobalConfig::IMAGE_TABLE,
rankList[i].index, GlobalConfig::IMAGE_HASH_COLUMN);
pReturnArray->push_back(path);
}
}
}
void qTickerStop()
{
g_Ticker.stop();
}