SIFT.cpp

#include "SIFT.h"
#include "opencv/SIFT.h"
#include "LinearMapper.h"
#include "macroses.h"

namespace DirectGraphicalModels { namespace fex
{
    Mat CSIFT::get(const Mat &img)
    {
        int i;                      // feature index
        int x, y;
        int width   = img.cols;
        int height  = img.rows;
        
        // Converting to one channel image
        Mat I;
        if (img.channels() != 1) cvtColor(img, I, cv::ColorConversionCodes::COLOR_RGB2GRAY);
        else img.copyTo(I);

        Ptr<xfeatures2d::SIFT> sift = xfeatures2d::SIFT::create();
        
        // Prepare key points
        std::vector<KeyPoint> vKeyPoints;
//      sift->detect(img, vKeyPoints);
        for (y = 0; y < height; y++)
            for (x = 0; x < width; x++)
                vKeyPoints.push_back(KeyPoint(static_cast<float>(x), static_cast<float>(y), 1.0f));

//      drawKeypoints(img, vKeyPoints, img, Scalar::all(-1), DrawMatchesFlags::DRAW_RICH_KEYPOINTS);

        Mat descriptors;
        sift->detectAndCompute(I, Mat(), vKeyPoints, descriptors, true);

        const int nFeatures = descriptors.cols;         // 128
        DGM_ASSERT(nFeatures == 128);
        DGM_ASSERT(nFeatures < CV_CN_MAX);

        // Initializing features
        vec_mat_t vFeatures(nFeatures);
        for (i = 0; i < nFeatures; i++) vFeatures[i].create(img.size(), CV_8UC1);

        std::vector<byte *> pFeatures(nFeatures);
        
        for (y = 0; y < height; y++) {
            for (i = 0; i < nFeatures; i++) pFeatures[i] = vFeatures[i].ptr<byte>(y);
            for (x = 0; x < width; x++) {
                float *pDescriptors = descriptors.ptr<float>(y * width + x);
                for (i = 0; i < nFeatures; i++) 
                    pFeatures[i][x] = linear_mapper<byte>(pDescriptors[i], 0.0f, 255.0f);           // features[i] (x, y) = descriptors (i, pixel_idx)
            } // x
        } // y

        Mat res;
        merge(vFeatures, res);

        return res;
    }
} }