static bool transform(cv::InputArray _src, cv::OutputArray _dst, cv::InputArray _lut,
        const int tilesX, const int tilesY, const cv::Size & tileSize)
    {

        cv::ocl::Kernel k("transform", cv::ocl::imgproc::clahe_oclsrc);
        if(k.empty())
            return false;

        int tile_size[2];
        tile_size[0] = tileSize.width;
        tile_size[1] = tileSize.height;

        cv::UMat src = _src.getUMat();
        _dst.create(src.size(), src.type());
        cv::UMat dst = _dst.getUMat();
        cv::UMat lut = _lut.getUMat();

        size_t localThreads[3]  = { 32, 8, 1 };
        size_t globalThreads[3] = { src.cols, src.rows, 1 };

        int idx = 0;
        idx = k.set(idx, cv::ocl::KernelArg::ReadOnlyNoSize(src));
        idx = k.set(idx, cv::ocl::KernelArg::WriteOnlyNoSize(dst));
        idx = k.set(idx, cv::ocl::KernelArg::ReadOnlyNoSize(lut));
        idx = k.set(idx, src.cols);
        idx = k.set(idx, src.rows);
        idx = k.set(idx, tile_size);
        idx = k.set(idx, tilesX);
        k.set(idx, tilesY);

        return k.run(2, globalThreads, localThreads, false);
    }

inline void bs::pattern::shiftPattern(cv::InputArray src, cv::OutputArray dst, const int direction, const int shift)
{
	cv::Mat d;
	cv::copyMakeBorder(src, d, shift, shift, shift, shift, cv::BORDER_WRAP);

	const auto direc = static_cast<Direction>(direction);

	switch (direc)
	{
	default:
		break;
	case Top:
		d = d(cv::Rect(cv::Point(shift, 2 * shift), src.size()));
		break;

	case Bottom:
		d = d(cv::Rect(cv::Point(shift, 0), src.size()));
		break;

	case Left:
		d = d(cv::Rect(cv::Point(2 * shift, shift), src.size()));
		break;

	case Right:
		d = d(cv::Rect(cv::Point(0, shift), src.size()));
		break;
	}
	d.copyTo(dst);
	return;
}

FilterCall::FilterCall(cv::InputArray in, cv::InputArray out,
                       impl::CallMetaData data, QString type,
                       QString description, QString requestedView)
    : Call{ data,                   std::move(type),
	    std::move(description), std::move(requestedView) },
      input_{ in.getMat().clone() }, output_{ out.getMat().clone() }
{
}

void sadTemplate(cv::InputArray tar, cv::InputArray tmp, cv::OutputArray res, int *minx, int *miny){
	//引数の入力をMatとして受け取る
	cv::Mat tarM = tar.getMat();	
	cv::Mat tmpM = tmp.getMat();
	cv::Mat resM = res.getMat();

	//sadが最小値のところがマッチングしたい箇所なので
	int minsad = std::numeric_limits<int>::max();	
	int sad = 0;  //各回のsadを格納
	int diff;	//sadに加算する前の作業変数
	int tarx,tary;	//目的のxy座標


	for(int y=0;y<tarM.rows - tmpM.rows;y++){
		for(int x=0;x<tarM.cols - tmpM.cols;x++){
			sad = 0;	//次の領域の計算の前に初期化
			//探索
			for(int yt = 0; yt < tmpM.rows; yt++){
				for(int xt = 0; xt < tmpM.cols; xt++){
					diff = (int)(tarM.at<uchar>(y+yt,x+xt) - tmpM.at<uchar>(yt,xt));
					if(diff < 0){  //負なら正に変換
						diff = -diff;
					}
					sad += diff;
					////残差逐次検定法
					if(sad > minsad){
						yt = tmpM.rows;	
						break;
					}
				}
			}

			//探索結果：sadが今までで最小なら
			if(sad < minsad){
				minsad = sad;	//最小値を更新
				//目的のxyを格納
				tarx = x;
				tary = y;
			}
		}
	}

	//outputに出力
	for(int y=0;y<resM.rows;y++){
		for(int x=0;x<resM.cols;x++){
			if(x==tarx && y==tary){
				resM.at<uchar>(y,x) = (uchar)0;
			}else{
				resM.at<uchar>(y,x) = (uchar)255;
			}
		}
	}
	std::cout << "最小値＝" << minsad << std::endl;
	std::cout << "最小点＝[" << tarx << ", " << tary << "]" <<  std::endl;
	*minx = tarx;
	*miny = tary;
}

void warmify(cv::InputArray src, cv::OutputArray dst, uchar delta)
{
    CV_Assert(src.type() == CV_8UC3);
    Mat imgSrc = src.getMat();
    CV_Assert(imgSrc.data);
    dst.create(src.size(), CV_8UC3);
    Mat imgDst = dst.getMat();

    imgDst = imgSrc + Scalar(0, delta, delta);
}

void stereo::stereoMatching(cv::InputArray _recImage1, cv::InputArray _recIamge2, cv::OutputArray _disparityMap, int minDisparity, int numDisparities, int SADWindowSize, int P1, int P2)
{
    Mat img1 = _recImage1.getMat();
    Mat img2 = _recIamge2.getMat();
    _disparityMap.create(img1.size(), CV_16S);
    Mat dis = _disparityMap.getMat();
    StereoSGBM matcher(minDisparity, numDisparities, SADWindowSize, P1, P2);
    matcher(img1, img2, dis);
    dis = dis / 16.0;
}

void testFunction(cv::InputArray ip, cv::InputArray op) {
    cv::Mat img = ip.getMat();
    cv::Mat obj = op.getMat();

    printMatrix(img);
    printMatrix(obj);

//	std::cerr<<img.checkVector()<<std::endl;

}

void stereo::stereoRectify(cv::InputArray _K1, cv::InputArray _K2, cv::InputArray _R, cv::InputArray _T,
                           cv::OutputArray _R1, cv::OutputArray _R2, cv::OutputArray _P1, cv::OutputArray _P2)
{
    Mat K1 = _K1.getMat(), K2 = _K2.getMat(), R = _R.getMat(), T = _T.getMat();
    _R1.create(3, 3, CV_32F);
    _R2.create(3, 3, CV_32F);
    Mat R1 = _R1.getMat();
    Mat R2 = _R2.getMat();
    _P1.create(3, 4, CV_32F);
    _P2.create(3, 4, CV_32F);
    Mat P1 = _P1.getMat();
    Mat P2 = _P2.getMat();

    if(K1.type()!=CV_32F)
        K1.convertTo(K1,CV_32F);
    if(K2.type()!=CV_32F)
        K2.convertTo(K2,CV_32F);
    if(R.type()!=CV_32F)
        R.convertTo(R,CV_32F);
    if(T.type()!=CV_32F)
        T.convertTo(T,CV_32F);
    if(T.rows != 3)
        T = T.t();

    // R and T is the transformation from the first to the second camera
    // Get the transformation from the second to the first camera
    Mat R_inv = R.t();
    Mat T_inv = -R.t()*T;

    Mat e1, e2, e3;
    e1 = T_inv.t() / norm(T_inv);
    /*Mat z = (Mat_<float>(1, 3) << 0.0,0.0,-1.0);
    e2 = e1.cross(z);
    e2 = e2 / norm(e2);*/
    e2 = (Mat_<float>(1,3) << T_inv.at<float>(1)*-1, T_inv.at<float>(0), 0.0 );
    e2 = e2 / (sqrt(e2.at<float>(0)*e2.at<float>(0) + e2.at<float>(1)*e2.at<float>(1)));
    e3 = e1.cross(e2);
    e3 = e3 / norm(e3);
    e1.copyTo(R1.row(0));
    e2.copyTo(R1.row(1));
    e3.copyTo(R1.row(2));
    R2 = R_inv * R1;

    P1.setTo(Scalar(0));
    R1.copyTo(P1.colRange(0, 3));
    P1 = K1 * P1;

    P2.setTo(Scalar(0));
    R2.copyTo(P2.colRange(0, 3));
    P2 = K2 * P2;

}

	void SLICSegment2Vector3D_(cv::InputArray segment, cv::InputArray signal, double invalidValue, std::vector<std::vector<cv::Point3_<S>>>& segmentPoint)
	{
		double minv, maxv;
		minMaxLoc(segment, &minv, &maxv);
		segmentPoint.clear();
		segmentPoint.resize((int)maxv + 1);
		if (signal.depth() == CV_8U) _SLICSegment2Vector3D_<uchar, S>(segment, signal, (uchar)invalidValue, segmentPoint);
		else if (signal.depth() == CV_16S) _SLICSegment2Vector3D_<short, S>(segment, signal, (short)invalidValue, segmentPoint);
		else if (signal.depth() == CV_16U) _SLICSegment2Vector3D_<ushort, S>(segment, signal, (ushort)invalidValue, segmentPoint);
		else if (signal.depth() == CV_32S) _SLICSegment2Vector3D_<int, S>(segment, signal, (int)invalidValue, segmentPoint);
		else if (signal.depth() == CV_32F) _SLICSegment2Vector3D_<float, S>(segment, signal, (float)invalidValue, segmentPoint);
		else if (signal.depth() == CV_64F) _SLICSegment2Vector3D_<double, S>(segment, signal, (double)invalidValue, segmentPoint);
	}

cv::Mat flutter::estimate_rigid_transform(cv::InputArray src1, cv::InputArray src2,
	double ransac_good_ratio, double ransac_threshold)
{
	Mat M(2, 3, CV_64F), A = src1.getMat(), B = src2.getMat();
	CvMat matA = A, matB = B, matM = M;
	int err = estimate_rigid_transform_detail(&matA, &matB, &matM,
		ransac_good_ratio, ransac_threshold);
	if (err == 1) {
		return M;
	} else {
		return Mat();
	}
}

void StereoMatch::StereoMatching(cv::InputArray rec_image1, cv::InputArray rec_image2,
    cv::OutputArray disparity_map, int min_disparity, int num_disparities, int SAD_window_size,
    int P1, int P2)
{
    cv::Mat img1 = rec_image1.getMat();
    cv::Mat img2 = rec_image2.getMat();
    disparity_map.create(img1.size(), CV_16S);
    cv::Mat dis = disparity_map.getMat();
    
    cv::StereoSGBM matcher(min_disparity, num_disparities, SAD_window_size, P1, P2);
    matcher(img1, img2, dis);
    dis = dis / 16.0;
}

void CV_HomographyTest::check_transform_quality(cv::InputArray src_points, cv::InputArray dst_points, const cv::Mat& H, const int norm_type)
{ 
	Mat src, dst_original; 
	cv::transpose(src_points.getMat(), src); cv::transpose(dst_points.getMat(), dst_original);
	cv::Mat src_3d(src.rows+1, src.cols, CV_32FC1);
	src_3d(Rect(0, 0, src.rows, src.cols)) = src;
	src_3d(Rect(src.rows, 0, 1, src.cols)) = Mat(1, src.cols, CV_32FC1, Scalar(1.0f));
	
	cv::Mat dst_found, dst_found_3d;
	cv::multiply(H, src_3d, dst_found_3d); 
	dst_found = dst_found_3d/dst_found_3d.row(dst_found_3d.rows-1);
    double reprojection_error = cv::norm(dst_original, dst_found, norm_type);
	CV_Assert ( reprojection_error > max_diff );
}

void Regression::write(cv::InputArray array)
{
    write() << "kind" << array.kind();
    write() << "type" << array.type();
    if (isVector(array))
    {
        int total = (int)array.total();
        int idx = regRNG.uniform(0, total);
        write() << "len" << total;
        write() << "idx" << idx;

        cv::Mat m = array.getMat(idx);

        if (m.total() * m.channels() < 26) //5x5 or smaller
            write() << "val" << m;
        else
            write(m);
    }
    else
    {
        if (array.total() * array.channels() < 26) //5x5 or smaller
            write() << "val" << array.getMat();
        else
            write(array.getMat());
    }
}

    FeatureValue FeatureShiCorner::calculate( cv::InputArray image )
    {
        Mat image_gray;

        cvtColor( image, image_gray, CV_BGR2GRAY );

        Mat corner( image.rows(), image.cols(), CV_32SC1, Scalar( 0 ) );
        computeRawCornerMat( image_gray, corner );
        auto points = genPoints( corner );


        // return genDescriptor;
        return FeatureValue();
    }

 void Tools::ReduceRowByMost(cv::InputArray _src, cv::OutputArray _dst, cv::InputArray _mask) {
   cv::Mat src = _src.getMat();
   _dst.create(1, src.cols, CV_8UC1);
   cv::Mat dst = _dst.getMat();
   cv::Mat mask = _mask.getMat();
   if (src.depth() != CV_8U || mask.depth() != CV_8U) {
     throw "TYPE DON'T SUPPORT";
   }
   int i, j, addr;
   cv::Size size = src.size();
   uchar *srcd = src.data;
   uchar *dstd = dst.data;
   uchar *maskd = mask.data;
   std::map<uchar, int> m;
   std::map<uchar, int>::iterator p;
   uchar mostVal;
   uchar rVal;
   int r;
   for (i = 0;i < size.width;++i) {
     m.clear();
     mostVal = 0;
     for (j = 0;j < size.height;++j) {
       addr = j*size.width + i;
       if (!maskd[addr])
         continue;
       for (r = -2;r <= 2;++r) {
         rVal = (uchar)(((int)srcd[addr] + r) % 180);
         if (m.find(rVal) != m.end()) {
           m[rVal]++;
         }
         else {
           m[rVal] = 1;
         }
       }
     }
     if (m.size() == 0) {
       dstd[i] = 0;
       continue;
     }
     mostVal = m.begin()->first;
     for (p = m.begin();p != m.end();++p) {
       if (p->second > m[mostVal]) {
         mostVal = p->first;
       }
     }
     dstd[i] = mostVal;
   }
 }

// Computes an Lidfaces model with images in src and corresponding labels
// in labels.
void Lidfaces::train(cv::InputArrayOfArrays src, cv::InputArray labels)
{
    std::vector<std::vector<cv::KeyPoint> > allKeyPoints;
    cv::Mat descriptors;

    // Get SIFT keypoints and LID descriptors
    detectKeypointsAndDescriptors(src, allKeyPoints, descriptors);

    // kmeans function requires points to be CV_32F
    descriptors.convertTo(descriptors, CV_32FC1);

    // Do k-means clustering
    const int CLUSTER_COUNT = params::lidFace::clustersAsPercentageOfKeypoints*descriptors.rows;
    cv::Mat histogramLabels;

    // This function populates histogram bin labels
    // The nth element of histogramLabels is an integer which represents the cluster that the
    // nth element of allKeyPoints is a member of.
    kmeans(
        descriptors, // The points we are clustering are the descriptors
        CLUSTER_COUNT, // The number of clusters (K)
        histogramLabels, // The label of the corresponding keypoint
        params::kmeans::termCriteria,
        params::kmeans::attempts,
        params::kmeans::flags,
        mCenters);

    // Convert to single channel 32 bit float as the matrix needs to be in a form supported
    // by calcHist
    histogramLabels.convertTo(histogramLabels, CV_32FC1);

    // We end up with a histogram for each image
    const size_t NUM_IMAGES = getSize(src);
    std::vector<cv::Mat> hists(NUM_IMAGES);
    // mCodebook.resize(NUM_IMAGES);

    // The histogramLabels vector contains ALL the points from EVERY image. We need to split
    // it up into groups of points for each image.
    // Because there are the same number of points in each image, and the points were put
    // into the labels vector in order, we can simply divide the labels vector evenly to get
    // the individual image's points.
    std::vector<cv::Mat> separatedLabels;
    for (unsigned int i = 0, startRow = 0; i < NUM_IMAGES; ++i)
    {
        separatedLabels.push_back(
            histogramLabels.rowRange(
                startRow,
                startRow + allKeyPoints[i].size()));
        startRow += allKeyPoints[i].size();
    }

    // Populate the hists vector
    generateHistograms(hists, separatedLabels, CLUSTER_COUNT);

    // Make the magnitude of each histogram equal to 1
    normalizeHistograms(hists);

    mCodebook = hists;
    mLabels = labels.getMat();
}

void RadiometricResponse::directMap(cv::InputArray _E, cv::OutputArray _I) const {
  if (_E.empty()) {
    _I.clear();
    return;
  }
  auto E = _E.getMat();
  _I.create(_E.size(), CV_8UC3);
  auto I = _I.getMat();
#if CV_MAJOR_VERSION > 2
  E.forEach<cv::Vec3f>(
      [&I, this](cv::Vec3f& v, const int* p) { I.at<cv::Vec3b>(p[0], p[1]) = inverseLUT(response_channels_, v); });
#else
  for (int i = 0; i < E.rows; i++)
    for (int j = 0; j < E.cols; j++) I.at<cv::Vec3b>(i, j) = inverseLUT(response_channels_, E.at<cv::Vec3f>(i, j));
#endif
}

    static bool calcLut(cv::InputArray _src, cv::OutputArray _dst,
        const int tilesX, const int tilesY, const cv::Size tileSize,
        const int clipLimit, const float lutScale)
    {
        cv::ocl::Kernel k("calcLut", cv::ocl::imgproc::clahe_oclsrc);
        if(k.empty())
            return false;

        cv::UMat src = _src.getUMat();
        _dst.create(tilesX * tilesY, 256, CV_8UC1);
        cv::UMat dst = _dst.getUMat();

        int tile_size[2];
        tile_size[0] = tileSize.width;
        tile_size[1] = tileSize.height;

        size_t localThreads[3]  = { 32, 8, 1 };
        size_t globalThreads[3] = { tilesX * localThreads[0], tilesY * localThreads[1], 1 };

        int idx = 0;
        idx = k.set(idx, cv::ocl::KernelArg::ReadOnlyNoSize(src));
        idx = k.set(idx, cv::ocl::KernelArg::WriteOnlyNoSize(dst));
        idx = k.set(idx, tile_size);
        idx = k.set(idx, tilesX);
        idx = k.set(idx, clipLimit);
        k.set(idx, lutScale);

        return k.run(2, globalThreads, localThreads, false);
    }

int boostColor(cv::InputArray src, cv::OutputArray dst, float intensity)
{
    const int MAX_INTENSITY = 255;

    Mat srcImg = src.getMat();

    CV_Assert(srcImg.channels() == 3);
    CV_Assert(intensity >= 0.0f && intensity <= 1.0f);

    if (srcImg.type() != CV_8UC3)
    {
        srcImg.convertTo(srcImg, CV_8UC3);
    }

    Mat srcHls;
    cvtColor(srcImg, srcHls, CV_BGR2HLS);
    
    int intensityInt = intensity * MAX_INTENSITY;
    srcHls += Scalar(0, 0, intensityInt);
    
    cvtColor(srcHls, dst, CV_HLS2BGR);
    
    dst.getMat().convertTo(dst, srcImg.type());
    return 0;
}

void StairDetection::CannyThreshold(cv::InputArray image, cv::Mat &edges) {
	if (image.channels() > 1) {
		/// Convert the image to grayscale
		cvtColor(image, edges, CV_RGB2GRAY);

		/// Reduce noise with a kernel 3x3
		blur(edges, edges, cv::Size(3, 3));
	}
	else {
		/// src is already a grayscale image.
		edges = image.getMat().clone();
	}

	/// Canny detector
	Canny(edges, edges, cannyLowThreshold, cannyLowThreshold*cannyRatio, cannyKernelSize);
}