mcc/src/checker_detector.cpp

// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.

/*
 * MIT License
 *
 * Copyright (c) 2018 Pedro Diamel Marrero Fernández
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in the Software without restriction, including without limitation the rights
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in all
 * copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include "precomp.hpp"

#include "checker_detector.hpp"
#include "graph_cluster.hpp"
#include "bound_min.hpp"
#include "wiener_filter.hpp"
#include "checker_model.hpp"
#include "debug.hpp"

namespace cv
{
namespace mcc
{

std::mutex mtx; // mutex for critical section

Ptr<CCheckerDetector> CCheckerDetector::create()
{
    return makePtr<CCheckerDetectorImpl>();
}

CCheckerDetectorImpl::
    CCheckerDetectorImpl()

{
}

CCheckerDetectorImpl::~CCheckerDetectorImpl()
{
}

bool CCheckerDetectorImpl::
    setNet(cv::dnn::Net _net)
{
    net = _net;
    return !net.empty();
}

bool CCheckerDetectorImpl::
    _no_net_process(InputArray image, const TYPECHART chartType, const int nc,
                    const Ptr<DetectorParameters> &params,
                    std::vector<cv::Rect> regionsOfInterest)
{
    m_checkers.clear();
    this->net_used = false;

    cv::Mat img = image.getMat();
    for (const cv::Rect &region : regionsOfInterest)
    {
        //-------------------------------------------------------------------
        // Run the model to find good regions
        //-------------------------------------------------------------------
        cv::Mat croppedImage = img(region);
#ifdef MCC_DEBUG
        std::string pathOut = "./";
#endif
        //-------------------------------------------------------------------
        // prepare image
        //-------------------------------------------------------------------

        cv::Mat img_bgr, img_gray;
        float asp;
        prepareImage(croppedImage, img_gray, img_bgr, asp, params);

#ifdef MCC_DEBUG
        showAndSave("prepare_image", img_gray, pathOut);
#endif
        //-------------------------------------------------------------------
        // thresholding
        //-------------------------------------------------------------------
        std::vector<cv::Mat> img_bw;
        performThreshold(img_gray, img_bw, params);

        cv::Mat3f img_rgb_f(img_bgr);
        cv::cvtColor(img_rgb_f, img_rgb_f, COLOR_BGR2RGB);
        img_rgb_f /= 255;

        cv::Mat img_rgb_org, img_ycbcr_org;
        std::vector<cv::Mat> rgb_planes(3), ycbcr_planes(3);

        // Convert to RGB and YCbCr space
        cv::cvtColor(croppedImage, img_rgb_org, COLOR_BGR2RGB);
        cv::cvtColor(croppedImage, img_ycbcr_org, COLOR_BGR2YCrCb);

        // Get chanels
        split(img_rgb_org, rgb_planes);
        split(img_ycbcr_org, ycbcr_planes);

        parallel_for_(
            Range(0, (int)img_bw.size()), [&](const Range &range) {
                const int begin = range.start;
                const int end = range.end;
                for (int i = begin; i < end; i++)
                {

#ifdef MCC_DEBUG
                    showAndSave("threshold_image", img_bw[i], pathOut);
#endif
                    // find contour
                    //-------------------------------------------------------------------
                    ContoursVector contours;
                    findContours(img_bw[i], contours, params);

                    if (contours.empty())
                        continue;
#ifdef MCC_DEBUG
                    cv::Mat im_contour(img_bgr.size(), CV_8UC1);
                    im_contour = cv::Scalar(0);
                    cv::drawContours(im_contour, contours, -1, cv::Scalar(255), 2, LINE_AA);
                    showAndSave("find_contour", im_contour, pathOut);
#endif
                    //-------------------------------------------------------------------
                    // find candidate
                    //-------------------------------------------------------------------

                    std::vector<CChart> detectedCharts;
                    findCandidates(contours, detectedCharts, params);

                    if (detectedCharts.empty())
                        continue;

#ifdef MCC_DEBUG
                    cv::Mat img_chart;
                    img_bgr.copyTo(img_chart);

                    for (size_t ind = 0; ind < detectedCharts.size(); ind++)
                    {

                        CChartDraw chrtdrw((detectedCharts[ind]), img_chart);
                        chrtdrw.drawCenter();
                        chrtdrw.drawContour();
                    }
                    showAndSave("find_candidate", img_chart, pathOut);
#endif
                    //-------------------------------------------------------------------
                    // clusters analysis
                    //-------------------------------------------------------------------

                    std::vector<int> G;
                    clustersAnalysis(detectedCharts, G, params);

                    if (G.empty())
                        continue;

#ifdef MCC_DEBUG
                    cv::Mat im_gru;
                    img_bgr.copyTo(im_gru);
                    RNG rng(0xFFFFFFFF);
                    int radius = 10, thickness = -1;

                    std::vector<int> g;
                    unique(G, g);
                    size_t Nc = g.size();
                    std::vector<cv::Scalar> colors(Nc);
                    for (size_t ind = 0; ind < Nc; ind++)
                        colors[ind] = randomcolor(rng);

                    for (size_t ind = 0; ind < detectedCharts.size(); ind++)
                        cv::circle(im_gru, detectedCharts[ind].center, radius, colors[G[ind]],
                                   thickness);
                    showAndSave("clusters_analysis", im_gru, pathOut);
#endif
                    //-------------------------------------------------------------------
                    // checker color recognize
                    //-------------------------------------------------------------------

                    std::vector<std::vector<cv::Point2f>> colorCharts;
                    checkerRecognize(img_bgr, detectedCharts, G, chartType, colorCharts, params);

                    if (colorCharts.empty())
                        continue;

#ifdef MCC_DEBUG
                    cv::Mat image_box;
                    img_bgr.copyTo(image_box);
                    for (size_t ind = 0; ind < colorCharts.size(); ind++)
                    {
                        std::vector<cv::Point2f> ibox = colorCharts[ind];
                        cv::Scalar color_box = CV_RGB(0, 0, 255);
                        int thickness_box = 2;
                        cv::line(image_box, ibox[0], ibox[1], color_box, thickness_box, LINE_AA);
                        cv::line(image_box, ibox[1], ibox[2], color_box, thickness_box, LINE_AA);
                        cv::line(image_box, ibox[2], ibox[3], color_box, thickness_box, LINE_AA);
                        cv::line(image_box, ibox[3], ibox[0], color_box, thickness_box, LINE_AA);
                        //cv::circle(image_box, ibox[0], 10, cv::Scalar(0, 0, 255), 3);
                        //cv::circle(image_box, ibox[1], 10, cv::Scalar(0, 255, 0), 3);
                    }
                    showAndSave("checker_recognition", image_box, pathOut);
#endif
                    //-------------------------------------------------------------------
                    // checker color analysis
                    //-------------------------------------------------------------------
                    std::vector<Ptr<CChecker>> checkers;
                    checkerAnalysis(img_rgb_f, chartType, nc, colorCharts, checkers, asp, params,
                                    img_rgb_org, img_ycbcr_org, rgb_planes, ycbcr_planes);

#ifdef MCC_DEBUG
                    cv::Mat image_checker;
                    croppedImage.copyTo(image_checker);
                    for (size_t ck = 0; ck < checkers.size(); ck++)
                    {
                        Ptr<CCheckerDraw> cdraw = CCheckerDraw::create((checkers[ck]));
                        cdraw->draw(image_checker);
                    }
                    showAndSave("checker_analysis", image_checker, pathOut);
#endif
                    for (Ptr<CChecker> checker : checkers)
                    {
                        for (cv::Point2f &corner : checker->getBox())
                            corner += static_cast<cv::Point2f>(region.tl());

                        mtx.lock(); // push_back is not thread safe
                        m_checkers.push_back(checker);
                        mtx.unlock();
                    }
                }
#ifdef MCC_DEBUG
            },
            1); //Run only one thread in debug mode
#else
            });
#endif
    }
    //remove too close detections
    removeTooCloseDetections(params);
    m_checkers.resize(min(nc, (int)m_checkers.size()));
    return !m_checkers.empty();
}

bool CCheckerDetectorImpl::
    process(InputArray image, const TYPECHART chartType,const std::vector<cv::Rect> &regionsOfInterest,
            const int nc /*= 1*/, bool useNet /*=false*/, const Ptr<DetectorParameters> &params)
{
    m_checkers.clear();

    if (this->net.empty() || !useNet)
    {
        return _no_net_process(image, chartType, nc, params, regionsOfInterest);
    }
    this->net_used = true;

    cv::Mat img = image.getMat();

    cv::Mat img_rgb_org, img_ycbcr_org;
    std::vector<cv::Mat> rgb_planes(3), ycbcr_planes(3);

    // Convert to RGB and YCbCr space
    cv::cvtColor(img, img_rgb_org, COLOR_BGR2RGB);
    cv::cvtColor(img, img_ycbcr_org, COLOR_BGR2YCrCb);

    // Get chanels
    split(img_rgb_org, rgb_planes);
    split(img_ycbcr_org, ycbcr_planes);

    for (const cv::Rect &region : regionsOfInterest)
    {
        //-------------------------------------------------------------------
        // Run the model to find good regions
        //-------------------------------------------------------------------

        cv::Mat croppedImage = img(region);

        int rows = croppedImage.size[0];
        int cols = croppedImage.size[1];
        net.setInput(cv::dnn::blobFromImage(croppedImage, 1.0, cv::Size(), cv::Scalar(), true));
        cv::Mat output = net.forward();

        Mat detectionMat(output.size[2], output.size[3], CV_32F, output.ptr<float>());

        for (int i = 0; i < detectionMat.rows; i++)
        {
            float confidence = detectionMat.at<float>(i, 2);
            if (confidence > params->confidenceThreshold)
            {
                float xTopLeft = max(0.0f, detectionMat.at<float>(i, 3) * cols - params->borderWidth);
                float yTopLeft = max(0.0f, detectionMat.at<float>(i, 4) * rows - params->borderWidth);
                float xBottomRight = min((float)cols - 1, detectionMat.at<float>(i, 5) * cols + params->borderWidth);
                float yBottomRight = min((float)rows - 1, detectionMat.at<float>(i, 6) * rows + params->borderWidth);

                cv::Point2f topLeft = {xTopLeft, yTopLeft};
                cv::Point2f bottomRight = {xBottomRight, yBottomRight};

                cv::Rect innerRegion(topLeft, bottomRight);
                cv::Mat innerCroppedImage = croppedImage(innerRegion);

#ifdef MCC_DEBUG
                std::string pathOut = "./";
#endif
                //-------------------------------------------------------------------
                // prepare image
                //-------------------------------------------------------------------

                cv::Mat img_bgr, img_gray;
                float asp;
                prepareImage(innerCroppedImage, img_gray, img_bgr, asp, params);

                //-------------------------------------------------------------------
                // thresholding
                //-------------------------------------------------------------------

                std::vector<cv::Mat> img_bw;
                performThreshold(img_gray, img_bw, params);

                cv::Mat3f img_rgb_f(img_bgr);
                cv::cvtColor(img_rgb_f, img_rgb_f, COLOR_BGR2RGB);
                img_rgb_f /= 255;

                parallel_for_(
                    Range(0, (int)img_bw.size()), [&](const Range &range) {
                        const int begin = range.start;
                        const int end = range.end;

                        for (int ind = begin; ind < end; ind++)
                        {

#ifdef MCC_DEBUG
                            showAndSave("threshold_image", img_bw[ind], pathOut);
#endif
                            //------------------------------------------------------------------
                            // find contour
                            //-------------------------------------------------------------------
                            ContoursVector contours;
                            findContours(img_bw[ind], contours, params);

                            if (contours.empty())
                                continue;
#ifdef MCC_DEBUG
                            cv::Mat im_contour(img_bgr.size(), CV_8UC1);
                            im_contour = cv::Scalar(0);
                            cv::drawContours(im_contour, contours, -1, cv::Scalar(255), 2, LINE_AA);
                            showAndSave("find_contour", im_contour, pathOut);
#endif
                            //-------------------------------------------------------------------
                            // find candidate
                            //-------------------------------------------------------------------

                            std::vector<CChart> detectedCharts;
                            findCandidates(contours, detectedCharts, params);

                            if (detectedCharts.empty())
                                continue;

#ifdef MCC_DEBUG
                            cv::Mat img_chart;
                            img_bgr.copyTo(img_chart);

                            for (size_t index = 0; index < detectedCharts.size(); index++)
                            {

                                CChartDraw chrtdrw((detectedCharts[index]), img_chart);
                                chrtdrw.drawCenter();
                                chrtdrw.drawContour();
                            }
                            showAndSave("find_candidate", img_chart, pathOut);
#endif
                            //-------------------------------------------------------------------
                            // clusters analysis
                            //-------------------------------------------------------------------

                            std::vector<int> G;
                            clustersAnalysis(detectedCharts, G, params);

                            if (G.empty())
                                continue;
#ifdef MCC_DEBUG
                            cv::Mat im_gru;
                            img_bgr.copyTo(im_gru);
                            RNG rng(0xFFFFFFFF);
                            int radius = 10, thickness = -1;

                            std::vector<int> g;
                            unique(G, g);
                            size_t Nc = g.size();
                            std::vector<cv::Scalar> colors(Nc);
                            for (size_t index = 0; index < Nc; index++)
                                colors[index] = randomcolor(rng);

                            for (size_t index = 0; index < detectedCharts.size(); index++)
                                cv::circle(im_gru, detectedCharts[index].center, radius, colors[G[index]],
                                           thickness);
                            showAndSave("clusters_analysis", im_gru, pathOut);
#endif

                            //-------------------------------------------------------------------
                            // checker color recognize
                            //-------------------------------------------------------------------

                            std::vector<std::vector<cv::Point2f>> colorCharts;
                            checkerRecognize(img_bgr, detectedCharts, G, chartType, colorCharts, params);

                            if (colorCharts.empty())
                                continue;

#ifdef MCC_DEBUG
                            cv::Mat image_box;
                            img_bgr.copyTo(image_box);
                            for (size_t index = 0; index < colorCharts.size(); index++)
                            {
                                std::vector<cv::Point2f> ibox = colorCharts[index];
                                cv::Scalar color_box = CV_RGB(0, 0, 255);
                                int thickness_box = 2;
                                cv::line(image_box, ibox[0], ibox[1], color_box, thickness_box, LINE_AA);
                                cv::line(image_box, ibox[1], ibox[2], color_box, thickness_box, LINE_AA);
                                cv::line(image_box, ibox[2], ibox[3], color_box, thickness_box, LINE_AA);
                                cv::line(image_box, ibox[3], ibox[0], color_box, thickness_box, LINE_AA);
                                //cv::circle(image_box, ibox[0], 10, cv::Scalar(0, 0, 255), 3);
                                //cv::circle(image_box, ibox[1], 10, cv::Scalar(0, 255, 0), 3);
                            }
                            showAndSave("checker_recognition", image_box, pathOut);
#endif
                            //-------------------------------------------------------------------
                            // checker color analysis
                            //-------------------------------------------------------------------
                            std::vector<Ptr<CChecker>> checkers;
                            checkerAnalysis(img_rgb_f, chartType, nc, colorCharts, checkers, asp, params,
                                            img_rgb_org, img_ycbcr_org, rgb_planes, ycbcr_planes);
#ifdef MCC_DEBUG
                            cv::Mat image_checker;
                            innerCroppedImage.copyTo(image_checker);
                            for (size_t ck = 0; ck < checkers.size(); ck++)
                            {
                                Ptr<CCheckerDraw> cdraw = CCheckerDraw::create((checkers[ck]));
                                cdraw->draw(image_checker);
                            }
                            showAndSave("checker_analysis", image_checker, pathOut);
#endif
                            for (Ptr<CChecker> checker : checkers)
                            {
                                for (cv::Point2f &corner : checker->getBox())
                                    corner += static_cast<cv::Point2f>(region.tl() + innerRegion.tl());
                                mtx.lock(); // push_back is not thread safe
                                m_checkers.push_back(checker);
                                mtx.unlock();
                            }
                        }
#ifdef MCC_DEBUG
                    },
                    1); //Run only one thread in debug mode
#else
                    });
#endif
            }
        }
    }
    // As a failsafe try the classical method
    if (m_checkers.empty())
    {
        return _no_net_process(image, chartType, nc, params, regionsOfInterest);
    }
    //remove too close detections
    removeTooCloseDetections(params);

    m_checkers.resize(min(nc, (int)m_checkers.size()));

    return !m_checkers.empty();
}


//Overload for the above function
bool CCheckerDetectorImpl::
    process(InputArray image, const TYPECHART chartType,
            const int nc /*= 1*/, bool useNet /*=false*/, const Ptr<DetectorParameters> &params)
{
    return process(image, chartType, std::vector<cv::Rect>(1, Rect(0, 0, image.cols(), image.rows())),
                   nc,useNet, params);
}


void CCheckerDetectorImpl::
    prepareImage(InputArray bgr, OutputArray grayOut,
                 OutputArray bgrOut, float &aspOut,
                 const Ptr<DetectorParameters> &params) const
{

    int min_size;
    cv::Size size = bgr.size();
    aspOut = 1;
    bgr.copyTo(bgrOut);

    // Resize image
    min_size = std::min(size.width, size.height);
    if (params->minImageSize > min_size)
    {
        aspOut = (float)params->minImageSize / min_size;
        cv::resize(bgr, bgrOut, cv::Size(int(size.width * aspOut), int(size.height * aspOut)));
    }

    // Convert to grayscale
    cv::cvtColor(bgrOut, grayOut, COLOR_BGR2GRAY);

    // PDiamel: wiener adaptative methods to minimize the noise effets
    // by illumination

    CWienerFilter filter;
    filter.wiener2(grayOut, grayOut, 5, 5);

    //JLeandro: perform morphological open on the equalized image
    //to minimize the noise effects by CLAHE and to even intensities
    //inside the MCC patches (regions)

    cv::Mat strelbox = cv::getStructuringElement(cv::MORPH_RECT, Size(5, 5));
    cv::morphologyEx(grayOut, grayOut, MORPH_OPEN, strelbox);
}

void CCheckerDetectorImpl::
    performThreshold(InputArray grayscaleImg,
                     OutputArrayOfArrays thresholdImgs,
                     const Ptr<DetectorParameters> &params) const
{
    // number of window sizes (scales) to apply adaptive thresholding
    int nScales = (params->adaptiveThreshWinSizeMax - params->adaptiveThreshWinSizeMin) / params->adaptiveThreshWinSizeStep + 1;
    thresholdImgs.create(nScales, 1, CV_8U);
    std::vector<cv::Mat> _thresholdImgs;
    for (int i = 0; i < nScales; i++)
    {
        int currScale = params->adaptiveThreshWinSizeMin + i * params->adaptiveThreshWinSizeStep;

        cv::Mat tempThresholdImg;
        cv::adaptiveThreshold(grayscaleImg, tempThresholdImg, 255, cv::ADAPTIVE_THRESH_MEAN_C,
                              cv::THRESH_BINARY_INV, currScale, params->adaptiveThreshConstant);

        _thresholdImgs.push_back(tempThresholdImg);
    }

    thresholdImgs.assign(_thresholdImgs);
}

void CCheckerDetectorImpl::
    findContours(
        InputArray srcImg,
        ContoursVector &contours,
        const Ptr<DetectorParameters> &params) const
{
    // contour detected
    // [Suzuki85] Suzuki, S. and Abe, K., Topological Structural Analysis of Digitized
    // Binary Images by Border Following. CVGIP 30 1, pp 32-46 (1985)
    ContoursVector allContours;
    cv::findContours(srcImg, allContours, RETR_LIST, CHAIN_APPROX_NONE);

    //select contours
    contours.clear();

    const long long int srcImgArea = srcImg.rows() * srcImg.cols();
    for (size_t i = 0; i < allContours.size(); i++)
    {

        PointsVector contour;
        contour = allContours[i];

        int contourSize = (int)contour.size();
        if (contourSize <= params->minContourPointsAllowed)
            continue;

        double area = cv::contourArea(contour);
        // double perm = cv::arcLength(contour, true);

        if (this->net_used && area / srcImgArea < params->minContoursAreaRate)
            continue;

        if (!this->net_used && area < params->minContoursArea)
            continue;
        // Circularity factor condition
        // KORDECKI, A., & PALUS, H. (2014). Automatic detection of colour charts in images.
        // Przegl?d Elektrotechniczny, 90(9), 197-202.
        // 0.65 < \frac{4*pi*A}{P^2} < 0.97
        // double Cf = 4 * CV_PI * area / (perm * perm);
        // if (Cf < 0.5 || Cf > 0.97) continue;

        // Soliditys
        // This measure is proposed in this work.
        PointsVector hull;
        cv::convexHull(contour, hull);
        double area_hull = cv::contourArea(hull);
        double S = area / area_hull;
        if (S < params->minContourSolidity)
            continue;

        // Texture analysis
        // ...

        contours.push_back(allContours[i]);
    }
}

void CCheckerDetectorImpl::
    findCandidates(
        const ContoursVector &contours,
        std::vector<CChart> &detectedCharts,
        const Ptr<DetectorParameters> &params)
{
    std::vector<cv::Point> approxCurve;
    std::vector<CChart> possibleCharts;

    // For each contour, analyze if it is a parallelepiped likely to be the chart
    for (size_t i = 0; i < contours.size(); i++)
    {
        // Approximate to a polygon
        //  It uses the Douglas-Peucker algorithm
        // http://en.wikipedia.org/wiki/Ramer-Douglas-Peucker_algorithm
        double eps = contours[i].size() * params->findCandidatesApproxPolyDPEpsMultiplier;
        cv::approxPolyDP(contours[i], approxCurve, eps, true);

        // We interested only in polygons that contains only four points
        if (approxCurve.size() != 4)
            continue;

        // And they have to be convex
        if (!cv::isContourConvex(approxCurve))
            continue;

        // Ensure that the distance between consecutive points is large enough
        float minDist = INFINITY;

        for (size_t j = 0; j < 4; j++)
        {
            cv::Point side = approxCurve[j] - approxCurve[(j + 1) % 4];
            float squaredSideLength = (float)side.dot(side);
            minDist = std::min(minDist, squaredSideLength);
        }

        // Check that distance is not very small
        if (minDist < params->minContourLengthAllowed)
            continue;

        // All tests are passed. Save chart candidate:
        CChart chart;

        std::vector<cv::Point2f> corners(4);
        for (int j = 0; j < 4; j++)
            corners[j] = cv::Point2f((float)approxCurve[j].x, (float)approxCurve[j].y);
        chart.setCorners(corners);

        possibleCharts.push_back(chart);
    }

    // Remove these elements which corners are too close to each other.
    // Eliminate overlaps!!!
    // First detect candidates for removal:
    std::vector<std::pair<int, int>> tooNearCandidates;
    for (int i = 0; i < (int)possibleCharts.size(); i++)
    {
        const CChart &m1 = possibleCharts[i];

        //calculate the average distance of each corner to the nearest corner of the other chart candidate
        for (int j = i + 1; j < (int)possibleCharts.size(); j++)
        {
            const CChart &m2 = possibleCharts[j];

            float distSquared = 0;

            for (int c = 0; c < 4; c++)
            {
                cv::Point v = m1.corners[c] - m2.corners[c];
                distSquared += v.dot(v);
            }

            distSquared /= 4;

            if (distSquared < params->minInterContourDistance)
            {
                tooNearCandidates.push_back(std::pair<int, int>(i, j));
            }
        }
    }

    // Mark for removal the element of the pair with smaller perimeter
    std::vector<bool> removalMask(possibleCharts.size(), false);

    for (size_t i = 0; i < tooNearCandidates.size(); i++)
    {
        float p1 = perimeter(possibleCharts[tooNearCandidates[i].first].corners);
        float p2 = perimeter(possibleCharts[tooNearCandidates[i].second].corners);

        size_t removalIndex;
        if (p1 > p2)
            removalIndex = tooNearCandidates[i].second;
        else
            removalIndex = tooNearCandidates[i].first;

        removalMask[removalIndex] = true;
    }

    // Return candidates
    detectedCharts.clear();
    for (size_t i = 0; i < possibleCharts.size(); i++)
    {
        if (removalMask[i])
            continue;
        detectedCharts.push_back(possibleCharts[i]);
    }
}

void CCheckerDetectorImpl::
    clustersAnalysis(
        const std::vector<CChart> &detectedCharts,
        std::vector<int> &groups,
        const Ptr<DetectorParameters> &params)
{
    size_t N = detectedCharts.size();
    std::vector<cv::Point> X(N);
    std::vector<double> B0(N), W(N);
    std::vector<int> G;

    CChart chart;
    double b0;
    for (size_t i = 0; i < N; i++)
    {
        chart = detectedCharts[i];
        b0 = chart.large_side * params->B0factor;
        X[i] = chart.center;
        W[i] = chart.area;
        B0[i] = b0;
    }

    CB0cluster bocluster;
    bocluster.setVertex(X);
    bocluster.setWeight(W);
    bocluster.setB0(B0);
    bocluster.group();
    bocluster.getGroup(G);
    groups = G;
}

void CCheckerDetectorImpl::
    checkerRecognize(
        InputArray img,
        const std::vector<CChart> &detectedCharts,
        const std::vector<int> &G,
        const TYPECHART chartType,
        std::vector<std::vector<cv::Point2f>> &colorChartsOut,
        const Ptr<DetectorParameters> &params)
{
    std::vector<int> gU;
    unique(G, gU);
    size_t Nc = gU.size();                //numero de grupos
    size_t Ncc = detectedCharts.size(); //numero de charts

    std::vector<std::vector<cv::Point2f>> colorCharts;

    for (size_t g = 0; g < Nc; g++)
    {

        ///-------------------------------------------------
        /// selecionar grupo i-esimo

        std::vector<CChart> chartSub;
        for (size_t i = 0; i < Ncc; i++)
            if (G[i] == (int)g)
                chartSub.push_back(detectedCharts[i]);

        size_t Nsc = chartSub.size();
        if (Nsc < params->minGroupSize)
            continue;

        ///-------------------------------------------------
        /// min box estimation

        CBoundMin bm;
        std::vector<cv::Point2f> points;

        bm.setCharts(chartSub);
        bm.calculate();
        bm.getCorners(points);

        // boundary condition
        if (points.size() == 0)
            continue;

        // sort the points in anti-clockwise order
        polyanticlockwise(points);

        ///-------------------------------------------------
        /// box projective transformation

        // get physical char box model
        std::vector<cv::Point2f> chartPhy;
        cv::Size size_box_phy;
        get_subbox_chart_physical(points, chartPhy, size_box_phy);

        // Find the perspective transformation that brings current chart to rectangular form
        Matx33f ccT = cv::getPerspectiveTransform(points, chartPhy);

        // transformer
        std::vector<cv::Point2f> c(Nsc), ct;
        std::vector<cv::Point2f> ch(4 * Nsc), cht;

        for (size_t i = 0; i < Nsc; i++)
        {

            CChart cc = chartSub[i];
            for (size_t j = 0; j < 4; j++)
                ch[i * 4 + j] = cc.corners[j];
            c[i] = chartSub[i].center;
        }

        transform_points_forward(ccT, c, ct);
        transform_points_forward(ccT, ch, cht);

        float wchart = 0, hchart = 0;
        std::vector<float> cx(Nsc), cy(Nsc);
        for (size_t i = 0, k = 0; i < Nsc; i++)
        {
            k = i * 4;
            cv::Point2f v1 = cht[k + 1] - cht[k + 0];
            cv::Point2f v2 = cht[k + 3] - cht[k + 0];
            wchart += (float)norm(v1);
            hchart += (float)norm(v2);
            cx[i] = ct[i].x;
            cy[i] = ct[i].y;
        }

        wchart /= Nsc;
        hchart /= Nsc;

        ///-------------------------------------------------
        /// centers and color estimate

        float tolx = wchart / 2, toly = hchart / 2;
        std::vector<float> cxr, cyr;
        reduce_array(cx, cxr, tolx);
        reduce_array(cy, cyr, toly);

        if (cxr.size() == 1 || cyr.size() == 1) //no information can be extracted if \
                                                //only one row or columns in present
            continue;
        // color and center rectificate
        cv::Size2i colorSize = cv::Size2i((int)cxr.size(), (int)cyr.size());
        cv::Mat colorMat(colorSize, CV_32FC3);
        std::vector<cv::Point2f> cte(colorSize.area());

        int k = 0;

        for (int i = 0; i < colorSize.height; i++)
        {
            for (int j = 0; j < colorSize.width; j++)
            {
                cv::Point2f vc = cv::Point2f(cxr[j], cyr[i]);
                cte[k] = vc;

                // recovery color
                cv::Point2f cti;
                cv::Matx31f p, xt;

                p(0, 0) = vc.x;
                p(1, 0) = vc.y;
                p(2, 0) = 1;
                xt = ccT.inv() * p;
                cti.x = xt(0, 0) / xt(2, 0);
                cti.y = xt(1, 0) / xt(2, 0);

                // color
                int x, y;
                x = (int)cti.x;
                y = (int)cti.y;
                Vec3f &srgb = colorMat.at<Vec3f>(i, j);
                Vec3b rgb;
                if (0 <= y && y < img.rows() && 0 <= x && x < img.cols())
                    rgb = img.getMat().at<Vec3b>(y, x);

                srgb[0] = (float)rgb[0] / 255;
                srgb[1] = (float)rgb[1] / 255;
                srgb[2] = (float)rgb[2] / 255;

                k++;
            }
        }

        CChartModel::SUBCCMModel scm;
        scm.centers = cte;
        scm.color_size = colorSize;
        colorMat = colorMat.t();
        scm.sub_chart = colorMat.reshape(3, colorSize.area());

        ///-------------------------------------------------

        // color chart model
        CChartModel cccm(chartType);

        int iTheta;     // rotation angle of chart
        int offset;     // offset
        float error; // min error
        if (!cccm.evaluate(scm, offset, iTheta, error))
            continue;
        if (iTheta >= 4)
            cccm.flip();

        for (int i = 0; i < iTheta % 4; i++)
            cccm.rotate90();

        ///-------------------------------------------------
        /// calculate coordanate

        cv::Size2i dim = cccm.size;
        std::vector<cv::Point2f> center = cccm.center;
        std::vector<cv::Point2f> box = cccm.box;
        int cols = dim.height - colorSize.width + 1;

        int x = (offset) / cols;
        int y = (offset) % cols;

        // seleccionar sub grid centers of model
        std::vector<cv::Point2f> ctss(colorSize.area());
        cv::Point2f point_ac = cv::Point2f(0, 0);
        int p = 0;

        for (int i = x; i < (x + colorSize.height); i++)
        {
            for (int j = y; j < (y + colorSize.width); j++)
            {
                int iter = i * dim.height + j;
                ctss[p] = center[iter];
                point_ac += ctss[p];
                p++;
            }
        }
        // is colineal point
        if (point_ac.x == ctss[0].x * p || point_ac.y == ctss[0].y * p)
            continue;
        // Find the perspective transformation
        cv::Matx33f ccTe = cv::findHomography(ctss, cte);

        std::vector<cv::Point2f> tbox, ibox;
        transform_points_forward(ccTe, box, tbox);
        transform_points_inverse(ccT, tbox, ibox);

        // sort the points in anti-clockwise order
        if (iTheta < 4)
            mcc::polyanticlockwise(ibox);
        else
            mcc::polyclockwise(ibox);
        // circshift(ibox, 4 - iTheta);
        colorCharts.push_back(ibox);
    }

    // return
    colorChartsOut = colorCharts;
}

void CCheckerDetectorImpl::
    checkerAnalysis(
        InputArray img_f,
        const TYPECHART chartType,
        const unsigned int nc,
        const std::vector<std::vector<cv::Point2f>> &colorCharts,
        std::vector<Ptr<CChecker>> &checkers,
        float asp,
        const Ptr<DetectorParameters> &params,
        const cv::Mat &img_rgb_org,
        const cv::Mat &img_ycbcr_org,
        std::vector<cv::Mat> &rgb_planes,
        std::vector<cv::Mat> &ycbcr_planes)
{
    size_t N;
    std::vector<cv::Point2f> ibox;

    // color chart classic model
    CChartModel cccm(chartType);
    cv::Mat lab;
    cccm.copyToColorMat(lab, 0);
    lab = lab.reshape(3, lab.size().area());
    lab /= 255;

    cv::Mat mask(img_f.size(), CV_8U);
    mask.setTo(Scalar::all(0));

    N = colorCharts.size();
    std::vector<float> J(N);
    for (size_t i = 0; i < N; i++)
    {
        ibox = colorCharts[i];
        J[i] = cost_function(img_f, mask, lab, ibox, chartType);
    }

    std::vector<int> idx;
    sort(J, idx);
    float invAsp = 1 / asp;
    size_t n = cv::min(nc, (unsigned)N);
    checkers.clear();

    for (size_t i = 0; i < n; i++)
    {
        ibox = colorCharts[idx[i]];

        if (J[i] > params->maxError)
            continue;

        // redimention box
        for (size_t j = 0; j < 4; j++)
            ibox[j] = invAsp * ibox[j];

        cv::Mat charts_rgb, charts_ycbcr;
        get_profile(ibox, chartType, charts_rgb, charts_ycbcr, img_rgb_org,
                    img_ycbcr_org, rgb_planes, ycbcr_planes);

        // result
        Ptr<CChecker> checker = CChecker::create();
        checker->setBox(ibox);
        checker->setTarget(chartType);
        checker->setChartsRGB(charts_rgb);
        checker->setChartsYCbCr(charts_ycbcr);
        checker->setCenter(mace_center(ibox));
        checker->setCost(J[i]);

        checkers.push_back(checker);
    }
}

void CCheckerDetectorImpl::
    removeTooCloseDetections(const Ptr<DetectorParameters> &params)
{
    // Remove these elements which corners are too close to each other.
    // Eliminate overlaps!!!
    // First detect candidates for removal:
    std::vector<std::pair<int, int>> tooNearCandidates;
    for (int i = 0; i < (int)m_checkers.size(); i++)
    {
        const Ptr<CChecker> &m1 = m_checkers[i];

        //calculate the average distance of each corner to the nearest corner of the other chart candidate
        for (int j = i + 1; j < (int)m_checkers.size(); j++)
        {
            const Ptr<CChecker> &m2 = m_checkers[j];

            float distSquared = 0;

            for (int c = 0; c < 4; c++)
            {
                cv::Point v = m1->getBox()[c] - m2->getBox()[c];
                distSquared += v.dot(v);
            }

            distSquared /= 4;

            if (distSquared < params->minInterCheckerDistance)
            {
                tooNearCandidates.push_back(std::pair<int, int>(i, j));
            }
        }
    }

    // Mark for removal the element of the pair with smaller cost
    std::vector<bool> removalMask(m_checkers.size(), false);

    for (size_t i = 0; i < tooNearCandidates.size(); i++)
    {
        float p1 = m_checkers[tooNearCandidates[i].first]->getCost();
        float p2 = m_checkers[tooNearCandidates[i].second]->getCost();

        size_t removalIndex;
        if (p1 < p2)
            removalIndex = tooNearCandidates[i].second;
        else
            removalIndex = tooNearCandidates[i].first;

        removalMask[removalIndex] = true;
    }

    std::vector<Ptr<CChecker>> copy_m_checkers = m_checkers;
    m_checkers.clear();

    for (size_t i = 0; i < copy_m_checkers.size(); i++)
    {
        if (removalMask[i])
            continue;
        m_checkers.push_back(copy_m_checkers[i]);
    }

    sort( m_checkers.begin(), m_checkers.end(),
          [&](const Ptr<CChecker> &a, const Ptr<CChecker> &b)
          {
              return a->getCost() < b->getCost();
          });
}

void CCheckerDetectorImpl::
    get_subbox_chart_physical(const std::vector<cv::Point2f> &points, std::vector<cv::Point2f> &chartPhy, cv::Size &size)
{
    float w, h;
    cv::Point2f v1 = points[1] - points[0];
    cv::Point2f v2 = points[3] - points[0];
    float asp = (float)(norm(v2) / norm(v1));

    w = 100;
    h = (float)floor(100 * asp + 0.5);

    chartPhy.clear();
    chartPhy.resize(4);
    chartPhy[0] = cv::Point2f(0, 0);
    chartPhy[1] = cv::Point2f(w, 0);
    chartPhy[2] = cv::Point2f(w, h);
    chartPhy[3] = cv::Point2f(0, h);

    size = cv::Size((int)w, (int)h);
}

void CCheckerDetectorImpl::
    reduce_array(const std::vector<float> &x, std::vector<float> &x_new, float tol)
{
    size_t n = x.size(), nn;
    std::vector<float> xx = x;
    x_new.clear();

    // sort array
    std::sort(xx.begin(), xx.end());

    // label array
    std::vector<int> label(n);
    for (size_t i = 0; i < n; i++)
        label[i] = abs(xx[(n + i - 1) % n] - xx[i]) > tol;

    // diff array
    for (size_t i = 1; i < n; i++)
        label[i] += label[i - 1];

    // unique array
    std::vector<int> ulabel;
    unique(label, ulabel);

    // mean for group
    nn = ulabel.size();
    x_new.resize(nn);
    for (size_t i = 0; i < nn; i++)
    {
        float mu = 0, s = 0;
        for (size_t j = 0; j < n; j++)
        {
            mu += (label[j] == ulabel[i]) * xx[j];
            s += (label[j] == ulabel[i]);
        }
        x_new[i] = mu / s;
    }

    // diff array
    std::vector<float> dif(nn - 1);
    for (size_t i = 0; i < nn - 1; i++)
        dif[i] = (x_new[(i + 1) % nn] - x_new[i]);

    // max and idx
    float fmax = 0;
    size_t idx = 0;
    for (size_t i = 0; i < nn - 1; i++)
        if (fmax < dif[i])
        {
            fmax = dif[i];
            idx = i;
        }

    // add ... X[i] MAX X[i+] ...
    if (fmax > 4 * tol)
        x_new.insert(x_new.begin() + idx + 1, (x_new[idx] + x_new[idx + 1]) / 2);
}

void CCheckerDetectorImpl::
    transform_points_forward(InputArray T, const std::vector<cv::Point2f> &X, std::vector<cv::Point2f> &Xt)
{
    size_t N = X.size();
    if (N == 0)
        return;

    Xt.clear();
    Xt.resize(N);
    cv::Matx31f p, xt;
    cv::Point2f pt;

    cv::Matx33f _T = T.getMat();
    for (int i = 0; i < (int)N; i++)
    {
        p(0, 0) = X[i].x;
        p(1, 0) = X[i].y;
        p(2, 0) = 1;
        xt = _T * p;
        pt.x = xt(0, 0) / xt(2, 0);
        pt.y = xt(1, 0) / xt(2, 0);
        Xt[i] = pt;
    }
}

void CCheckerDetectorImpl::
    transform_points_inverse(InputArray T, const std::vector<cv::Point2f> &X, std::vector<cv::Point2f> &Xt)
{
    cv::Matx33f _T = T.getMat();
    cv::Matx33f Tinv = _T.inv();
    transform_points_forward(Tinv, X, Xt);
}
void CCheckerDetectorImpl::
    get_profile(
        const std::vector<cv::Point2f> &ibox,
        const TYPECHART chartType,
        OutputArray charts_rgb,
        OutputArray charts_ycbcr,
        InputArray im_rgb,
        InputArray im_ycbcr,
        std::vector<cv::Mat> &rgb_planes,
        std::vector<cv::Mat> &ycbcr_planes)
{
    // color chart classic model
    CChartModel cccm(chartType);
    cv::Mat lab;
    size_t N;
    std::vector<cv::Point2f> fbox = cccm.box;
    std::vector<cv::Point2f> cellchart = cccm.cellchart;

    // tranformation
    Matx33f ccT = cv::getPerspectiveTransform(fbox, ibox);

    cv::Mat mask(im_rgb.size(), CV_8U);
    mask.setTo(Scalar::all(0));
    std::vector<cv::Point2f> bch(4), bcht(4);
    N = cellchart.size() / 4;

    // Create table charts information
    //          |p_size|average|stddev|max|min|
    //    RGB   |      |       |      |   |   |
    //  YCbCr |

    Mat _charts_rgb = cv::Mat(cv::Size(5, 3 * (int)N), CV_64F);
    Mat _charts_ycbcr = cv::Mat(cv::Size(5, 3 * (int)N), CV_64F);

    cv::Scalar mu_rgb, st_rgb, mu_ycb, st_ycb, p_size;
    double max_rgb[3], min_rgb[3], max_ycb[3], min_ycb[3];

    for (int i = 0, k; i < (int)N; i++)
    {
        k = 4 * i;
        bch[0] = cellchart[k + 0];
        bch[1] = cellchart[k + 1];
        bch[2] = cellchart[k + 2];
        bch[3] = cellchart[k + 3];
        polyanticlockwise(bch);
        transform_points_forward(ccT, bch, bcht);

        cv::Point2f c(0, 0);
        for (int j = 0; j < 4; j++)
            c += bcht[j];
        c /= 4;
        for (size_t j = 0; j < 4; j++)
            bcht[j] = ((bcht[j] - c) * 0.50) + c;

        Rect roi = poly2mask(bcht, im_rgb.size(), mask);
        Mat submask = mask(roi);
        p_size = cv::sum(submask);

        // rgb space
        cv::meanStdDev(im_rgb.getMat()(roi), mu_rgb, st_rgb, submask);
        cv::minMaxLoc(rgb_planes[0](roi), &min_rgb[0], &max_rgb[0], NULL, NULL, submask);
        cv::minMaxLoc(rgb_planes[1](roi), &min_rgb[1], &max_rgb[1], NULL, NULL, submask);
        cv::minMaxLoc(rgb_planes[2](roi), &min_rgb[2], &max_rgb[2], NULL, NULL, submask);

        // create tabla
        //|p_size|average|stddev|max|min|
        // raw_r
        _charts_rgb.at<double>(3 * i + 0, 0) = p_size(0);
        _charts_rgb.at<double>(3 * i + 0, 1) = mu_rgb(0);
        _charts_rgb.at<double>(3 * i + 0, 2) = st_rgb(0);
        _charts_rgb.at<double>(3 * i + 0, 3) = min_rgb[0];
        _charts_rgb.at<double>(3 * i + 0, 4) = max_rgb[0];
        // raw_g
        _charts_rgb.at<double>(3 * i + 1, 0) = p_size(0);
        _charts_rgb.at<double>(3 * i + 1, 1) = mu_rgb(1);
        _charts_rgb.at<double>(3 * i + 1, 2) = st_rgb(1);
        _charts_rgb.at<double>(3 * i + 1, 3) = min_rgb[1];
        _charts_rgb.at<double>(3 * i + 1, 4) = max_rgb[1];
        // raw_b
        _charts_rgb.at<double>(3 * i + 2, 0) = p_size(0);
        _charts_rgb.at<double>(3 * i + 2, 1) = mu_rgb(2);
        _charts_rgb.at<double>(3 * i + 2, 2) = st_rgb(2);
        _charts_rgb.at<double>(3 * i + 2, 3) = min_rgb[2];
        _charts_rgb.at<double>(3 * i + 2, 4) = max_rgb[2];

        // YCbCr space
        cv::meanStdDev(im_ycbcr.getMat()(roi), mu_ycb, st_ycb, submask);
        cv::minMaxLoc(ycbcr_planes[0](roi), &min_ycb[0], &max_ycb[0], NULL, NULL, submask);
        cv::minMaxLoc(ycbcr_planes[1](roi), &min_ycb[1], &max_ycb[1], NULL, NULL, submask);
        cv::minMaxLoc(ycbcr_planes[2](roi), &min_ycb[2], &max_ycb[2], NULL, NULL, submask);

        // create tabla
        //|p_size|average|stddev|max|min|
        // raw_Y
        _charts_ycbcr.at<double>(3 * i + 0, 0) = p_size(0);
        _charts_ycbcr.at<double>(3 * i + 0, 1) = mu_ycb(0);
        _charts_ycbcr.at<double>(3 * i + 0, 2) = st_ycb(0);
        _charts_ycbcr.at<double>(3 * i + 0, 3) = min_ycb[0];
        _charts_ycbcr.at<double>(3 * i + 0, 4) = max_ycb[0];
        // raw_Cb
        _charts_ycbcr.at<double>(3 * i + 1, 0) = p_size(0);
        _charts_ycbcr.at<double>(3 * i + 1, 1) = mu_ycb(1);
        _charts_ycbcr.at<double>(3 * i + 1, 2) = st_ycb(1);
        _charts_ycbcr.at<double>(3 * i + 1, 3) = min_ycb[1];
        _charts_ycbcr.at<double>(3 * i + 1, 4) = max_ycb[1];
        // raw_Cr
        _charts_ycbcr.at<double>(3 * i + 2, 0) = p_size(0);
        _charts_ycbcr.at<double>(3 * i + 2, 1) = mu_ycb(2);
        _charts_ycbcr.at<double>(3 * i + 2, 2) = st_ycb(2);
        _charts_ycbcr.at<double>(3 * i + 2, 3) = min_ycb[2];
        _charts_ycbcr.at<double>(3 * i + 2, 4) = max_ycb[2];

        submask.setTo(Scalar::all(0));
    }
    charts_rgb.assign(_charts_rgb);
    charts_ycbcr.assign(_charts_ycbcr);
}

float CCheckerDetectorImpl::
    cost_function(InputArray im_rgb, InputOutputArray mask, InputArray lab,
                  const std::vector<cv::Point2f> &ibox, const TYPECHART chartType)
{
    CChartModel cccm(chartType);
    std::vector<cv::Point2f> fbox = cccm.box;
    std::vector<cv::Point2f> cellchart = cccm.cellchart;

    // tranformation
    Matx33f ccT = cv::getPerspectiveTransform(fbox, ibox);
    std::vector<cv::Point2f> bch(4), bcht(4);

    int N = (int)(cellchart.size() / 4);

    cv::Mat _lab = lab.getMat();
    cv::Mat _im_rgb = im_rgb.getMat();

    float ec = 0, es = 0;
    for (int i = 0, k; i < N; i++)
    {
        cv::Vec3f r = _lab.at<cv::Vec3f>(i);

        k = 4 * i;
        bch[0] = cellchart[k + 0];
        bch[1] = cellchart[k + 1];
        bch[2] = cellchart[k + 2];
        bch[3] = cellchart[k + 3];
        polyanticlockwise(bch);
        transform_points_forward(ccT, bch, bcht);

        cv::Point2f c(0, 0);
        for (int j = 0; j < 4; j++)
            c += bcht[j];
        c /= 4;
        for (int j = 0; j < 4; j++)
            bcht[j] = ((bcht[j] - c) * 0.75) + c;

        cv::Scalar mu, st;
        Rect roi = poly2mask(bcht, _im_rgb.size(), mask);
        if (!roi.empty())
        {
            Mat submask = mask.getMat()(roi);
            cv::meanStdDev(_im_rgb(roi), mu, st, submask);
            submask.setTo(Scalar::all(0));

            // cos error
            float costh;
            costh = (float)(mu.dot(cv::Scalar(r)) / (norm(mu) * norm(r) + FLT_EPSILON));
            ec += (1 - (1 + costh) / 2);

            // standar desviation
            es += (float)st.dot(st);
        }
    }

    // J = arg min ec + es
    float J = ec + es;
    return J / N;
}

} // namespace mcc
} // namespace cv