Fork me on GitHub

基于opencv.js实现二维码定位

    通过分析OpenCV.JS(官方下载地址 https://docs.opencv.org/_VERSION_/opencv.js)的白名单,我们可以了解目前官方PreBuild版本并没有实现QR识别。
# Classes and methods whitelist
core = {'': ['absdiff''add''addWeighted''bitwise_and''bitwise_not''bitwise_or''bitwise_xor''cartToPolar',\
             'compare''convertScaleAbs''copyMakeBorder''countNonZero''determinant''dft''divide''eigen', \
             'exp''flip''getOptimalDFTSize','gemm''hconcat''inRange''invert''kmeans''log''magnitude', \
             'max''mean''meanStdDev''merge''min''minMaxLoc''mixChannels''multiply''norm''normalize', \
             'perspectiveTransform''polarToCart''pow''randn''randu''reduce''repeat''rotate''setIdentity''setRNGSeed', \
             'solve''solvePoly''split''sqrt''subtract''trace''transform''transpose''vconcat'],
        'Algorithm': []}

imgproc = {'': ['Canny''GaussianBlur''Laplacian''HoughLines''HoughLinesP''HoughCircles''Scharr','Sobel', \
                'adaptiveThreshold','approxPolyDP','arcLength','bilateralFilter','blur','boundingRect','boxFilter',\
                'calcBackProject','calcHist','circle','compareHist','connectedComponents','connectedComponentsWithStats', \
                'contourArea''convexHull''convexityDefects''cornerHarris','cornerMinEigenVal','createCLAHE', \
                'createLineSegmentDetector','cvtColor','demosaicing','dilate''distanceTransform','distanceTransformWithLabels', \
                'drawContours','ellipse','ellipse2Poly','equalizeHist','erode''filter2D''findContours','fitEllipse', \
                'fitLine''floodFill','getAffineTransform''getPerspectiveTransform''getRotationMatrix2D''getStructuringElement', \
                'goodFeaturesToTrack','grabCut','initUndistortRectifyMap''integral','integral2''isContourConvex''line', \
                'matchShapes''matchTemplate','medianBlur''minAreaRect''minEnclosingCircle''moments''morphologyEx', \
                'pointPolygonTest''putText','pyrDown','pyrUp','rectangle','remap''resize','sepFilter2D','threshold', \
                'undistort','warpAffine','warpPerspective','warpPolar','watershed', \
                'fillPoly''fillConvexPoly'],
           'CLAHE': ['apply''collectGarbage''getClipLimit''getTilesGridSize''setClipLimit''setTilesGridSize']}

objdetect = {'': ['groupRectangles'],
             'HOGDescriptor': ['load''HOGDescriptor''getDefaultPeopleDetector''getDaimlerPeopleDetector''setSVMDetector''detectMultiScale'],
             'CascadeClassifier': ['load''detectMultiScale2''CascadeClassifier''detectMultiScale3''empty''detectMultiScale']}

video = {'': ['CamShift''calcOpticalFlowFarneback''calcOpticalFlowPyrLK''createBackgroundSubtractorMOG2', \
             'findTransformECC''meanShift'],
         'BackgroundSubtractorMOG2': ['BackgroundSubtractorMOG2''apply'],
         'BackgroundSubtractor': ['apply''getBackgroundImage']}

dnn = {'dnn_Net': ['setInput''forward'],
       '': ['readNetFromCaffe''readNetFromTensorflow''readNetFromTorch''readNetFromDarknet',
            'readNetFromONNX''readNet''blobFromImage']}

features2d = {'Feature2D': ['detect''compute''detectAndCompute''descriptorSize''descriptorType''defaultNorm''empty''getDefaultName'],
              'BRISK': ['create''getDefaultName'],
              'ORB': ['create''setMaxFeatures''setScaleFactor''setNLevels''setEdgeThreshold''setFirstLevel''setWTA_K''setScoreType''setPatchSize''getFastThreshold''getDefaultName'],
              'MSER': ['create''detectRegions''setDelta''getDelta''setMinArea''getMinArea''setMaxArea''getMaxArea''setPass2Only''getPass2Only''getDefaultName'],
              'FastFeatureDetector': ['create''setThreshold''getThreshold''setNonmaxSuppression''getNonmaxSuppression''setType''getType''getDefaultName'],
              'AgastFeatureDetector': ['create''setThreshold''getThreshold''setNonmaxSuppression''getNonmaxSuppression''setType''getType''getDefaultName'],
              'GFTTDetector': ['create''setMaxFeatures''getMaxFeatures''setQualityLevel''getQualityLevel''setMinDistance''getMinDistance''setBlockSize''getBlockSize''setHarrisDetector''getHarrisDetector''setK''getK''getDefaultName'],
              # 'SimpleBlobDetector': ['create'],
              'KAZE': ['create''setExtended''getExtended''setUpright''getUpright''setThreshold''getThreshold''setNOctaves''getNOctaves''setNOctaveLayers''getNOctaveLayers''setDiffusivity''getDiffusivity''getDefaultName'],
              'AKAZE': ['create''setDescriptorType''getDescriptorType''setDescriptorSize''getDescriptorSize''setDescriptorChannels''getDescriptorChannels''setThreshold''getThreshold''setNOctaves''getNOctaves''setNOctaveLayers''getNOctaveLayers''setDiffusivity''getDiffusivity''getDefaultName'],
              'DescriptorMatcher': ['add''clear''empty''isMaskSupported''train''match''knnMatch''radiusMatch''clone''create'],
              'BFMatcher': ['isMaskSupported''create'],
              '': ['drawKeypoints''drawMatches''drawMatchesKnn']}

photo = {'': ['createAlignMTB''createCalibrateDebevec''createCalibrateRobertson', \
              'createMergeDebevec''createMergeMertens''createMergeRobertson', \
              'createTonemapDrago''createTonemapMantiuk''createTonemapReinhard''inpaint'],
        'CalibrateCRF': ['process'],
        'AlignMTB' : ['calculateShift''shiftMat''computeBitmaps''getMaxBits''setMaxBits', \
                      'getExcludeRange''setExcludeRange''getCut''setCut'],
        'CalibrateDebevec' : ['getLambda''setLambda''getSamples''setSamples''getRandom''setRandom'],
        'CalibrateRobertson' : ['getMaxIter''setMaxIter''getThreshold''setThreshold''getRadiance'],
        'MergeExposures' : ['process'],
        'MergeDebevec' : ['process'],
        'MergeMertens' : ['process''getContrastWeight''setContrastWeight''getSaturationWeight', \
                          'setSaturationWeight''getExposureWeight''setExposureWeight'],
        'MergeRobertson' : ['process'],
        'Tonemap' : ['process' , 'getGamma''setGamma'],
        'TonemapDrago' : ['getSaturation''setSaturation''getBias''setBias', \
                          'getSigmaColor''setSigmaColor''getSigmaSpace','setSigmaSpace'],
        'TonemapMantiuk' : ['getScale''setScale''getSaturation''setSaturation'],
        'TonemapReinhard' : ['getIntensity''setIntensity''getLightAdaptation''setLightAdaptation', \
                             'getColorAdaptation''setColorAdaptation']
        }

aruco = {'': ['detectMarkers''drawDetectedMarkers''drawAxis''estimatePoseSingleMarkers''estimatePoseBoard''estimatePoseCharucoBoard''interpolateCornersCharuco''drawDetectedCornersCharuco'],
        'aruco_Dictionary': ['get''drawMarker'],
        'aruco_Board': ['create'],
        'aruco_GridBoard': ['create''draw'],
        'aruco_CharucoBoard': ['create''draw'],
        }

calib3d = {'': ['findHomography''calibrateCameraExtended''drawFrameAxes''estimateAffine2D''getDefaultNewCameraMatrix''initUndistortRectifyMap''Rodrigues']}


white_list = makeWhiteList([core, imgproc, objdetect, video, dnn, features2d, photo, aruco, calib3d])

    但是我们仍然可以通过轮廓分析的相关方法,去实现“基于opencv.js实现二维码定位”,这就是本篇BLOG的主要内容。
一、基本原理
        主要内容请参考《OpenCV使用FindContours进行二维码定位》,这里重要的回顾一下。
        使用过FindContours直接寻找联通区域的函数。典型的运用在二维码上面:
    对于它的3个定位点,这种重复包含的特性,在图上只有不容易重复的三处,这是具有排它性的。
    那么轮廓识别的结果是如何展示的了?比如在这幅图中(白色区域为有数据的区域,黑色为无数据),0,1,2是第一层,然后里面是3,3的里面是4和5。(2a表示是2的内部),他们的关系应该是这样的:
hierarchy.png
所以我们只需要寻找某一个轮廓“有无爷爷轮廓”,就可以判断出来它是否“重复包含”
值得参考的C++代码应该是这样的,其中注释部分已经说明的比较清楚。
#include "opencv2/highgui/highgui.hpp"
#include "opencv2/imgproc/imgproc.hpp"
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#include <math.h>
using namespace cv;
using namespace std;
//找到所提取轮廓的中心点
//在提取的中心小正方形的边界上每隔周长个像素提取一个点的坐标,求所提取四个点的平均坐标(即为小正方形的大致中心)
Point Center_cal(vector<vector<Point> > contours,int i)
{
    int centerx=0,centery=0,n=contours[i].size();
    centerx = (contours[i][n/4].x + contours[i][n*2/4].x + contours[i][3*n/4].x + contours[i][n-1].x)/4;
    centery = (contours[i][n/4].y + contours[i][n*2/4].y + contours[i][3*n/4].y + contours[i][n-1].y)/4;
    Point point1=Point(centerx,centery);
    return point1;
}
int main( int argc, char** argv[] )
{
    Mat src = imread( "e:/sandbox/qrcode.jpg"1 );
    resize(src,src,Size(800,600));//标准大小
    Mat src_gray;
    Mat src_all=src.clone();
    Mat threshold_output;
    vector<vector<Point> > contours,contours2;
    vector<Vec4i> hierarchy;
    //预处理
    cvtColor( src, src_gray, CV_BGR2GRAY );
    blur( src_gray, src_gray, Size(3,3) ); //模糊,去除毛刺
    threshold( src_gray, threshold_output, 100255, THRESH_OTSU );
    //寻找轮廓 
    //第一个参数是输入图像 2值化的
    //第二个参数是内存存储器,FindContours找到的轮廓放到内存里面。
    //第三个参数是层级,**[Next, Previous, First_Child, Parent]** 的vector
    //第四个参数是类型,采用树结构
    //第五个参数是节点拟合模式,这里是全部寻找
    findContours( threshold_output, contours, hierarchy,  CV_RETR_TREE, CHAIN_APPROX_NONE, Point(00) );
    //轮廓筛选
    int c=0,ic=0,area=0;
    int parentIdx=-1;
    forint i = 0; i< contours.size(); i++ )
    {
        //hierarchy[i][2] != -1 表示不是最外面的轮廓
        if (hierarchy[i][2!= -1 && ic==0)
        {
            parentIdx = i; 
            ic++;
        }
        else if (hierarchy[i][2!= -1)
        {
            ic++;
        }
        //最外面的清0
        else if(hierarchy[i][2== -1)
        {
            ic = 0;
            parentIdx = -1;
        }
        //找到定位点信息
        if ( ic >= 2)
        {
            contours2.push_back(contours[parentIdx]);
            ic = 0;
            parentIdx = -1;
        }
    }
    //填充定位点
    for(int i=0; i<contours2.size(); i++)
        drawContours( src_all, contours2, i,  CV_RGB(0,255,0) , -1 );
    //连接定位点
    Point point[3];
    for(int i=0; i<contours2.size(); i++)
    {
        point[i] = Center_cal( contours2, i );
    }
    
    line(src_all,point[0],point[1],Scalar(0,0,255),2);
    line(src_all,point[1],point[2],Scalar(0,0,255),2);
    line(src_all,point[0],point[2],Scalar(0,0,255),2);
     
    imshow( "结果", src_all );
    waitKey(0);
    return(0);
}

二、算法重点
        由于hierarchy这块是比较缺乏文档的,在转换为JS的过程中存在一定困难,最终得到了以下的正确结果:
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<title>Hello OpenCV.js</title>
<script async src="opencv.js" onload="onOpenCvReady();" type="text/javascript"></script>
</head>
<body>
<h2>Hello OpenCV.js</h2>
<p id="status">OpenCV.js is loading...</p>
<div>
  <div class="inputoutput">
    <img id="imageSrc" alt="No Image" />
    <div class="caption">imageSrc <input type="file" id="fileInput" name="file" /></div>
  </div>
  <div class="inputoutput">
    <canvas id="canvasOutput" ></canvas>
    <div class="caption">canvasOutput</div>
  </div>
  <div class="inputoutput2">
    <canvas id="canvasOutput2" ></canvas>
    <div class="caption">canvasOutput2</div>
  </div>
</div>
<script type="text/javascript">
let imgElement = document.getElementById('imageSrc');
let inputElement = document.getElementById('fileInput');
inputElement.addEventListener('change', (e) => {
  imgElement.src = URL.createObjectURL(e.target.files[0]);
}, false);
imgElement.onload = function() {
let src = cv.imread(imgElement);
let src_clone = cv.imread(imgElement);
let dsize = new cv.Size(800600);
// You can try more different parameters
cv.resize(src, src, dsize);cv.resize(src_clone, src_clone, dsize);
let dst = cv.Mat.zeros(src.rows,src.cols, cv.CV_8UC3);
cv.cvtColor(src, src, cv.COLOR_RGBA2GRAY, 0);
let ksize = new cv.Size(33);
// You can try more different parameters 
cv.blur(src, src, ksize); 

cv.threshold(src, src, 100255, cv.THRESH_OTSU);
let contours = new cv.MatVector();
let contours2 = new cv.MatVector();
let hierarchy = new cv.Mat();
// You can try more different parameters
cv.findContours(src, contours, hierarchy, cv.RETR_TREE, cv.CHAIN_APPROX_NONE);
//轮廓筛选
let c=0,ic=0,area=0;
let parentIdx = -1;
debugger
forlet i = 0; i< contours.size(); i++ )
{
    //let hier = hierarchy.intPtr(0, i)
    if (hierarchy.intPtr(0,i)[2] != -1 && ic==0)
    {
        parentIdx = i; 
        ic++;
    }
    else if (hierarchy.intPtr(0,i)[2] != -1)
    {
        ic++;
    }
    else if(hierarchy.intPtr(0,i)[2] == -1)
    {
        ic = 0;
        parentIdx = -1;
    }
    //找到定位点信息
    if ( ic >= 2)
    {
      //let cnt = matVec.get(0);
      contours2.push_back(contours.get(parentIdx));
      ic = 0;
      parentIdx = -1;
    }
    
}
console.log(contours2.size());

//填充定位点
for(let i=0; i<contours.size(); i++)
{
  let color = new cv.Scalar(25500255); 
  cv.drawContours(src_clone, contours, i,color,1);
}
cv.imshow('canvasOutput', src_clone);

for(let i=0; i<contours2.size(); i++)
{
  let color = new cv.Scalar(Math.round(Math.random() * 255), Math.round(Math.random() * 255),
                              Math.round(Math.random() * 255));
  cv.drawContours(dst, contours2, i, color, 1);
}
cv.imshow('canvasOutput2', dst);
src.delete(); src_clone.delete();
dst.delete(); contours.delete(); hierarchy.delete();

};
function onOpenCvReady() {
  document.getElementById('status').innerHTML = 'OpenCV.js is ready.';
}
</script>

</body>
</html>

其中绝大多数部分都和C++相似,不同的地方已经标红。它能够成功运行,并且得到正确的定位。(这里OpenCVJS的相关运行情况请参考官方教程)
三、研究收获
        这次研究的关键节点, 是建立了Debug机制。在JS代码中加入debugger语句,并且开启F12,则在调试的过程中,可以查看各个变量的信息。
此外,非常重要的参考资料,就是OpenCV的官方教程。如果希望进一步进行研究的话,首先需要先收集掌握所有现有资料。
感谢阅读至此,希望有所帮助。




目前方向:图像处理,人工智能
posted @ 2020-06-03 22:28  jsxyhelu  阅读(1053)  评论(0编辑  收藏  举报