Android:使用 JavaCV 进行形状检测

Question

我是 JavaCV 的新手。我正在尝试检测图像中最大的矩形并在原始图像上用颜色勾勒出它的轮廓。我在下面发布了我已经尝试过但无法正常工作的代码。我正在正确获取 edgeDetectedImage。我正确地得到 4 个角点。只是 cvDrawLine 不起作用。如果我遗漏了什么，请帮助:

按钮的

OnClick 我正在处理图像并在 ImageView 上再次显示它。在按钮的onClickListener中:

if ((new File(path + "trial.jpg")).exists()) {
   opencv_core.IplImage originalImage = opencv_imgcodecs.cvLoadImage(path + "trial.jpg", opencv_imgcodecs.CV_IMWRITE_JPEG_QUALITY);
   opencv_core.IplImage iplImage = opencv_imgcodecs.cvLoadImage(path + "trial.jpg", opencv_imgcodecs.CV_LOAD_IMAGE_GRAYSCALE);
   opencv_core.IplImage edgeDetectedImage = applyCannyRectangleEdgeDetection(iplImage, 80);
   opencv_core.CvSeq largestContour = findLargestContour(edgeDetectedImage);
   opencv_core.CvPoint[] cvPoints = new opencv_core.CvPoint[4];
   for(int i=0; i<largestContour.total();i++)
   {
     opencv_core.CvPoint cvPoint = new opencv_core.CvPoint(cvGetSeqElem(largestContour, i));
     cvPoints[i] = cvPoint;
   }
   cvDrawLine(originalImage, cvPoints[0], cvPoints[1], opencv_core.CvScalar.YELLOW, 10, 10, 10);
   cvDrawLine(originalImage, cvPoints[1], cvPoints[2], opencv_core.CvScalar.YELLOW, 10, 10, 10);
   cvDrawLine(originalImage, cvPoints[2], cvPoints[3], opencv_core.CvScalar.YELLOW, 10,10, 10);
   cvDrawLine(originalImage, cvPoints[3], cvPoints[0], opencv_core.CvScalar.YELLOW, 10, 10,10);
   opencv_imgcodecs.cvSaveImage(path + "img1.jpg", originalImage);
                    if ((new File(path + "img1.jpg").exists())) {
                        imageView.setImageDrawable(Drawable.createFromPath(path + "img1.jpg"));
                    }
                }

方法applyCannyRectangleEdgeDetection(IplImage, int):

private opencv_core.IplImage applyCannyRectangleEdgeDetection(opencv_core.IplImage iplImage, int percent) {
    opencv_core.IplImage destImage = downScaleImage(iplImage, percent);
    OpenCVFrameConverter.ToMat converterToMat = new OpenCVFrameConverter.ToMat();
    Frame grayImageFrame = converterToMat.convert(destImage);
    opencv_core.Mat grayImageMat = converterToMat.convertToMat(grayImageFrame);
    GaussianBlur(grayImageMat, grayImageMat, new opencv_core.Size(5, 5), 0.0, 0.0, BORDER_DEFAULT);
    destImage = converterToMat.convertToIplImage(grayImageFrame);
    cvErode(destImage, destImage);
    cvDilate(destImage, destImage);
    cvCanny(destImage, destImage, 20, 55);
    return destImage;
}

方法 downScaleImage(IplImage, int)

private opencv_core.IplImage downScaleImage(opencv_core.IplImage srcImage, int percent) {
    opencv_core.IplImage destImage = cvCreateImage(cvSize((srcImage.width() * percent) / 100, (srcImage.height() * percent) / 100), srcImage.depth(), srcImage.nChannels());
    cvResize(srcImage, destImage);
    return destImage;
}

方法 findLargestContour(IplImage)

private opencv_core.CvSeq findLargestContour(opencv_core.IplImage edgeDetectedImage) {
    opencv_core.IplImage foundContoursOfImage = cvCloneImage(edgeDetectedImage);
    opencv_core.CvMemStorage memory = new opencv_core.CvMemStorage().create();
    opencv_core.CvSeq contours = new opencv_core.CvSeq();
    cvFindContours(foundContoursOfImage, memory, contours, Loader.sizeof(opencv_core.CvContour.class), CV_RETR_LIST, CV_CHAIN_APPROX_SIMPLE, new opencv_core.CvPoint(0, 0));
    int maxWidth = 0;
    int maxHeight = 0;
    opencv_core.CvRect contr = null;
    opencv_core.CvSeq seqFound = null;
    opencv_core.CvSeq nextSeq;
    for (nextSeq = contours; nextSeq != null; nextSeq = nextSeq.h_next()) {
        contr = cvBoundingRect(nextSeq, 0);
        if ((contr.width() >= maxWidth) && (contr.height() >= maxHeight)) {
            maxHeight = contr.height();
            maxWidth = contr.width();
            seqFound = nextSeq;
        }
    }
    opencv_core.CvSeq result = cvApproxPoly(seqFound, Loader.sizeof(opencv_core.CvContour.class), memory, CV_POLY_APPROX_DP, cvContourPerimeter(seqFound) * 0.1, 0);
    return result;
}

Answer 1

抱歉，这应该在评论中，但我没有足够的声誉。我从您的代码中可以看出，canny 应用于缩小的图像，轮廓也是如此。您在原始图像上绘制线条(未按百分比缩小)所以它自然看起来不正确(如果它看起来不正确但正在绘制某些东西)。否则，您应该提及图像的色彩空间，这对绘图无关紧要，但对精明而言却很重要。