How to set a mask image for grabCut in OpenCV?

问题

How can I set a mask image for the grabCut function in OpenCV? I want to do GC_INIT_WITH_MASK with the options

GC_BGD    = 0,
GC_FGD    = 1,
GC_PR_BGD = 2,
GC_PR_FGD = 3,

If you can answer this with JavaCV it would be great as I am doing this in Scala/Java.

回答1:

Input images (lena.png, lena_mask.png):

Code (JAVA):

import org.opencv.core.Core;
import org.opencv.core.CvType;
import org.opencv.core.Mat;
import org.opencv.core.Rect;
import org.opencv.core.Scalar;
import org.opencv.highgui.Highgui;
import org.opencv.imgproc.Imgproc;

/**
 * Example how to use grabCut algorithm in OpenCV using GC_INIT_WITH_MASK.
 *
 */
public class ImageSegmentByMask {

    public static void main(String[] args) {
        System.loadLibrary(Core.NATIVE_LIBRARY_NAME);

        // will read image
        Mat image = Highgui.imread("lena.png");
        Mat mask = Highgui.imread("lena_mask.png", Highgui.CV_LOAD_IMAGE_GRAYSCALE);

        Rect rectangle = new Rect(10, 10, image.cols() - 20, image.rows() - 20);

        Mat bgdModel = new Mat(); // extracted features for background
        Mat fgdModel = new Mat(); // extracted features for foreground
        Mat source = new Mat(1, 1, CvType.CV_8U, new Scalar(0));

        convertToOpencvValues(mask); // from human readable values to OpenCV values 

        int iterCount = 1;
        Imgproc.grabCut(image, mask, rectangle, bgdModel, fgdModel, iterCount, Imgproc.GC_INIT_WITH_MASK);


        convertToHumanValues(mask); // back to human readable values
        Imgproc.threshold(mask,mask,128,255,Imgproc.THRESH_TOZERO);

        Mat foreground = new Mat(image.size(), CvType.CV_8UC1, new Scalar(0, 0, 0));
        image.copyTo(foreground, mask);

        Highgui.imwrite("sandbox/sucess1.jpg", foreground);
    }

    private static void convertToHumanValues(Mat mask) {
        byte[] buffer = new byte[3];
        for (int x = 0; x < mask.rows(); x++) {
            for (int y = 0; y < mask.cols(); y++) {
                mask.get(x, y, buffer);
                int value = buffer[0];
                if (value == Imgproc.GC_BGD) {
                    buffer[0] = 0; // for sure background
                } else if (value == Imgproc.GC_PR_BGD) {
                    buffer[0] = 85; // probably background
                } else if (value == Imgproc.GC_PR_FGD) {
                    buffer[0] = (byte) 170; // probably foreground
                } else {
                    buffer[0] = (byte) 255; // for sure foreground

                }
                mask.put(x, y, buffer);
            }
        }
    }

    /**
     * Converts level of grayscale into OpenCV values. White - foreground, Black
     * - background.
     * 
     * @param mask
     */
    private static void convertToOpencvValues(Mat mask) {
        byte[] buffer = new byte[3];
        for (int x = 0; x < mask.rows(); x++) {
            for (int y = 0; y < mask.cols(); y++) {
                mask.get(x, y, buffer);
                int value = buffer[0];
                if (value >= 0 && value < 64) {
                    buffer[0] = Imgproc.GC_BGD; // for sure background
                } else if (value >= 64 && value < 128) {
                    buffer[0] = Imgproc.GC_PR_BGD; // probably background
                } else if (value >= 128 && value < 192) {
                    buffer[0] = Imgproc.GC_PR_FGD; // probably foreground
                } else {
                    buffer[0] = Imgproc.GC_FGD; // for sure foreground

                }
                mask.put(x, y, buffer);
            }
        }

    }

}

Result:

回答2:

Working example in OpenCV:

//Fill with the background value
Mat mask = cv::Mat::ones(src.size(), CV_8U) * cv::GC_BGD;

//Fill a smaller rectangle with the probably-foreground value.
Rect area;
area.x=10;  area.y=10;
area.width=250; area.height=250;
rectangle(mask, area , cv::Scalar(cv::GC_PR_FGD),-1,8,0);

//Fill a smaller rectangle with the foreground value.
area.x=50;  area.y=50;
area.width=20;  area.height=20;
rectangle(mask, area , cv::Scalar(cv::GC_FGD),-1,8,0);

Mat bgd,fgd;
cv::grabCut(src, mask, area, bgd, fgd, 1, cv::GC_INIT_WITH_MASK);

//Visualize results.
compare(mask,cv::GC_FGD,mask,cv::CMP_EQ);
Mat foreground(src.size(),CV_8UC3, cv::Scalar(0,0,0));
src.copyTo(foreground,mask);
imshow("segm",foreground);

回答3:

public class ImageSegment {

    public static void main(String[] args) {
        System.loadLibrary("opencv_java244");

        Mat image = null;
        image = Highgui.imread("syh.jpg");

        Rect rectangle = new Rect(25,25,image.cols()-64,image.rows()-64);
        Mat result = new Mat();
        Mat bgdModel = new Mat();
        Mat fgdModel = new Mat();
        Mat source = new Mat(1, 1, CvType.CV_8U, new Scalar(3));
        Imgproc.grabCut(image, result, rectangle, bgdModel, fgdModel, 1, 0);
        Core.compare(result, source,result, Core.CMP_EQ);
        Mat foreground= new Mat(image.size(), CvType.CV_8UC1, new Scalar(0, 0, 0));
        image.copyTo(foreground, result);
        Highgui.imwrite("sucess1.jpg", foreground);
        System.out.println("grabcut sucess!");
    }

}

回答4:

The initial image mask is the second argument to grabCut, so you should just be able to create an image with CvMat.create() and then set the values appropriately to GC_BGD, GC_FGC, GC_PR_BGD, GC_PR_FGD, for example (not able to test code on this computer):

val mask = cvCreateImage(cvGetSize(image), IPL_DEPTH_8U, 1)
// Set mask pixel values to what you want
mask.put(offset, GC_BGD)
...
grabCut(image, mask, ... <your other options here>)

I'm not the most familiar with javacv, there is probably a good way to set the values in ranges or blocks for the mask instead of one pixel at a time.

Also, according to the docs for grabCut:

"Note that GC_INIT_WITH_RECT and GC_INIT_WITH_MASK can be combined, then all the pixels outside of the ROI are automatically initialized with GC_BGD"

来源：https://stackoverflow.com/questions/14111716/how-to-set-a-mask-image-for-grabcut-in-opencv

标签

OpenCV

javacv