OpenCV java alpha 混合

Question

我正在尝试移植以下代码（从 C++ 到 Java）以在我的图像之间进行良好的 alpha 混合，但它不起作用：

#include opencv2/opencv.hpp

using namespace cv;
using namespace std;

int main(int argc, char** argv)
{

    // Read the images
    Mat foreground = imread("puppets.png");
    Mat background = imread("ocean.png");
    Mat alpha = imread("puppets_alpha.png");

    // Convert Mat to float data type
    foreground.convertTo(foreground, CV_32FC3);
    background.convertTo(background, CV_32FC3);

    // Normalize the alpha mask to keep intensity between 0 and 1
    alpha.convertTo(alpha, CV_32FC3, 1.0/255); // 

    // Storage for output image
    Mat ouImage = Mat::zeros(foreground.size(), foreground.type());

    // Multiply the foreground with the alpha matte
    multiply(alpha, foreground, foreground); 

    // Multiply the background with ( 1 - alpha )
    multiply(Scalar::all(1.0)-alpha, background, background); 

    // Add the masked foreground and background.
    add(foreground, background, ouImage); 

    // Display image
    imshow("alpha blended image", ouImage/255);
    waitKey(0);

    return 0;
}

代码可以在这里找到：https://www.learnopencv.com/alpha-blending-using-opencv-cpp-python/ 还有我的 Java 版本：

public static Mat alphaBlend(Mat background, Mat foreground) {
        Vector<Mat> rgba = new Vector<Mat>();
        // split RBGA image for separate channels
        Core.split(background, rgba);
        // get alpha channel
        Mat alpha = rgba.get(3);

        // Convert Mat to float data type
        foreground.convertTo(foreground, CvType.CV_32FC3);
        background.convertTo(background, CvType.CV_32FC3);

        // Normalize the alpha mask to keep intensity between 0 and 1
        alpha.convertTo(alpha, CvType.CV_32FC3, 1.0/255); //
        Imgproc.cvtColor(alpha,alpha, Imgproc.COLOR_GRAY2BGRA,4);

        Mat outImage = Mat.zeros(foreground.size(),foreground.type());

        // Multiply the foreground with the alpha matte
        Core.multiply(alpha, foreground, foreground);

        Mat kernel = new MatOfDouble(1.0);

        Core.subtract(kernel, alpha, alpha);

        // Multiply the background with ( 1 - alpha )
        Core.multiply(alpha, background, background);

        // Add the masked foreground and background.
        Core.add(foreground, background, outImage);

        Core.divide(new MatOfDouble(255), outImage,outImage);
        return outImage;
    }

我认为我的问题是在 java 中移植这个：

multiply(Scalar::all(1.0)-alpha, background, background); 

任何帮助将不胜感激！

谢谢

编辑： 这是一个工作版本，但边界不混合透明像素，导致图像交集边界处出现空白像素/线：

    private static Mat merge(Mat background, Mat foreground) {
        Vector<Mat> rgba = new Vector<Mat>();
        // split RBGA image for separate channels
        Core.split(background, rgba);
        // get alpha channel
        Mat alpha = rgba.get(3);
        // Convert Mat to float data type
        // Normalize the alpha mask to keep intensity between 0 and 1
        alpha.convertTo(alpha, CvType.CV_32FC3, 1.0/255); //
        Mat dst = new Mat(256,256,CvType.CV_8UC4);
        alpha.convertTo(alpha,CvType.CV_8UC1);
        Mat alphaInv = new Mat();
        // invert the mask
        Core.absdiff(new MatOfDouble(1.0), alpha, alphaInv);
        foreground.copyTo(dst, alphaInv);
        // case where foreground is full JPEG in BGR
        if(dst.type() != CvType.CV_8UC4) {
            Imgproc.cvtColor(dst, dst, Imgproc.COLOR_BGR2BGRA, 4);
        }
        background.copyTo(dst, alpha);
        return dst;
    }

Answer 1

根据您分享的链接，看起来目标只是使用图像遮罩，我认为有一种更简单的方法可以做到这一点，使用 copyTo 函数。试着把它放在你的 main 中，当它工作时转移到你的函数：

// Read the images
Mat foreground = Imgcodecs.imread("puppets.png");
Mat background = Imgcodecs.imread("ocean.png");
Mat mask = Imgcodecs.imread("puppets_alpha.png");
Mat notMask = new Mat();
Core.bitwise_not(mask, notMask);
Mat result = new Mat();
Core.copyTo(background, result, mask);
Core.copyTo(foreground, result, notMask);
Imgcodecs.imwrite("result.png", result);

Answer 2

我建议使用 Core 的 addWeighted 方法进行基本的 alpha 混合。

看起来你想要一个基于掩码的可变 alpha 混合，你需要按照 api 指定的顺序排列你的参数，比如addition/subtraction/multiplication。对于您的乘法，这意味着标量应该是第二个参数。对于您的减法，这意味着标量应该再次成为第二个参数（我发现自己将矩阵乘以负 1，然后添加到标量值）。