卷积:将conv2 matlab转换为opencv

时间:2018-11-23 16:50:21

标签: c++ matlab opencv convolution cross-correlation

如果我将输入图像和内核设置为:

cv::Mat conv2D(cv::Mat input,cv::Mat kernel){
    printf("\nINPUT [ %d x %d ]\n",input.rows,input.cols);
    cout << input <<endl;
    printf("\nKERNEL [ %d x %d ]\n",kernel.rows,kernel.cols);
    cout << kernel <<endl;

    cv::Mat flipped_kernel;
    cv::Point2i pad;
    cv::Mat result;
    cv::Mat padded;
    cv::Point anchor(kernel.cols -kernel.cols/2 -1,kernel.rows -kernel.rows/2 -1 );
    cv::flip( kernel, flipped_kernel, -1 );
    printf("\nFLIPPED KERNEL [ %d x %d ]\n",flipped_kernel.rows,flipped_kernel.cols);
    cout << flipped_kernel <<endl;
    pad = cv::Point2i( kernel.cols - 1, kernel.rows - 1);
    cv::copyMakeBorder( input, padded, pad.y, pad.y, pad.x, pad.x, cv::BORDER_CONSTANT,cv::Scalar(0) );
    cv::Rect region = cv::Rect( pad.x / 2, pad.y / 2, padded.cols - pad.x, padded.rows - pad.y);

    cv::filter2D( padded, result , padded.depth(), flipped_kernel, anchor, 0, cv::BORDER_CONSTANT );
    cv::Mat out = result( region );

    printf("\nOUT [ %d x %d ]\n",out.rows,out.cols);
    cout << out <<endl;
    return out;
}

输出不同于:

conv2(input,kernel)

在网络上阅读时,有人说翻转是错误,而其他人则不是。所以我使用对称旋转矩阵作为内核,结果仍然是错误的。我如何在c ++中执行2d卷积和2d互相关?

1 个答案:

答案 0 :(得分:0)

问题在于我通过了CV_32F类型的opencv垫,这给了我错误的结果,我使用以下方法将图像转换为CV_8UC1:

input.convertTo(input,CV_8UC1); flipped_kernel.convertTo(input,CV_8UC1);

并在操作后更改输出类型:

out.convertTo(input,CV_32F);

因此该方法有效!