OpenCV C ++使用高斯内核函数计算晕影效果

Question

I try to convert this reply from python to C++ and I'm stuck on the first call to multiply... small is my input Mat with 2 dims and about 600x400 cols/rows.

Mat a = getGaussianKernel(small.cols, .3);
Mat b = getGaussianKernel(small.rows, .3);
Mat ta;
transpose(a, ta);
Mat c = *new Mat(ta.rows, ta.cols, ta.type());
cv::multiply(ta, b, c);
Mat d;
cv::max(c, d);
d = c / d;
Mat e;
multiply(small, d, e);

The error message reads:

The operation is neither 'array op array' (where arrays have the same size and the same number of channels), nor 'array op scalar', nor 'scalar op array' in function arithm_op

I'm not familiar with either numpy or matplotlib but I'm constraint to C++ for reasons to lengthy to explain...

Answer 1

If you read the documentation for cv::multiply you will see that it expects the first two inputs to be the same size and type. This is because it is attempting to do element-wise multiplication .

Your Gaussian kernel a is defined to be 600 x 1 (which getGaussianKernel recommends the first input to be odd ) so the transpose ( ta ) is going to be 1 x 600.

The second input, b is defined to be 400 x 1 (again, should really be an odd size).

Obviously these two are not the same size dimensions/size.

Instead of using cv::multiply which is element-wise multiplication, you'll want to just use the * operator which is the matrix product that you likely want.

Mat C = b * ta

Which should yield a 400 x 600 matrix.

Also, I switched the order of b and ta so that the matrix dimensions worked out for multiplication.

Note you also have some other issues with your code. You're using cv::max() to compare an empty matrix with c . You should likely use cv::minMaxLoc .

The multiplication on your last line looks like it should be fine since you want to apply the combination of Gaussians image directly to the input image on a pixel-by-pixel or element-by-element basis.