Question

I have sample image processing code which applies a convolution filter over an image using following kernel:

float kernel[] = {-1,0,1};

and the application of filter:

new ConvolveOp(new Kernel(1,3, kernel), ConvolveOp.EDGE_NO_OP, null).filter(copy, img);

I am wondering how ConvolveOp will behave, dealing with a non-square kernel matrix? I far as I know we must use a square matrix with odd number of rows in Convolution algorithm.

PS. I thought that it (java) may pad it with zeros (e.g. {0,0,0,-1,0,1,0,0,0}) but I a more sophisticated case we can also have a kernel of this form with no run-time error:

    float data[] = {
        -1,0,1,
        -1,0,1,
        -1,0,1,
        -1,0,1
    };

which cannot be padded to become square with odd number of rows.

thanks

Answer 1

There is nothing that says that a convolution kernel has to be a square matrix.