Shapes not matching in numpy.convolve

Question

Error message:

operands could not be broadcast together with shapes (603) (613)

What should I do?
Do both of the list need to be the same length?
Or should I zero-pad it?

Here's my code:

def gaussian_smooth1(img, sigma): 
    '''
    Do gaussian smoothing with sigma.
    Returns the smoothed image.
    '''
    result = np.zeros_like(img)

    #get the filter
    filter = gaussian_filter(sigma)

    #get the height and width of img
    width = len(img[0])
    height = len(img)

    #smooth every color-channel
    for c in range(3):
        #smooth the 2D image img[:,:,c]
        #tip: make use of numpy.convolve
        for x in range(height):
            result[x,:,c] = np.convolve(filter,img[x,:,c])
        for y in range(width):
            result[:,y,c] = np.convolve(filter,img[:,y,c])
    return result

Answer 1

The problem arises because you do not specify the right mode .
Read it up in the documentation:
numpy.convolve

The default for numpy.convolve is mode='full' .

This returns the convolution at each point of overlap, with an output shape of (N+M-1,).

N is the size of the input array, M is the size of the filter. So the output is larger than the input.

Instead you want to use np.convolve(filter,img[...],mode='same') .

Also have a look at scipy.convolve which allows 2D convolution using the FFT.