bayesian_matting.py 文件源码

def matlab_style_gauss2d(shape=(3, 3), sigma=0.5):
    """
    2D gaussian mask - should give the same result as MATLAB's
    fspecial('gaussian',[shape],[sigma])
    """
    m, n = [(ss-1.)/2. for ss in shape]
    y, x = np.ogrid[-m:m+1, -n:n+1]
    h = np.exp(-(x*x + y*y)/(2.*sigma*sigma))
    h[h < np.finfo(h.dtype).eps*h.max()] = 0
    sumh = h.sum()
    if sumh != 0:
        h /= sumh
    return h


# returns the surrounding N-rectangular neighborhood of matrix m, centered
# at pixel (x,y), (odd valued N)