通过dft找到一个块到框架

Question

我正在尝试解决运动跟踪的对应问题：

在帧n中从帧n-1中找到给定的维数m * n块。我正在使用opencv和python（这两个都是第一次），我正在计算归一化的平方差，但它太慢了。我想在某种程度上我可以使用离散傅里叶变换，但我无法弄清楚我是如何做到的！

def match(img, block):
    # img is the frame n, block is from frame n-1
    w, h = img.shape[:2]

    output = np.zeros( (w,h) ) + 255

    for x in range( w ):
        for y in range( h ):
           output[x, y] = evaluate(img, (x,y) , block)

    # the minimum value is the position of the block into the frame n
    return output    

def evaluate( img, point, block):
    m, n = block.shape[:2]  
    w, h = img.shape[:2]

    a = (m-1)/2
    b = (n-1)/2

    x, y = point
    response = 0 

    for s in range( -a, a+1 ):
       for t in range(-b, b+1 ):

         if x+s >= w or x+s < 0 or y+t >= h or y+t < 0:  
            pixel = 0
         else:
            pixel = img[x+s, y+t]

         # normalized squared difference
         response = response + (pow((block[ 1+s, 1+t] - pixel), 2) / (m*n))

    return response