在多维数组中查找模式

Question

I've got a 4 dimensional array in which I need to find some patters. The patter is to be evaluated in the last two dimensions, scanning across the first two. It is somewhat like using a hit or miss function over an image, but instead of evaluating across the two dimensions of the image for patters on the values of these points, I need scan across the two dimensions looking for a pattern in the two other dimensions of the point and its surroundings. My current code looks like this, but considering it will look for patterns other than the two ones described bellow, as well as run in real time, I need to make in run much faster.

image = np.empty((240,320,2,3), np.bool)

# image receive some data

for y in range(1, 239):
    for x in range(1, 319):
        if image[y][x-1][1][1] and image[y][x][1][2]:
            FindPattern1[y][x] = True
        if image[y-1][x][0][1] and image[y][x][0][2]:
            FindPattern2[y][x] = True

I thought about using the hit or miss transform with some multidimensional structure to scan across image[y][x] , but was unable to get it working perfectly. Do you guys have some idea for solving this problem?

Answer 1

Seems like you simply just need to do some fancy indexing and a logical_and .

find_pattern_1 = np.logical_and(image[1:, :-1, 1, 1], image[1:, 1:, 1, 2])
find_pattern_2 = np.logical_and(image[:-1, 1:, 0, 1], image[1:, 1:, 0, 2])