我正在尝试检测图像比例空间中的局部最小值和最大值。 Scale Space只是一个3D numpy数组，其中前2个维对应于图像像素值，第3个维对应于其比例尺（在每个比例中，我都使用高斯滤波器对图像进行滤波，使其更加模糊）。
这是我计算比例空间的方式：
def scalespace (image, size = 5, laplacian = True, fact = np.sqrt(2)):
"""Returns 3d numpy array containing scale space of image
Each step corresponds to the gaussian filter applied with a step = sqrt(2)
The first scale corresponds to the original image.
"""
scale_space = np.zeros((image.shape[0],image.shape[1],size))
image_copy = image.copy()
scale_space[:,:,0] = image_copy
for s in range(1,size):
#convolve on previous scale
filtered_image = ndi.filters.gaussian_filter(scale_space[:,:,s-1], sigma = fact*np.sqrt(s))
if laplacian:
filtered_image = ndi.filters.laplace(filtered_image)
scale_space[:,:,s] = filtered_image
return scale_space
现在，我必须找到局部最小值和最大值（我正在进行斑点检测）。对于单个图像（2D情况），我发现它非常简单，但是现在，我还需要考虑像素值在所有比例上是否为最大值/最小值，我发现这很困难。在2d情况下，我是这样的：
if len(image.shape) == 2:
min_x,min_y = scipy.signal.argrelextrema(image, np.less, order = n_order, axis = 0)
min_2d = [(j,i) for i,j in zip(min_x,min_y)]
max_x,max_y = scipy.signal.argrelextrema(image, np.greater, order = n_order, axis = 0)
max_2d = [(j,i) for i,j in zip(max_x,max_y)]
#check on the other direction
min_x,min_y = scipy.signal.argrelextrema(image, np.less, order = n_order, axis = 1)
new_min_2d = [(j,i) for i,j in zip(min_x,min_y) if (j,i) in min_2d]
max_x,max_y = scipy.signal.argrelextrema(image, np.greater, order = n_order, axis = 1)
new_max_2d = [(j,i) for i,j in zip(max_x,max_y) if (j,i) in max_2d]
我一直在尝试使其适应3D空间，但这只是一团糟，我觉得必须有一种更简单的方法来做到这一点？有想法吗？除了您知道的此scipy功能之外，还有其他资源吗？
谢谢