使用Gabor滤波器处理图像时异常的问题排查求助

我尝试复现某篇已发表的目标识别论文中的Gabor函数及参数，给图像应用Gabor滤波器，但输出结果不符合预期——仅在图像左上角出现部分白/灰色圆形Gabor渐变，其余区域均为黑色像素。我排查了很久没找到问题，求帮忙。

最初的循环代码是这样的：

output = gabor_output
if output > 0:
    output = output
else:
    output = 0

这段代码直接导致输出全黑，所以我改成了现在的代码来排查问题。但我希望最终能保留上面output = output的逻辑，而不是现在用的output = 1。

下面是我当前使用的完整代码：

# Gabor Function from S1 of "Robust Object Recognition" paper
import numpy as np
import matplotlib.pyplot as plt
from PIL import Image
import math


orientation_1 = 0.0
orientation_2 = 45.0
orientation_3 = 90.0
orientation_4 = 135.0

def compute_x0(x, y, orientation):
    x0 = (x * math.cos(orientation)) + (y * math.sin(orientation))
    return x0

def compute_y0(x, y, orientation):
    y0 = (-x * math.sin(orientation)) + (y * math.cos(orientation))
    return y0

def gabor_function(x, y, orientation, wavelength, width, aspect_ratio=0.3):
    x0 = compute_x0(x, y, orientation)
    y0 = compute_y0(x, y, orientation)
    term_1 = np.exp(-(x0**2 + aspect_ratio**2 * y0**2) / (2 * width**2))
    term_2 = math.cos(((2 * np.pi) / wavelength) * x0)
    output = term_1 * term_2
    return output


# Trying it out

# Upload Image, convert to grayscale, convert to np.array
image = Image.open('/Users/damondtrowbridge/Desktop/MATLAB/Face.png')
bw = image.convert('L')
bw_array = np.array(bw) / 255.0

# Crop to fit different images and adjust for pooling
og_size = np.shape(bw_array)
Xdim = (og_size[0]//4)*4
Ydim = (og_size[1]//4)*4
cropped = bw_array[0:Xdim, 0:Ydim]

# Set up output array

out_1 = np.zeros((Xdim-3, Ydim-3))

for x in range(3, Xdim - 3):
    for y in range(3, Ydim - 3):
        gabor_output = gabor_function(x, y, orientation_1, 3.5, 2.8, aspect_ratio=0.3)
        output = gabor_output * cropped[x, y]
        if output > 0:
            output = 1
        else:
            output = 0
        out_1[x, y] = output



plt.imshow(out_1[:, :], cmap='gray', aspect='auto', vmin=0, vmax=1)
plt.suptitle('Filtered Image')
plt.show()

内容的提问来源于stack exchange，提问作者Sydney Trowbridge