Python图像处理中带通滤波器的优化方案咨询

我正在开展一项Python图像处理任务，尝试通过傅里叶变换为图像应用带通滤波器，目标是将原始图像（）转换为预期输出图像（）。

以下是我已尝试的代码：

import cv2
import numpy as np
from matplotlib import pyplot as plt

image = cv2.imread('path-to-image', cv2.IMREAD_GRAYSCALE)

# Apply Fourier Transform
f_transform = np.fft.fft2(image)
f_transform_shifted = np.fft.fftshift(f_transform)

# Create a bandpass filter
rows, cols = image.shape
crow, ccol = rows // 2, cols // 2
inner_radius = 0.1 * crow
outer_radius = 0.8 * crow

# Create a meshgrid for the frequency coordinates
x = np.arange(-ccol, ccol)
y = np.arange(-crow, crow)
x, y = np.meshgrid(x, y)

# Create the bandpass filter
mask = ((x**2 + y**2 >= inner_radius**2) & (x**2 + y**2 <= outer_radius**2))

# Apply the mask to the shifted Fourier transform
f_transform_shifted_filtered = f_transform_shifted * mask

# Inverse Fourier Transform to get the image back
f_inverse_shifted = np.fft.ifftshift(f_transform_shifted_filtered)
image_filtered = np.fft.ifft2(f_inverse_shifted)
image_filtered = np.abs(image_filtered)

# Convert back to uint8 and normalize the values
image_filtered = np.uint8(image_filtered)
image_filtered = cv2.normalize(image_filtered, None, 0, 255, cv2.NORM_MINMAX)

当前遇到的问题是输出图像与预期不符，我推测问题源于带通滤波器的定义与应用逻辑，但不确定如何调整才能达成目标效果，恳请提供代码优化建议以接近预期输出。

核心问题出在滤波器的创建逻辑和后处理步骤上，以下是具体优化点及完整修改代码：

1. 修正频率网格坐标 #

原代码生成的网格可能因图像尺寸奇偶性出现匹配问题，改用基于图像尺寸的线性空间生成，确保网格与图像完全对齐：

x = np.linspace(-ccol, ccol-1, cols)
y = np.linspace(-crow, crow-1, rows)
x, y = np.meshgrid(x, y)

2. 使用平滑带通滤波器避免振铃效应 #

原硬边界滤波器会引入伪影，改用高斯平滑的带通滤波器，平衡频率保留与噪声抑制：

distance = np.sqrt(x**2 + y**2)
# 高斯带通=低通（过滤低频）*高通（过滤高频）
sigma_inner = inner_radius / 3
sigma_outer = outer_radius / 3
low_pass = 1 - np.exp(-(distance**2) / (2 * sigma_inner**2))
high_pass = np.exp(-(distance**2) / (2 * sigma_outer**2))
mask = low_pass * high_pass

3. 优化后处理顺序 #

先归一化再转uint8，避免精度丢失：

image_filtered = cv2.normalize(image_filtered, None, 0, 255, cv2.NORM_MINMAX)
image_filtered = np.uint8(image_filtered)

完整修改代码 #

import cv2
import numpy as np
from matplotlib import pyplot as plt

image = cv2.imread('path-to-image', cv2.IMREAD_GRAYSCALE)

# 傅里叶变换
f_transform = np.fft.fft2(image)
f_transform_shifted = np.fft.fftshift(f_transform)

rows, cols = image.shape
crow, ccol = rows // 2, cols // 2

# 微调半径参数（可根据效果调整）
inner_radius = 0.05 * crow
outer_radius = 0.6 * crow

# 生成匹配的频率网格
x = np.linspace(-ccol, ccol-1, cols)
y = np.linspace(-crow, crow-1, rows)
x, y = np.meshgrid(x, y)
distance = np.sqrt(x**2 + y**2)

# 创建高斯平滑带通滤波器
sigma_inner = inner_radius / 3
sigma_outer = outer_radius / 3
low_pass = 1 - np.exp(-(distance**2) / (2 * sigma_inner**2))
high_pass = np.exp(-(distance**2) / (2 * sigma_outer**2))
mask = low_pass * high_pass

# 应用滤波器并逆变换
f_transform_shifted_filtered = f_transform_shifted * mask
f_inverse_shifted = np.fft.ifftshift(f_transform_shifted_filtered)
image_filtered = np.fft.ifft2(f_inverse_shifted)
image_filtered = np.abs(image_filtered)

# 后处理
image_filtered = cv2.normalize(image_filtered, None, 0, 255, cv2.NORM_MINMAX)
image_filtered = np.uint8(image_filtered)

# 可选：对比显示结果
plt.subplot(121), plt.imshow(image, cmap='gray')
plt.title('Original'), plt.xticks([]), plt.yticks([])
plt.subplot(122), plt.imshow(image_filtered, cmap='gray')
plt.title('Filtered'), plt.xticks([]), plt.yticks([])
plt.show()

额外说明 #

• 半径参数需根据图像实际尺寸微调：若输出仍有噪声，缩小outer_radius；若丢失细节，增大inner_radius。
• 高斯平滑滤波器可有效消除硬边界带来的振铃效应，让输出更接近预期的平滑效果。

内容的提问来源于stack exchange，提问作者Estiven