OpenCV图像噪点消除五大滤波方法

cv2.blur(src, ksize[, dst[, anchor[, borderType]]]) → dst

import cv2
import numpy as np

# 读取图像并添加高斯噪声
img = cv2.imread('test.jpg')
noise = np.random.normal(0, 30, img.shape).astype(np.uint8)
noisy_img = cv2.add(img, noise)

# 应用不同尺寸的均值滤波
blur_3x3 = cv2.blur(noisy_img, (3,3))  # 3×3小窗口
blur_7x7 = cv2.blur(noisy_img, (7,7))  # 7×7大窗口

# 显示结果
cv2.imshow('Noisy Image', noisy_img)
cv2.imshow('3x3 Blur', blur_3x3)
cv2.imshow('7x7 Blur', blur_7x7)
cv2.waitKey(0)

cv2.boxFilter(src, ddepth, ksize[, dst[, anchor[, normalize[, borderType]]]]) → dst

# 非归一化方框滤波（像素值可能溢出）
box_nonorm = cv2.boxFilter(noisy_img, -1, (3,3), normalize=False)
# 归一化方框滤波
box_norm = cv2.boxFilter(noisy_img, -1, (3,3), normalize=True)

# 比较结果
cv2.imshow('Non-normalized Box', box_nonorm)
cv2.imshow('Normalized Box', box_norm)

cv2.GaussianBlur(src, ksize, sigmaX[, dst[, sigmaY[, borderType]]]) → dst

# 不同标准差的高斯滤波
gauss_small = cv2.GaussianBlur(noisy_img, (5,5), 1)  # 小标准差
gauss_large = cv2.GaussianBlur(noisy_img, (5,5), 3)  # 大标准差

# 比较边缘保持效果
edge_img = cv2.Canny(img, 100, 200)
edge_gauss_small = cv2.Canny(gauss_small, 100, 200)
edge_gauss_large = cv2.Canny(gauss_large, 100, 200)

cv2.imshow('Original Edge', edge_img)
cv2.imshow('Small Sigma Edge', edge_gauss_small)
cv2.imshow('Large Sigma Edge', edge_gauss_large)

cv2.medianBlur(src, ksize[, dst]) → dst

# 添加椒盐噪声
def salt_pepper_noise(image, prob=0.05):
    output = np.zeros(image.shape, np.uint8)
    for i in range(image.shape[0]):
        for j in range(image.shape[1]):
            r = random.random()
            if r < prob/2:
                output[i,j] = 0    # 椒噪声
            elif r < prob:
                output[i,j] = 255  # 盐噪声
            else:
                output[i,j] = image[i,j]
    return output

sp_noisy = salt_pepper_noise(img)

# 中值滤波去噪
median_3 = cv2.medianBlur(sp_noisy, 3)
median_5 = cv2.medianBlur(sp_noisy, 5)

# 计算PSNR评估去噪效果
def psnr(img1, img2):
    mse = np.mean((img1 - img2) ** 2)
    return 10 * np.log10(255**2 / mse)

print(f"PSNR 3x3: {psnr(img, median_3):.2f} dB")
print(f"PSNR 5x5: {psnr(img, median_5):.2f} dB")

cv2.bilateralFilter(src, d, sigmaColor, sigmaSpace[, dst[, borderType]]) → dst

# 不同参数的双边滤波
bilateral_weak = cv2.bilateralFilter(noisy_img, 9, 25, 25)
bilateral_strong = cv2.bilateralFilter(noisy_img, 9, 75, 75)

# 边缘保持度比较
def edge_preserve_ratio(orig, filtered):
    orig_edge = cv2.Laplacian(orig, cv2.CV_64F).var()
    filt_edge = cv2.Laplacian(filtered, cv2.CV_64F).var()
    return filt_edge / orig_edge

print(f"Weak edge preserve: {edge_preserve_ratio(img, bilateral_weak):.2%}")
print(f"Strong edge preserve: {edge_preserve_ratio(img, bilateral_strong):.2%}")

import time

methods = {
    'Mean': lambda img: cv2.blur(img, (5,5)),
    'Gaussian': lambda img: cv2.GaussianBlur(img, (5,5), 0),
    'Median': lambda img: cv2.medianBlur(img, 5),
    'Bilateral': lambda img: cv2.bilateralFilter(img, 9, 75, 75)
}

results = {}
timings = {}

for name, func in methods.items():
    start = time.time()
    results[name] = func(noisy_img)
    timings[name] = time.time() - start
    print(f"{name}: {timings[name]:.4f}s")

# 可视化比较
plt.figure(figsize=(12,8))
for i, (name, img) in enumerate(results.items()):
    plt.subplot(2,2,i+1)
    plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
    plt.title(f"{name} ({timings[name]:.3f}s)")
    plt.axis('off')
plt.tight_layout()
plt.show()

# 多级滤波处理
def advanced_denoise(img):
    # 第一步：去除椒盐噪声
    temp = cv2.medianBlur(img, 3)
    # 第二步：平滑高斯噪声
    temp = cv2.bilateralFilter(temp, 5, 50, 50)
    # 第三步：锐化边缘
    kernel = np.array([[-1,-1,-1], [-1,9,-1], [-1,-1,-1]])
    return cv2.filter2D(temp, -1, kernel)

enhanced = advanced_denoise(noisy_img)