全部水下图像增强算法python

以下是一些常见的水下图像增强算法的Python实现：

1. 对数拉伸算法

import numpy as np
import cv2

def log_stretch(img):
    # 对数拉伸算法
    c = 255 / np.log(1 + np.max(img))
    log_image = c * (np.log(img + 1))
    log_image = np.array(log_image, dtype=np.uint8)
    return log_image

2. 直方图均衡化算法

import numpy as np
import cv2

def hist_equalization(img):
    # 直方图均衡化算法
    hist, bins = np.histogram(img.flatten(), 256, [0, 256])
    cdf = hist.cumsum()
    cdf_normalized = cdf * hist.max() / cdf.max()
    cdf_m = np.ma.masked_equal(cdf, 0)
    cdf_m = (cdf_m - cdf_m.min()) * 255 / (cdf_m.max() - cdf_m.min())
    cdf = np.ma.filled(cdf_m, 0).astype('uint8')
    img2 = cdf[img]
    return img2

3. 多尺度Retinex算法

import numpy as np
import cv2

def multi_scale_retinex(img, sigma_list):
    # 多尺度Retinex算法
    img = np.array(img, dtype=np.float64)
    retinex = np.zeros_like(img)
    for sigma in sigma_list:
        img_blur = cv2.GaussianBlur(img, (0, 0), sigma)
        img_blur = np.array(img_blur, dtype=np.float64)
        retinex += np.log10(img) - np.log10(img_blur)
    retinex = retinex / len(sigma_list)
    retinex = np.array((retinex - np.min(retinex)) / (np.max(retinex) - np.min(retinex)) * 255)
    return np.array(retinex, dtype=np.uint8)

4. 双边滤波算法

import numpy as np
import cv2

def bilateral_filter(img):
    # 双边滤波算法
    img = cv2.bilateralFilter(img, 9, 75, 75)
    return img

5. 暗通道先验算法

import numpy as np
import cv2

def dark_channel_prior(img, w=15):
    # 暗通道先验算法
    img = np.array(img, dtype=np.float64) / 255.0
    m, n = img.shape[:2]
    padded = np.pad(img, ((w // 2, w // 2), (w // 2, w // 2), (0, 0)), mode='edge')
    dark_channel = np.zeros((m, n))
    for i, j in np.ndindex(dark_channel.shape):
        dark_channel[i, j] = np.min(padded[i:i + w, j:j + w, :])
    return dark_channel

def get_atmosphere(img, dark_channel):
    # 获取大气光照
    height, width, _ = img.shape
    flat_img = img.reshape(height * width, 3)
    flat_dark = dark_channel.ravel()
    search_idx = (-flat_dark).argsort()[:int(height * width * 0.01)]
    candidates = flat_img[search_idx]
    atmosphere = np.mean(candidates, axis=0)
    return atmosphere

def get_transmission(img, atmosphere, omega=0.95, w=15):
    # 获取透射率
    img = np.array(img, dtype=np.float64) / 255.0
    transmission = np.zeros_like(img)
    for i in range(3):
        transmission[:, :, i] = np.divide(img[:, :, i], atmosphere[i])
    transmission = 1 - omega * dark_channel_prior(transmission, w)
    return transmission

def dehaze(img, t, atm):
    # 去雾算法
    img = np.array(img, dtype=np.float64) / 255.0
    t = cv2.max(t, 0.1)
    dehazed = np.zeros_like(img)
    for i in range(3):
        dehazed[:, :, i] = (img[:, :, i] - atm[i]) / t + atm[i]
    dehazed = np.array(dehazed * 255, dtype=np.uint8)
    return dehazed