Retinex图像增强算法(SSR, MSR, MSRCR)详解及其OpenCV源码

转自：

https://blog.csdn.net/ajianyingxiaoqinghan/article/details/71435098

PYTHON:

https://www.cnblogs.com/wangyong/p/8665434.html

import cv2
import numpy as np
#单尺度Retinex
def singleScaleRetinex(img, sigma):
    #按照公式计算
    _temp=cv2.GaussianBlur(img,(0,0), sigma)
    gaussian=np.where(_temp==0, 0.001, _temp)
    img_ssr= np.log10(img+0.01) - np.log10(gaussian)
    #量化到0--255
    for i in range(img_ssr.shape[2]):
        img_ssr[:, :, i] = (img_ssr[:,:,i] - np.min(img_ssr[:, :, i])) / \
                (np.max(img_ssr[:,:,i]) - np.min(img_ssr[:, :, i]))*255
    img_ssr = np.uint8(np.minimum(np.maximum(img_ssr, 0), 255))
    return img_ssr
    
def singleScaleRetinexTemp(img, sigma):
    #按照公式计算
    _temp=cv2.GaussianBlur(img,(0,0), sigma)
    gaussian=np.where(_temp==0, 0.001, _temp)
    retinex = np.log10(img+0.01) - np.log10(gaussian)

    return retinex
    
#多尺度Retinex, sigma_list[15,80,250]
def multiScaleRetinex(img, sigma_list):
    retinex = np.zeros_like(img*1.0)
    for sigma in sigma_list:
        print("sigma:", sigma)
        retinex += singleScaleRetinexTemp(img, sigma)
    img_msr= retinex / len(sigma_list)
    for i in range(img_msr.shape[2]):
        img_msr[:, :, i] = (img_msr[:,:,i] - np.min(img_msr[:, :, i])) / \
                (np.max(img_msr[:,:,i]) - np.min(img_msr[:, :, i]))*255
    img_msr = np.uint8(np.minimum(np.maximum(img_msr, 0), 255))
    return img_msr
    
if __name__ == '__main__':
    imageSrc = cv2.imread("G:\\vc2010\\Opencv\\Test\\image\\test_MSRCR.jpg")
    cv2.imshow('src',imageSrc)

    dstsrc = singleScaleRetinex(imageSrc,300)
    cv2.imshow('ssr',dstsrc)
    
    sigma_list = [15, 80, 250]
    dstmsr = multiScaleRetinex(imageSrc, sigma_list)
    cv2.imshow('msr',dstmsr) 
    cv2.waitKey(0)