简单快速有效的低照度图像非线性增强方法

介绍的是一种比较实用并且去阴影效果很好的方法，选自2004年Tao的一篇论文，名称是《An Integrated Neighborhood Dependent Approach for Nonlinear Enhancement of Color Images 》，有兴趣的朋友仔细细读一下。论文篇幅很短，所以应该难度不大，而且效果极佳。

通过灰度图，动态范围压缩，高斯滤波结果，得到修复的灰度图。然后根据灰度图进行色彩恢复

Matlab

clc
clear
I=im2double(imread('2.png'));
I1=rgb2gray(I);
%其中的0.24和0.5都是可以调整的常数
In=(I1.^(0.24)+(1-I1).*0.5,I1.^2)/2;
%通过高斯核对灰度增强图像做卷积运算
sigma=3;
window_1 = double(uint8(3*sigma)*2+1);
G1=fspecial('gaussian',window_1,sigma);
Guass1=imfilter(I1,G1,'conv','replicate','same');
r1=Gauss1./I1;
R1=In.^r1;
sigma=20;
window_2=doule(uint8(3*sigma)*2+1);
G2=fspecial('gaussian',window_2,sigma);
Guass2=imfilter(I1,G2,'conv','replicate','same');
r2=Gauss2./I1;
R2=In.^r2;
sigma=240;
window_3 = double(uint8(3*sigma)*2+1);
G3=fspecial('gaussian',window_3,sigma);
Gauss3=imfilter(I1,G3,'conv','replicate','same');
r3=Gauss3./I1;
R3=In.^r3;
R=(R1+R2+R3)/3;
%R就是调整之后的I1，所以按照下面的式子来恢复
Rr=R.*(I(:,:,1)./I1);
Rg=R.*(I(:,:,2)./I2);
Rb=R.*(I(:,:,3)./I3);
rgb=cat(3,Rr,Rg,Rb);
imshow([I rgb])

Python

#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @Time : 2020/5/25 11:17
# @Author : Hran
# @Site :
# @File : work.py
# @Software: PyCharm
"""
论文：An Integrated Neighborhood Dependent Approach for Nonlinear Enhancement of Color Images
想复现一下之后，用暗通道去雾来代替
"""
import sys

import cv2
import numpy as np
e = np.exp(-10)

def Gaussion_kernel(kernel_size=3,sigma=1):
    m = n = (kernel_size - 1.) / 2.
    y, x = np.ogrid[-m:m + 1, -n:n + 1]
    h = np.exp(-(x * x + y * y) / (2. * sigma * sigma))
    h[h < np.finfo(h.dtype).eps * h.max()] = 0
    sumh = h.sum()
    if sumh != 0:
        h /= sumh
    return h


def gray_img(img):
    I1 = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    return I1 / 255 + e

def light_img(img):
    """
    对灰度图进行亮度恢复（动态范围压缩）
    :param img:灰度图
    :return:
    """
    np.power(img, 0.24)
    In = (np.power(img, 0.24) + (1 - img) * 0.5 + np.power(img, 2)) / 2
    return In

def contrast_enhancement(I1, In, sigma_list=[3,20,240]):
    """
    经过对比度增强获取R
    :param img:
    :return:
    """
    R = []
    for i in range(len(sigma_list)):
        sigma = sigma_list[i]
        windows_size = np.float32(np.uint8(sigma*3) * 2 + 1)
        if windows_size > 255:
            windows_size = 255
        print(windows_size)
        Gauss = cv2.GaussianBlur(I1, (0,0), sigma)
        # kernel = Gaussion_kernel(windows_size, sigma)
        # Gauss = cv2.filter2D(I1, -1, kernel)
        r = Gauss / I1
        R_temp = np.power(In, r)
        R.append(R_temp)

    R = R[0] + R[1] + R[2]
    return R / 3


def color_rebuild(img, I1, R):
    img = img / 255.
    b, g, r= cv2.split(img)
    r_1 = R * (r / I1)
    g_1 = R * (g / I1)
    b_1 = R * (b / I1)

    img = cv2.merge([b_1, g_1, r_1])
    return img


if __name__ == '__main__':
    img = cv2.imread('/Users/mac/Desktop/Test/6.jpeg')
    img = img.astype(np.float32)
    I1 = gray_img(img)
    a = np.where(I1 == 0)
    print(a)
    In = light_img(I1)
    R = contrast_enhancement(I1, In)
    result = color_rebuild(img, I1, R)
    result[result > 1] = 1
    result = result * 255
    result = result.astype(np.uint8)
    cv2.imshow('result', result)
    cv2.waitKey()