自动曝光修复算法 附完整C代码

众所周知，

图像方面的3A算法有：

AF自动对焦(Automatic Focus)
自动对焦即调节摄像头焦距自动得到清晰的图像的过程

AE自动曝光(Automatic Exposure)
自动曝光的是为了使感光器件获得合适的曝光量

AW自动白平衡(Automatic White Balance)
白平衡的本质是使白色物体在任何光源下都显示白色

 

前面的文章也有提及过，在刚开始做图像算法的时候，我是先攻克的自动白平衡算法。

后来攻克自动曝光的时候，傻啦吧唧的，踩了不少坑。

我相信一定不止我一个，一开始的时候抱着对图像均衡化，

软磨硬泡，想要做出兼顾自动曝光和自动白平衡的算法。

可惜，图像均衡化去做白平衡或者自动曝光，这条路是错的。

严格意义上来说，图像均衡化是拉伸曲线，这种做法有个弊端。

它没有考虑到图像的空间信息，也就是局部信息。

当然如果是处理音频之类的算法，肯定要考虑时间信息，因为数据是时序性为主的。

而图像，明显是空间信息为主的。

所以从理论上来说，用拉伸曲线这种不具备空间信息的操作，来做空间信息处理的事情，是不科学的。

我记得这博客刚开始写的时候，好多网友问我，为什么你要写那么多图像模糊算法，

图像模糊算法好像很鸡肋啊，没什么用的吧。

这就大错特错了，因为模糊算法是图像算法中，典型的包含空间信息的全局算法。

也就是说，如果要玩好图像算法，玩好模糊算法就是标配。

 

本次分享的算法为《Local Color Correction using Non-Linear Masking》，是ImageShop博主，

彭兄发出来的，安利一下他的博客https://www.cnblogs.com/imageshop 。

这个文章里的算法比较简单，

主要是通过图像模糊获取局域权重信息，然后映射回图片上。

matlab代码如下:

% Read the image
A=imread('input.jpg');

% Seperate the Channels
R=A(:,:,1);
G=A(:,:,2);
B=A(:,:,3);

% Calculate Intensity Component
I=(R+G+B)/3;

% Invert the image
I_inverted=255-I;

% Apply Average Filter to obtain the Mask Image
h_average=fspecial('average',15);
M=imfilter(I_inverted,h_average);

% Color Correction for R channel
R_new=zeros(size(R));
[c_y, c_x,~] = size(R);
for j = 1:c_x
        for i = 1:c_y
            p=double(R(i,j));
            q=double(M(i,j));
            R_new(i,j,:)=int8(255*((p/255)^(2^((128-q)/128))));
        end
end

% Color Correction for G channel
G_new=zeros(size(G));
[c_y, c_x,~] = size(G);
for j = 1:c_x
        for i = 1:c_y
            p=double(G(i,j));
            q=double(M(i,j));
            G_new(i,j,:)=int8(255*((p/255)^(2^((128-q)/128))));
        end
end

% Color Correction for B channel
B_new=zeros(size(B));
[c_y, c_x,~] = size(B);
for j = 1:c_x
        for i = 1:c_y
            p=double(B(i,j));
            q=double(M(i,j));
            B_new(i,j,:)=int8(255*((p/255)^(2^((128-q)/128))));
        end
end

% Output Image
O=zeros(size(A));
O(:,:,1)=R_new;
O(:,:,2)=G_new;
O(:,:,3)=B_new;

% Convert the double output image to uint8
O=uint8(O);

% Plot the images
subplot(1,3,1), imshow(A), title('Original Image');
subplot(1,3,2), imshow(M), title('Mask');
subplot(1,3,3), imshow(O), title('Output Image');

算法步骤很清晰，就不展开了。

有兴趣的同学，品读下论文吧。

论文链接直达

这个算法其实只是简单采用局部信息进行曝光调节，

但是并不能很好的适配很多图片情景。

需要进行二次改造，

例如：  白平衡，纹理处理更加自然诸如此类，之后就能更加美美哒。

师傅领进门，修行在个人。

改进的思路和方法就不展开一一细说了，

有兴趣的同学，可以考虑进一步改进。

效果图如下：

 

主要的算法函数实现如下：

void LocalColorCorrection(unsigned char *Input, unsigned char *Output, int Width, int Height, int Channels) {
    unsigned char *Mask = (unsigned char *) malloc(Width * Height * sizeof(unsigned char));
    if (Mask == NULL)
        return;
    unsigned char LocalLut[256 * 256];
    for (int mask = 0; mask < 256; ++mask) {
        unsigned char *pLocalLut = LocalLut + (mask << 8);
        for (int pix = 0; pix < 256; ++pix) {
            pLocalLut[pix] = ClampToByte(255.0f * powf(pix / 255.0f, powf(2.0f, (128.0f - mask) / 128.0f)));
        }
    }
    InvertGrayscale(Input, Output, Width, Height, Channels);
    int Radius = (MAX(Width, Height) / 512) + 1;
    BoxBlurGrayscale(Output, Mask, Width, Height, Radius);
    for (int Y = 0; Y < Height; Y++) {
        unsigned char *pOutput = Output + (Y * Width * Channels);
        unsigned char *pInput = Input + (Y * Width * Channels);
        unsigned char *pMask = Mask + (Y * Width);
        for (int X = 0; X < Width; X++) {
            unsigned char *pLocalLut = LocalLut + (pMask[X] << 8);
            for (int C = 0; C < Channels; C++) {
                pOutput[C] = pLocalLut[pInput[C]];
            }
            pOutput += Channels;
            pInput += Channels;
        }
    }
    free(Mask);
}

做了一些算法性能上的优化，720P,1080P下实时没半点问题。

至于进一步优化性能和效果，就留待下回分解，

当然有没有下回，得看心情。

附完整C代码：

/**
*implmentation of Local Color Correction using Non-Linear Masking published by Nathan Moroney Hewlett-Packard Laboratories, Palo Alto, California.
 **/
#include "browse.h"

#define USE_SHELL_OPEN

#define STB_IMAGE_STATIC
#define STB_IMAGE_IMPLEMENTATION

#include "stb_image.h"
/* ref:https://github.com/nothings/stb/blob/master/stb_image.h */
#define TJE_IMPLEMENTATION

#include "tiny_jpeg.h"
/* ref:https://github.com/serge-rgb/TinyJPEG/blob/master/tiny_jpeg.h */
#include <math.h>
#include <stdbool.h>
#include <stdio.h>
#include "timing.h"
#include <stdint.h>
#include <assert.h>

#ifndef _MAX_DRIVE
#define _MAX_DRIVE 3
#endif
#ifndef _MAX_FNAME
#define _MAX_FNAME 256
#endif
#ifndef _MAX_EXT
#define _MAX_EXT 256
#endif
#ifndef _MAX_DIR
#define _MAX_DIR 256
#endif
#ifndef MIN
#define MIN(a, b)    ( (a) > (b) ? (b) : (a) )
#endif
#ifndef MAX
#define MAX(a, b) (((a) > (b)) ? (a) : (b))
#endif
char saveFile[1024];

unsigned char *loadImage(const char *filename, int *Width, int *Height, int *Channels) {
    return (stbi_load(filename, Width, Height, Channels, 0));
}


void saveImage(const char *filename, int Width, int Height, int Channels, unsigned char *Output) {
    memcpy(saveFile + strlen(saveFile), filename, strlen(filename));
    *(saveFile + strlen(saveFile) + 1) = 0;

    if (!tje_encode_to_file(saveFile, Width, Height, Channels, true, Output)) {
        fprintf(stderr, "save JPEG fail.\n");
        return;
    }
#ifdef USE_SHELL_OPEN
    browse(saveFile);
#endif
}


void splitpath(const char *path, char *drv, char *dir, char *name, char *ext) {
    const char *end;
    const char *p;
    const char *s;
    if (path[0] && path[1] == ':') {
        if (drv) {
            *drv++ = *path++;
            *drv++ = *path++;
            *drv = '\0';
        }
    } else if (drv)
        *drv = '\0';
    for (end = path; *end && *end != ':';)
        end++;
    for (p = end; p > path && *--p != '\\' && *p != '/';)
        if (*p == '.') {
            end = p;
            break;
        }
    if (ext)
        for (s = end; (*ext = *s++);)
            ext++;
    for (p = end; p > path;)
        if (*--p == '\\' || *p == '/') {
            p++;
            break;
        }
    if (name) {
        for (s = p; s < end;)
            *name++ = *s++;
        *name = '\0';
    }
    if (dir) {
        for (s = path; s < p;)
            *dir++ = *s++;
        *dir = '\0';
    }
}

void getCurrentFilePath(const char *filePath, char *saveFile) {
    char drive[_MAX_DRIVE];
    char dir[_MAX_DIR];
    char fname[_MAX_FNAME];
    char ext[_MAX_EXT];
    splitpath(filePath, drive, dir, fname, ext);
    size_t n = strlen(filePath);
    memcpy(saveFile, filePath, n);
    char *cur_saveFile = saveFile + (n - strlen(ext));
    cur_saveFile[0] = '_';
    cur_saveFile[1] = 0;
}

int GetMirrorPos(int Length, int Pos) {
    if (Pos < 0)
        return -Pos;
    else if (Pos >= Length)
        return Length + Length - Pos - 2;
    else
        return Pos;
}

unsigned char ClampToByte(int Value) {
    if (Value < 0)
        return 0;
    else if (Value > 255)
        return 255;
    else
        return (unsigned char) Value;
}

void FillLeftAndRight_Mirror(int *Array, int Length, int Radius) {
    for (int X = 0; X < Radius; X++) {
        Array[X] = Array[Radius + Radius - X];
        Array[Radius + Length + X] = Array[Radius + Length - X - 2];
    }
}

int SumOfArray(const int *Array, int Length) {
    int Sum = 0;
    for (int X = 0; X < Length; X++) {
        Sum += Array[X];
    }
    return Sum;
}

void BoxBlurGrayscale(unsigned char *input, unsigned char *output, int Width, int Height, int Radius) {
    if ((input == NULL) || (output == NULL)) return;
    if ((Width <= 0) || (Height <= 0) || (Radius <= 0)) return;
    if (Radius < 1) return;
    Radius = MIN(MIN(Radius, Width - 1), Height - 1);
    int SampleAmount = (2 * Radius + 1) * (2 * Radius + 1);
    float Inv = 1.0f / SampleAmount;

    int *ColValue = (int *) malloc((Width + Radius + Radius) * sizeof(int));
    int *ColOffset = (int *) malloc((Height + Radius + Radius) * sizeof(int));
    if ((ColValue == NULL) || (ColOffset == NULL)) {
        if (ColValue != NULL) free(ColValue);
        if (ColOffset != NULL) free(ColOffset);
        return;
    }
    for (int Y = 0; Y < Height + Radius + Radius; Y++)
        ColOffset[Y] = GetMirrorPos(Height, Y - Radius);
    {
        for (int Y = 0; Y < Height; Y++) {
            unsigned char *scanLineOut = output + Y * Width;
            if (Y == 0) {
                memset(ColValue + Radius, 0, Width * sizeof(int));
                for (int Z = -Radius; Z <= Radius; Z++) {
                    unsigned char *scanLineIn = input + ColOffset[Z + Radius] * Width;
                    for (int X = 0; X < Width; X++) {
                        ColValue[X + Radius] += scanLineIn[X];
                    }
                }
            } else {
                unsigned char *RowMoveOut = input + ColOffset[Y - 1] * Width;
                unsigned char *RowMoveIn = input + ColOffset[Y + Radius + Radius] * Width;
                for (int X = 0; X < Width; X++) {
                    ColValue[X + Radius] -=
                            RowMoveOut[X] - RowMoveIn[X];
                }
            }
            FillLeftAndRight_Mirror(ColValue, Width, Radius);
            int LastSum = SumOfArray(ColValue, Radius * 2 + 1);
            scanLineOut[0] = ClampToByte((int) (LastSum * Inv));
            for (int X = 0 + 1; X < Width; X++) {
                int NewSum = LastSum - ColValue[X - 1] + ColValue[X + Radius + Radius];
                scanLineOut[X] = ClampToByte((int) (NewSum * Inv));
                LastSum = NewSum;
            }
        }
    }
    free(ColValue);
    free(ColOffset);
}

void InvertGrayscale(unsigned char *Input, unsigned char *Output, int Width, int Height, int Channels) {
    if (Channels == 1) {
        for (unsigned int Y = 0; Y < Height; Y++) {
            unsigned char *pOutput = Output + (Y * Width);
            unsigned char *pInput = Input + (Y * Width);
            for (unsigned int X = 0; X < Width; X++) {
                pOutput[X] = (unsigned char) (255 - pInput[X]);
            }
        }
    } else {
        for (unsigned int Y = 0; Y < Height; Y++) {
            unsigned char *pOutput = Output + (Y * Width);
            unsigned char *pInput = Input + (Y * Width * Channels);
            for (unsigned int X = 0; X < Width; X++) {
                pOutput[X] = (unsigned char) (255 - ClampToByte(
                        (21842 * pInput[0] + 21842 * pInput[1] + 21842 * pInput[2]) >> 16));
                pInput += Channels;
            }
        }
    }
}

void LocalColorCorrection(unsigned char *Input, unsigned char *Output, int Width, int Height, int Channels) {
    unsigned char *Mask = (unsigned char *) malloc(Width * Height * sizeof(unsigned char));
    if (Mask == NULL)
        return;
    unsigned char LocalLut[256 * 256];
    for (int mask = 0; mask < 256; ++mask) {
        unsigned char *pLocalLut = LocalLut + (mask << 8);
        for (int pix = 0; pix < 256; ++pix) {
            pLocalLut[pix] = ClampToByte(255.0f * powf(pix / 255.0f, powf(2.0f, (128.0f - mask) / 128.0f)));
        }
    }
    InvertGrayscale(Input, Output, Width, Height, Channels);
    int Radius = (MAX(Width, Height) / 512) + 1;
    BoxBlurGrayscale(Output, Mask, Width, Height, Radius);
    for (int Y = 0; Y < Height; Y++) {
        unsigned char *pOutput = Output + (Y * Width * Channels);
        unsigned char *pInput = Input + (Y * Width * Channels);
        unsigned char *pMask = Mask + (Y * Width);
        for (int X = 0; X < Width; X++) {
            unsigned char *pLocalLut = LocalLut + (pMask[X] << 8);
            for (int C = 0; C < Channels; C++) {
                pOutput[C] = pLocalLut[pInput[C]];
            }
            pOutput += Channels;
            pInput += Channels;
        }
    }
    free(Mask);
}

int main(int argc, char **argv) {
    printf("Local Color Correction demo\n ");
    printf("blog:http://cpuimage.cnblogs.com/ \n ");

    if (argc < 2) {
        printf("usage: %s   image \n ", argv[0]);
        printf("eg: %s   d:\\image.jpg \n ", argv[0]);

        return (0);
    }
    char *szfile = argv[1];

    getCurrentFilePath(szfile, saveFile);

    int Width = 0;
    int Height = 0;
    int Channels = 0;
    unsigned char *inputImage = NULL;

    double startTime = now();
    inputImage = loadImage(szfile, &Width, &Height, &Channels);

    double nLoadTime = calcElapsed(startTime, now());
    printf("load time: %d ms.\n ", (int) (nLoadTime * 1000));
    if ((Channels != 0) && (Width != 0) && (Height != 0)) {
        unsigned char *outputImg = (unsigned char *) stbi__malloc(Width * Channels * Height * sizeof(unsigned char));
        if (inputImage) {
            memcpy(outputImg, inputImage, (size_t) (Width * Channels * Height));
        } else {
            printf("load: %s fail!\n ", szfile);
        }
        startTime = now();
        LocalColorCorrection(inputImage, outputImg, Width, Height, Channels);
        double nProcessTime = calcElapsed(startTime, now());

        printf("process time: %d ms.\n ", (int) (nProcessTime * 1000));

        startTime = now();

        saveImage("done.jpg", Width, Height, Channels, outputImg);
        double nSaveTime = calcElapsed(startTime, now());

        printf("save time: %d ms.\n ", (int) (nSaveTime * 1000));

        if (outputImg) {
            stbi_image_free(outputImg);
        }

        if (inputImage) {
            stbi_image_free(inputImage);
        }
    } else {
        printf("load: %s fail!\n", szfile);
    }

    getchar();
    printf("press any key to exit. \n");

    return (EXIT_SUCCESS);
}

项目地址:https://github.com/cpuimage/LocalColorCorrection

再来一个效果前后对比：

 

以上，权当抛砖引玉。

若有其他相关问题或者需求也可以邮件联系俺探讨。

邮箱地址是: 
gaozhihan@vip.qq.com

转载于:https://www.cnblogs.com/cpuimage/p/9125646.html