图像处理-RGB24转YUV420遇到的坑以及执行效率对比
感兴趣可以加QQ群85486140，大家一起交流相互学习下！
文章目录
- 一、色彩空间简介及一些参考文档
- 二、YUV和RGB转换公式
- 三、RGB和YUV420转换提前需要了解的知识
- 1.YUV420内存布局
- 2.RGB内存布局
- 3.转换方式
- 四、源代码执行效率对比
- 五、编译过程中遇到的问题
一、色彩空间简介及一些参考文档
不同的色彩空间，颜色的表现形式不同(见http://colorizer.org/ 这个网站中简单介绍了各个色彩空间的模型。)。色彩空间中的颜色是可以相互转化的。我们常用的颜色空间有YUV和RGB色彩空间。
-
色彩空间简介：http://colorizer.org/
-
参考文章:1：一种简单的YUV转RGB的优化算法（之前微信推送的一篇文章）
-
参考文章2：YUV <——> RGB 转换算法（此文章分析的非常全面，并有github源码仓库，优先推荐参考）
-
JPG转RGB格式的网站：https://convertio.co/zh/jpg-rgb/
二、YUV和RGB转换公式
如果是RGB24的话，则它们和YUV的对应公式如下所示：
- 1.RGB转YUV
Y = 0.299R + 0.587G + 0.114B
U = -0.147R - 0.289G + 0.436B
V = 0.615R - 0.515G - 0.100B
- 2.YUV转RGB
R = Y + 1.14V
G = Y - 0.39U - 0.58V
B = Y + 2.03U
上面可以可以看到有浮点运算，这会导致耗时很长。下面的优化也是基于浮点运算，来的。
三、RGB和YUV420转换提前需要了解的知识
1.YUV420内存布局
W
+--------------------+
|Y0Y1Y2Y3... |
|... |
| | H
| |
| |
| |
+--------------------+
|U0U1... |
|... | H/2
| |
+----------+
|V0V1... |
|... | H/2
| |
+----------+
w/2
2.RGB内存布局
虽然名字上叫RGB,但是排列的时候是按着BGR的顺序排列的，如下所示。
B G R
+-----------------------------------------------------------------------------------------------+
高字节 | B | B | B | B | B | B | B | B | G | G | G | G | G | G | G | G | R | R | R | R | R | R | R | R | 低字节
+-----------------------------------------------------------------------------------------------+
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
3.转换方式
转换的时候由于不是RGB24转YUV444(一对一)，而是转成YUV420.所以必然有些数据时要舍弃的。下面即将处理的图片分辨率是320 x 240的大小，所以RGB24的格式就会产生320 * 240 *3 byte的数据 即230400，下图也能看到后缀为rgb图片的大小。对应YUV420图片大小则为320 * 240 * 1.5=115200 Byte的大小。同样下面实际生成的图片也是这么大。
RGB的存放方式如下所示，例如8*6像素的RGB图像，其数据分布式这样存放的。
pixe0 pixe1 pixe2 pixe3 pixe4 pixe5 pixe6 pixe7
+-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------+
Line0 | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R |
+-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------+
Line1 | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R |
+-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------+
Line2 | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R |
+-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------+
Line3 | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R |
+-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------+
Line4 | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R |
+-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------+
Line5 | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R | B | G | R |
+-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------+
那么由于UV是每4个点对应一组UV,则选择哪个点转换为U，那个点转换为V。针对这样的block选择每个block的左上角(pix0)像素转换成U，左下角(pix2)的转换成V，其实这4个点的像素都可以用来转换UV，只是大家都习惯这样来转。
|pix0|pix1|
|pix2|pix3|
其它网友YUV420和YUV444相互转换的的好博客：https://blog.csdn.net/guyuealian/article/details/82454945
四、源代码执行效率对比
VS2015源码包可以在这里下载：https://download.csdn.net/download/armwind/11589508
由于原始的转换过程中包含了浮点和乘法运算，所以需要优化算法。下面例程中也包含了优化前后的对比。
- 1.源码
#include "stdafx.h"
#include "iostream"
#include
#include
using namespace::std;
static void RGBToYUV_v1(int R, int G, int B, int* Y, int* U, int* V) {
*Y = 0.299*R + 0.587*G + 0.114*B;
*U = -0.147*R - 0.289*G + 0.436*B;
*V = 0.615*R - 0.515*G - 0.100*B;
}
static void YUVToRGB_v1(int Y, int U, int V, int* R, int* G, int* B) {
*R = Y + 1.14*V;
*G = Y - 0.39*U - 0.58*V;
*B = Y + 2.03*U;
}
static void RGBToYUV(int Red, int Green, int Blue, int* Y, int* U, int* V)
{
*Y = ((Red << 6) + (Red << 3) + (Red << 2) + Red + (Green << 7) + (Green << 4) + (Green << 2) + (Green << 1) + (Blue << 4) + (Blue << 3) + (Blue << 2) + Blue) >> 8;
*U = (-((Red << 5) + (Red << 2) + (Red << 1)) - ((Green << 6) + (Green << 3) + (Green << 1)) + ((Blue << 6) + (Blue << 5) + (Blue << 4))) >> 8;
*V = ((Red << 7) + (Red << 4) + (Red << 3) + (Red << 2) + (Red << 1) - ((Green << 7) + (Green << 2)) - ((Blue << 4) + (Blue << 3) + (Blue << 1))) >> 8;
}
static void YUVToRGB(int Y, int U, int V, int* Red, int* Green, int* Blue)
{
*Red = ((Y << 8) + ((V << 8) + (V << 5) + (V << 2))) >> 8;
*Green = ((Y << 8) - ((U << 6) + (U << 5) + (U << 2)) - ((V << 7) + (V << 4) + (V << 2) + V)) >> 8;
*Blue = ((Y << 8) + (U << 9) + (U << 3)) >> 8;
}
/*********************只不包含浮点运算********************/
//如果使用原始的转换出来的话，会出现偏色问题，下面在原始的基础上又做了微调，可以正常显示。
#define RGB2Y(r,g,b) \
((unsigned char)((66 * r + 129 * g + 25 * b + 128) >> 8) + 16)
// (((r << 6) + (r << 3) + (r << 2) + r + (g << 7) + (g << 4) + (g << 2) + (g << 1) + (b << 4) + (b << 3) + (b << 2) + b) >> 8)
#define RGB2U(r,g,b) \
((unsigned char)((-38 * r - 74 * g + 112 * b + 128) >> 8) + 128)
// ((-((r << 5) + (r << 2) + (r << 1)) - ((g << 6) + (g << 3) + (g << 1)) + ((b << 6) + (b << 5) + (b << 4))) >> 8)
#define RGB2V(r,g,b) \
((unsigned char)((112 * r - 94 * g - 18 * b + 128) >> 8) + 128)
// (((r << 7) + (r << 4) + (r << 3) + (r << 2) + (r << 1) - ((g << 7) + (g << 2)) - ((b << 4) + (b << 3) + (b << 1))) >> 8)
/**********************不包含浮点和乘法运算*****************/
#define RGB2Y_SHIFT(r,g,b) \
((unsigned char)((((r << 6) + (r << 3) + (r << 2) + r + (g << 7) + (g << 4) + (g << 2) + (g << 1) + (b << 4) + (b << 3) + (b << 2) + b) + 128) >> 8) + 16)
#define RGB2U_SHIFT(r,g,b) \
((unsigned char)(((-((r << 5) + (r << 2) + (r << 1)) - ((g << 6) + (g << 3) + (g << 1)) + ((b << 6) + (b << 5) + (b << 4))) + 128) >> 8) + 128)
#define RGB2V_SHIFT(r,g,b) \
((unsigned char)(((r << 7) + (r << 4) + (r << 3) + (r << 2) + (r << 1) - ((g << 7) + (g << 2)) - ((b << 4) + (b << 3) + (b << 1)) + 128) >> 8) + 128)
/*************************包含浮点和乘法运算，没有优化**************/
#define RGB2Y_FLOAT(r,g,b) \
((unsigned char)(0.299*r + 0.587*g + 0.114*b))
#define RGB2U_FLOAT(r,g,b) \
((unsigned char)(-0.147*r - 0.289*g + 0.436*b))
#define RGB2V_FLOAT(r,g,b) \
((unsigned char)(0.615*r - 0.515*g - 0.100*b))
void rgb2yuv420_multip(unsigned char *rgb_buf, unsigned char *yuv_buf, int width, int heigh) {
if (!rgb_buf || !yuv_buf || !width || !heigh) {
printf("invalid param\n");
return;
}
unsigned char *y = yuv_buf;
unsigned char *u = yuv_buf + width*heigh;
unsigned char *v = u + width*heigh/4;
for (int r = 0; r < heigh; r++ ) {
for (int c = 0; c < width; c++) {
int index = r*width + c;
*y++ = RGB2Y(rgb_buf[index * 3 + 2], rgb_buf[index * 3 + 1], rgb_buf[index * 3]);
if((r%2==0)&&(c%2 ==0)) //偶数行，偶数列获取U
*u++ = RGB2U(rgb_buf[index * 3 + 2], rgb_buf[index * 3 + 1], rgb_buf[index * 3]);
else if(c%2 == 0) //奇数行，偶数列获取V
*v++ = RGB2V(rgb_buf[index * 3 + 2], rgb_buf[index * 3 + 1], rgb_buf[index * 3]);
}
}
}
void rgb2yuv420_shift(unsigned char *rgb_buf, unsigned char *yuv_buf, int width, int heigh) {
if (!rgb_buf || !yuv_buf || !width || !heigh) {
printf("invalid param\n");
return;
}
unsigned char *y = yuv_buf;
unsigned char *u = yuv_buf + width*heigh;
unsigned char *v = u + width*heigh/4;
for (int r = 0; r < heigh; r++ ) {
for (int c = 0; c < width; c++) {
int index = r*width + c;
*y++ = RGB2Y(rgb_buf[index * 3 + 2], rgb_buf[index * 3 + 1], rgb_buf[index * 3]);
if((r%2==0)&&(c%2 ==0)) //偶数行，偶数列获取U
*u++ = RGB2U(rgb_buf[index * 3 + 2], rgb_buf[index * 3 + 1], rgb_buf[index * 3]);
else if(c%2 == 0) //奇数行，偶数列获取V
*v++ = RGB2V(rgb_buf[index * 3 + 2], rgb_buf[index * 3 + 1], rgb_buf[index * 3]);
}
}
}
void rgb2yuv420_float(unsigned char *rgb_buf, unsigned char *yuv_buf, int width, int heigh) {
if (!rgb_buf || !yuv_buf || !width || !heigh) {
printf("invalid param\n");
return;
}
unsigned char *y = yuv_buf;
unsigned char *u = yuv_buf + width*heigh;
unsigned char *v = u + width*heigh/4;
for (int r = 0; r < heigh; r++ ) {
for (int c = 0; c < width; c++) {
int index = r*width + c;
*y++ = RGB2Y_FLOAT(rgb_buf[index * 3 + 2], rgb_buf[index * 3 + 1], rgb_buf[index * 3]);
if((r%2==0)&&(c%2 ==0)) //偶数行，偶数列获取U
*u++ = RGB2U_FLOAT(rgb_buf[index * 3 + 2], rgb_buf[index * 3 + 1], rgb_buf[index * 3]);
else if(c%2 == 0) //奇数行，偶数列获取V
*v++ = RGB2V_FLOAT(rgb_buf[index * 3 + 2], rgb_buf[index * 3 + 1], rgb_buf[index * 3]);
}
}
}
int main(int arc, const char ** argv)
{
int ret = 0;
FILE *pFile = NULL,*pFile_out = NULL;
int size = 0;
int width = 0, heigh = 0;
char path[] = "D:\\vs2015_project\\yuv22rgb\\girl320x240.yuv";
unsigned char * rgb_buf = NULL;
unsigned char *yuv_buf = nullptr;
cout << "arc:" << arc << endl;
if (arc == 1) {
cout << "please input parameters!" << endl;
return -1;
}
sscanf(argv[2],"%d", &width);
sscanf(argv[3],"%d", &heigh);
printf("file name:%s,width:%d,heigh:%d\n", argv[1], width, heigh);
//cout << "file name:" << argv[1] << endl;
pFile = fopen(argv[1], "rb");
if (NULL == pFile) {
cout << "open file failed!" << endl;
}
else {
fseek(pFile, 0, SEEK_END);
size = ftell(pFile);
fseek(pFile, 0, SEEK_SET);
cout << "file size is " << size << endl;
}
rgb_buf = (unsigned char *)malloc(size);
if (NULL == rgb_buf) {
cout << "malloc buffer failed, return" << endl;
fclose(pFile);
return -1;
}
int r_size = fread(rgb_buf, 1, size, pFile);
if (r_size != size) {
cout << "read error" << endl;
ret = -1;
goto release;
}
int yuv_size = width * heigh * 1.5;
yuv_buf = (unsigned char*)malloc(yuv_size);
if (NULL == yuv_buf)
goto release;
LARGE_INTEGER t1, t2,t3,t4;
LARGE_INTEGER f;
QueryPerformanceFrequency(&f); //得到CPU频率
QueryPerformanceCounter(&t1); //得到开始时间
rgb2yuv420_multip(rgb_buf, yuv_buf, width, heigh);
QueryPerformanceCounter(&t2); //得到优化了浮点运算的计算时间
double total1_time = (t2.QuadPart - t1.QuadPart) / (double)f.QuadPart; //单位秒
rgb2yuv420_shift(rgb_buf, yuv_buf, width, heigh);
QueryPerformanceCounter(&t3); //得到优化浮点和乘法运算之后结束时间
double total2_time = (t3.QuadPart - t2.QuadPart) / (double)f.QuadPart; //单位秒
rgb2yuv420_float(rgb_buf, yuv_buf, width, heigh);
QueryPerformanceCounter(&t4); //原始没有优化的图像
double total3_time_float = (t4.QuadPart - t3.QuadPart) / (double)f.QuadPart; //单位秒
printf("shift_time:%f s,multip_time:%f s,time_float:%f s\n", total2_time,total1_time,total3_time_float);
pFile_out = fopen(path, "wb+");
if (pFile_out == nullptr) {
printf("null fd\n");
goto release;
}
printf("yuv_size:%d\n", yuv_size);
int w_count = fwrite(yuv_buf, 1, yuv_size, pFile_out);
printf("write_count:%d\n", w_count);
fclose(pFile_out);
release:
if (rgb_buf) {
free(rgb_buf);
rgb_buf = nullptr;
}
if (yuv_buf) {
free(yuv_buf);
yuv_buf = nullptr;
}
fclose(pFile);
return 0;
}
运行结果如下所示：
D:\vs2015_project\yuv22rgb\yuv22rgb\Debug>yuv22rgb.exe …\girl320x240.rgb 320 240
arc:4
file name:…\girl320x240.rgb,width:320,heigh:240
file size is 230400
shift_time:0.000367 s,multip_time:0.000366 s,time_float:0.000593 s
yuv_size:115200
write_count:115200
从上面能够发现在同样的运行环境下，处理时间优化前后差异很大，其中优化移位操作，优化的时间不是很大，不知道跑在arm架构上时间会是怎么样的，后面会尝试一下。
| 优化项 | 时间(单位S) |
|---|---|
| 没有任何优化 | 0.000593 |
| 优化浮点计算 | 0.000367 |
| 优化浮点和乘法计算 | 0.000366 |
- 2.CPU硬件信息
- 3.图片对比,下图左边是原始的图片，右边是转换之后的YUV，可以发现转换之后的图片有点偏红，不过色调基本一致。
五、编译过程中遇到的问题
- 1.读取文件识别
打开文件时应该加上“rb"模式 - 2.写入文件有问题(写入文件的大小和预期不一致)
写入文件时应该使用”wb+" - 3.图像偏色
uv反色或者公式有问题