[图像处理]YUV图像处理入门5

12 yuv420转换为rgb(opencv mat)

yuv格式具有亮度信息和色彩信息分离的特点，但大多数图像处理操作都是基于RGB格式，而且自己造轮子工作量太大。因此通常都会将yuv转换为rgb，再用opencv等视觉库进行图像处理。

yuv转换为rgb有多种方法，比如公式法。但是推荐使用第三方库进行转换，比如ffmpeg，libyuv，opencv。其中ffmpeg是专门的视频音频处理软件，libyuv是谷歌开发的专门用于yuv基本图像处理（如旋转，缩放，格式转换）的视频库，libyuv主要用于android端。

ffmpeg，libyuv，opencv都是开源的。可以在网上查找资料。

本文简单介绍ffmpeg和libyuv的安装，opencv的安装教程很多就不介绍了。具体见文章：

ffmpeg和libyuv的安装：

下载最新的ffmpeg的dev版和share版，ffmpeg严格区分x64和x86。下载网站为：

​​http://ffmpeg.zeranoe.com/builds/​​

Libyuv需要编译源文件，源文件地址：

​​https://chromium.googlesource.com/libyuv/libyuv/​​

​​https://github.com/seungrye/libyuv​​。

编译步骤见：

获得源文件后先建立vs工程，然后将ffmpegdev版本文件夹中的include和lib整个目录复制到vs工程目录下。如图所示：

对于libyuv的libyuv文件和lib文件，将其分别复制到vs工程目录下的include目录和lib目录。如图所示：

通常include中包含的是所调用库头文件，lib包含的是静态链接库，当然ffmpeg需要将其动态链接库复制到vs工程目录下，即将ffmpeg，share版本文件夹中bin目录下对应的所有dll复制到项目路径下如图所示：：

Dll和lib是windows系统下的动态链接库和静态链接库，linux系统下的静态链接库以.a结尾，linux系统下的动态链接库以.so或.so.y结尾。具体可以见文章：

对于ffmpeg，libyuv在linux系统下的编译使用，通过编译下载相关源代码，通过cmake或者make命令进行项目构建。推荐使用cmake软件，cmake非常有用，应有十分广泛。入门教程见：

在windows平台下，通过vs就能够减少大量工作。vs平台链接ffmpeg和libyuv的头文件和lib文件，先在项目工程属性>C/C++>常规>附加包含目录，添加include目录，但是ffmpeg有许多错误，vs通常会开启SDL检查后，某些警告会成为错误。所以将sdl检查置为否。如下图所示：

接着在在项目工程属性>链接器>常规>附加库目录下，添加lib文件夹，如下图所示：

最后如果使用ffmpeg和libyuv库，需要添加头文件完成整个工作的配置。代码如下：

extern "C"

{

#include "include\libavcodec\avcodec.h"

#include "include\libavformat\avformat.h"

#include "include\libavutil\channel_layout.h"

#include "include\libavutil\common.h"

#include "include\libavutil\imgutils.h"

#include "include\libswscale\swscale.h"

#include "include\libavutil\imgutils.h"

#include "include\libavutil\opt.h"

#include "include\libavutil\mathematics.h"

#include "include\libavutil\samplefmt.h"

//libyuv

#include "include\libyuv\libyuv.h"

};

#pragma comment(lib, "avcodec.lib")

#pragma comment(lib, "avformat.lib")

#pragma comment(lib, "avdevice.lib")

#pragma comment(lib, "avfilter.lib")

#pragma comment(lib, "avutil.lib")

#pragma comment(lib, "postproc.lib")

#pragma comment(lib, "swresample.lib")

#pragma comment(lib, "swscale.lib")

//libyuv

#pragma comment(lib, "yuv.lib")

yuv420转rgb

接下来通过ffmpeg，libyuv，opencv实现yuv420转rgb，并进行性能分析。函数的代码如下所示：

/**
 * @file 12 yuv_transform.cpp
 * @author luohen
 * @brief YUV image transform to opencv rgb image
 * @date 2018-12-11
 * 
 */

#include "pch.h"
#include <iostream>
#include <opencv2/opencv.hpp>
#include <time.h>

extern "C"
{
#include "include\libavcodec\avcodec.h"
#include "include\libavformat\avformat.h"
#include "include\libavutil\channel_layout.h"
#include "include\libavutil\common.h"
#include "include\libavutil\imgutils.h"
#include "include\libswscale\swscale.h"
#include "include\libavutil\imgutils.h"
#include "include\libavutil\opt.h"
#include "include\libavutil\mathematics.h"
#include "include\libavutil\samplefmt.h"
//libyuv
#include "include\libyuv\libyuv.h"
};
#pragma comment(lib, "avcodec.lib")
#pragma comment(lib, "avformat.lib")
#pragma comment(lib, "avdevice.lib")
#pragma comment(lib, "avfilter.lib")
#pragma comment(lib, "avutil.lib")
#pragma comment(lib, "postproc.lib")
#pragma comment(lib, "swresample.lib")
#pragma comment(lib, "swscale.lib")
//libyuv
#pragma comment(lib, "yuv.lib")

using namespace std;
using namespace cv;

/**
 * @brief
 *
 * @param pYUV    input yuv420 image
 * @param pBGR24  output bgr24 image
 * @param width   width of input yuv420p image
 * @param height  height of input yuv420p image
 * @return
 */
bool ffmpeg_yuv2bgr(unsigned char *pYUV, unsigned char *pBGR24, int width, int height)
{
  AVPicture pFrameYUV, pFrameBGR;

  avpicture_fill(&pFrameYUV, pYUV, AV_PIX_FMT_YUV420P, width, height);
  avpicture_fill(&pFrameBGR, pBGR24, AV_PIX_FMT_BGR24, width, height);

  struct SwsContext *imgCtx = NULL;
  //初始化函数
  //原图高，宽，图像类型；输出图高，宽，图像类型；算法种类；其他
  imgCtx = sws_getContext(width, height, AV_PIX_FMT_YUV420P, width, height, AV_PIX_FMT_BGR24, SWS_BILINEAR, 0, 0, 0);

  if (imgCtx != NULL)
  {
    //执行函数
    //函数返回值；输入图像指针数组，图像颜色通道数组；扫描起点；扫描行数；输出图像指针数组，图像颜色通道数组；
    sws_scale(imgCtx, pFrameYUV.data, pFrameYUV.linesize, 0, height, pFrameBGR.data, pFrameBGR.linesize);
    //end
    if (imgCtx)
    {
      sws_freeContext(imgCtx);
      imgCtx = NULL;
    }
    return true;
  }
  else
  {
    sws_freeContext(imgCtx);
    imgCtx = NULL;
    return false;
  }
}

/**
 * @brief   transform function of ffmpeg
 *
 * @param w   width of input yuv420p image
 * @param h   height of input yuv420p image
 * @param pic input yuv image
 * @return Mat  output rgb image(opencv mat)
 */
Mat yuv420_ffmpeg(int w, int h, unsigned char *pic)
{
  Mat bgrImg(h, w, CV_8UC3);
  unsigned char *pBGR24 = new unsigned char[w * h * 3];
  ffmpeg_yuv2bgr(pic, bgrImg.data, w, h);

  return bgrImg;
}

/**
 * @brief   transform function of libyuv
 *
 * @param w   width of input yuv420p image
 * @param h   height of input yuv420p image
 * @param pic input yuv image
 * @return Mat  output rgb image(opencv mat)
 */
Mat yuv420_libyuv(int w, int h, unsigned char *pic)
{
  int size_src = w * h * 3 / 2;
  int size_dest = w * h * 4;

  //BGRA, A:Alpha(transparency，透明度)
  Mat matI420 = cv::Mat(h, w, CV_8UC4);

  libyuv::I420ToARGB((const uint8 *)pic, w,
             (const uint8 *)(pic + w * h), w / 2,
             (const uint8 *)(pic + w * h * 5 / 4), w / 2,
             matI420.data, w * 4, w, h);
  //bgr
  Mat bgrImg;
  cvtColor(matI420, bgrImg, COLOR_BGRA2BGR);
  return bgrImg;
}

/**
 * @brief
 *
 * @param w
 * @param h
 * @param pic
 * @return Mat
 */
Mat yuv420_opencv(int w, int h, unsigned char *pic)
{
  //创建YUV mat
  cv::Mat yuvImg;
  yuvImg.create(h * 3 / 2, w, CV_8UC1);
  //数据保存为yuvImg.data
  memcpy(yuvImg.data, pic, w * h * 3 / 2 * sizeof(unsigned char));

  //转化为RGB图像
  cv::Mat bgrImg;
  cv::cvtColor(yuvImg, bgrImg, CV_YUV2BGR_I420);

  return bgrImg;
}

/**
 * @brief main
 *
 * @return int
 */
int main()
{
  clock_t start, end;
  double endtime;
  //Frequency of reading image
  int count_frame = 300;
  //视频路径
  char *url = (char *)"video/akiyo.yuv";
  int w = 352, h = 288;
  FILE *input_fp;
  if ((input_fp = fopen(url, "rb")) == NULL)
  {
    printf("%s open error!\n", url);
    return -1;
  }
  else
  {
    printf("%s open.\n", url);
  }

  unsigned char *pYuvBuf = new unsigned char[w * h * 3 / 2];

  fseek(input_fp, 0, SEEK_SET);
  //Timing starts
  start = clock();
  Mat ffmpeg_mat;
  for (int i = 0; i < count_frame; i++)
  {
    fread(pYuvBuf, sizeof(unsigned char), w * h * 3 / 2, input_fp);
    ffmpeg_mat = yuv420_ffmpeg(w, h, pYuvBuf);
  }

  //Timing end
  end = clock();
  endtime = (double)(end - start) / CLOCKS_PER_SEC;
  cout << "ffmpeg Total time:" << endtime << "s" << endl;
  cout << "ffmpeg Total time:" << endtime * 1000 << "ms" << endl;

  fseek(input_fp, 0, SEEK_SET);
  start = clock();
  Mat libyuv_mat;
  for (int i = 0; i < count_frame; i++)
  {
    fread(pYuvBuf, sizeof(unsigned char), w * h * 3 / 2, input_fp);
    libyuv_mat = yuv420_libyuv(w, h, pYuvBuf);
  }
  end = clock();
  endtime = (double)(end - start) / CLOCKS_PER_SEC;
  cout << "libyuv Total time:" << endtime << "s" << endl;     //s为单位
  cout << "libyuv Total time:" << endtime * 1000 << "ms" << endl; //ms为单位

  fseek(input_fp, 0, SEEK_SET);
  start = clock();
  Mat opencv_mat;
  for (int i = 0; i < count_frame; i++)
  {
    fread(pYuvBuf, sizeof(unsigned char), w * h * 3 / 2, input_fp);
    opencv_mat = yuv420_opencv(w, h, pYuvBuf);
  }
  end = clock();
  endtime = (double)(end - start) / CLOCKS_PER_SEC;
  cout << "opencv Total time:" << endtime << "s" << endl;
  cout << "opencv Total time:" << endtime * 1000 << "ms" << endl;

  system("pause");
  return 0;
}

调用函数为：

Mat yuv420_ffmpeg(int w, int h, unsigned char *pic);

Mat yuv420_libyuv(int w, int h, unsigned char *pic);

Mat yuv420_opencv(int w, int h, unsigned char *pic);

这段代码主要是分别用ffmpeg，libyuv，opencv实现yuv420转换为rgb，每种方法转换300张yuv420图像。对比三种方法转换所用时间，结果如下：

综合三种方法来说，ffmpeg速度最快，且ffmpeg最常用，因此推荐使用ffmpeg。如果仅仅对yuv图像进行处理或者android端，libyuv最为推荐。如果是安装ffmpeg或者libyuv较为麻烦，仅限于研究项目，建议使用opencv。