C++ cpp rgb到yuv422的转换
我正在尝试将RGB/RGBA格式（可以更改）的图像（最初来自QImage）转换为YUV422格式。 我最初的意图是使用OpenCV来完成这项工作，但它无法将RGB/RGBA转换为422格式 我寻找替代方案，甚至考虑根据需要编写自己的转换，但速度不够快 我找了另一个图书馆使用，但它是旧的，不太相关 所以我的问题是，对于RGB->YUV422转换，我有什么好的选择？如果他们在GPU而不是CPU上执行转换，效果会更好C++ cpp rgb到yuv422的转换,c++,opencv,opencl,yuv,color-space,C++,Opencv,Opencl,Yuv,Color Space,我正在尝试将RGB/RGBA格式（可以更改）的图像（最初来自QImage）转换为YUV422格式。 我最初的意图是使用OpenCV来完成这项工作，但它无法将RGB/RGBA转换为422格式 我寻找替代方案，甚至考虑根据需要编写自己的转换，但速度不够快 我找了另一个图书馆使用，但它是旧的，不太相关 所以我的问题是，对于RGB->YUV422转换，我有什么好的选择？如果他们在GPU而不是CPU上执行转换，效果会更好 在这方面，他们建议使用，并且OP似乎达到了预期的效果（实时转换许多PAL流）。我使用
在这方面，他们建议使用，并且OP似乎达到了预期的效果（实时转换许多PAL流）。我使用OpenCL解决了我的问题，如下所示： 我将转换更改为格式\u ARGB32\u预乘为YUV422，但它可以轻松更改为任何格式 openclwrapper.h:
class OpenClWrapper
{
public:
OpenClWrapper(size_t width, size_t height);
~OpenClWrapper();
void RGB2YUV422(unsigned int * yuvImg, unsigned char * rgbImg);
private:
std::vector<cl::Platform> m_all_platforms;
std::vector<cl::Device> m_all_devices;
cl::Platform m_default_platform;
cl::Device m_default_device;
cl::Context m_context;
cl::Program::Sources m_sources;
cl::Program m_program;
cl::CommandQueue m_queue;
cl::Buffer m_buffer_yuv;
cl::Buffer m_buffer_rgb;
std::string m_kernel_code;
size_t m_width;
size_t m_height;
};
#include "openclwrapper.h"
#include <iostream>
#include <sstream>
OpenClWrapper::OpenClWrapper(size_t width, size_t height) :
m_height(height),
m_width(width)
{
//get all platforms (drivers)
cl::Platform::get(&m_all_platforms);
if(m_all_platforms.size()==0){
std::cout<<" No platforms found. Check OpenCL installation!\n";
exit(1);
}
m_default_platform=m_all_platforms[0];
//get default device of the default platform
m_default_platform.getDevices(CL_DEVICE_TYPE_ALL, &m_all_devices);
if(m_all_devices.size()==0){
std::cout<<" No devices found. Check OpenCL installation!\n";
exit(1);
}
m_default_device=m_all_devices[0];
m_context = *(new cl::Context({m_default_device}));
std::ostringstream oss;
oss <<
" void kernel RGB2YUV422(global const unsigned char rgbImg[" << m_height << "][" << m_width << "*4], global unsigned int yuvImg[" << m_height << "][" << m_width << "/2]){ \n"
" int x_idx = get_global_id(0); \n"
" int y_idx = get_global_id(1)*8; \n"
" int alpha1 = rgbImg[x_idx][y_idx+3]; \n"
" int alpha2 = rgbImg[x_idx][y_idx+7]; \n"
" unsigned char R1 = rgbImg[x_idx][y_idx+2] * (255 / alpha1); \n"
" unsigned char G1 = rgbImg[x_idx][y_idx+1] * (255 / alpha1); \n"
" unsigned char B1 = rgbImg[x_idx][y_idx] * (255 / alpha1); \n"
" unsigned char R2 = rgbImg[x_idx][y_idx+6] * (255 / alpha2); \n"
" unsigned char G2 = rgbImg[x_idx][y_idx+5] * (255 / alpha2); \n"
" unsigned char B2 = rgbImg[x_idx][y_idx+4] * (255 / alpha2); \n"
" unsigned char Y1 = (unsigned char)(0.299000*R1 + 0.587000*G1 + 0.114000*B1); \n"
" unsigned char Y2 = (unsigned char)(0.299000*R2 + 0.587000*G2 + 0.114000*B2); \n"
" unsigned char U = (unsigned char)(-0.168736*R1-0.331264*G1+0.500000*B1+128);//(0.492*(B1-Y1)); \n"
" unsigned char V = (unsigned char)(0.500000*R1-0.418688*G1-0.081312*B1+128);//(0.877*(R1-Y1)); \n"
" yuvImg[get_global_id(0)][get_global_id(1)] = (unsigned int)(Y2 << 24 | V << 16 | Y1 << 8 | U); \n"
" } ";
m_kernel_code = oss.str();
m_sources.push_back({m_kernel_code.c_str(),m_kernel_code.length()});
m_program = *(new cl::Program(m_context,m_sources));
if(m_program.build({m_default_device})!=CL_SUCCESS){
std::cout<<" Error building: "<<m_program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(m_default_device)<<"\n";
exit(1);
}
// create buffers on the device
m_buffer_yuv = *(new cl::Buffer(m_context,CL_MEM_READ_WRITE,sizeof(unsigned int)*(m_width*m_height/2))); //each cell is int so it is 4 times the mem nedded, but each pixel is represented by 16 bits
m_buffer_rgb = *(new cl::Buffer(m_context,CL_MEM_READ_WRITE,sizeof(unsigned char)*(m_width*m_height*4))); // each pixel is represented by 4 bytes (alpha, RGB)
}
OpenClWrapper::~OpenClWrapper(){
free(&m_buffer_rgb);
free(&m_buffer_yuv);
}
void OpenClWrapper::RGB2YUV422(unsigned int * yuvImg, unsigned char * rgbImg){
cl::CommandQueue queue(m_context,m_default_device);
//write rgb image to the OpenCl buffer
queue.enqueueWriteBuffer(m_buffer_rgb,CL_TRUE,0,sizeof(unsigned char)*(m_width*m_height*4),rgbImg);
//run the kernel
cl::Kernel kernel_yuv2rgb=cl::Kernel(m_program,"RGB2YUV422");
kernel_yuv2rgb.setArg(0,m_buffer_rgb);
kernel_yuv2rgb.setArg(1,m_buffer_yuv);
queue.enqueueNDRangeKernel(kernel_yuv2rgb,cl::NullRange,cl::NDRange(m_height,(m_width/2)),cl::NullRange); //range is divided by 2 because we have width is represented in integers instead of 16bit (as needed in yuv422).
queue.finish();
//read result yuv Image from the device to yuv Image pointer
queue.enqueueReadBuffer(m_buffer_yuv,CL_TRUE,0,sizeof(unsigned int)*(m_width*m_height/2),yuvImg);
}
类OpenClWrapper
{
公众：
开口包装（尺寸、宽度、尺寸、高度）；
~OpenClWrapper（）；
void RGB2YUV422（无符号int*yuvImg，无符号char*rgbImg）；
私人：
std：：矢量m_所有_平台；
std：：矢量m_所有_设备；
cl：：平台m_默认_平台；
cl:：设备m_默认_设备；
cl:：Context m_Context；
cl:：Program:：Sources m_Sources；
cl：：程序m_程序；
cl：：命令队列m_队列；
cl：：Buffer m_Buffer_yuv；
cl：：缓冲区m_缓冲区rgb；
std：：字符串m_内核代码；
尺寸和宽度；
大小和高度；
};
class OpenClWrapper
{
public:
OpenClWrapper(size_t width, size_t height);
~OpenClWrapper();
void RGB2YUV422(unsigned int * yuvImg, unsigned char * rgbImg);
private:
std::vector<cl::Platform> m_all_platforms;
std::vector<cl::Device> m_all_devices;
cl::Platform m_default_platform;
cl::Device m_default_device;
cl::Context m_context;
cl::Program::Sources m_sources;
cl::Program m_program;
cl::CommandQueue m_queue;
cl::Buffer m_buffer_yuv;
cl::Buffer m_buffer_rgb;
std::string m_kernel_code;
size_t m_width;
size_t m_height;
};
#include "openclwrapper.h"
#include <iostream>
#include <sstream>
OpenClWrapper::OpenClWrapper(size_t width, size_t height) :
m_height(height),
m_width(width)
{
//get all platforms (drivers)
cl::Platform::get(&m_all_platforms);
if(m_all_platforms.size()==0){
std::cout<<" No platforms found. Check OpenCL installation!\n";
exit(1);
}
m_default_platform=m_all_platforms[0];
//get default device of the default platform
m_default_platform.getDevices(CL_DEVICE_TYPE_ALL, &m_all_devices);
if(m_all_devices.size()==0){
std::cout<<" No devices found. Check OpenCL installation!\n";
exit(1);
}
m_default_device=m_all_devices[0];
m_context = *(new cl::Context({m_default_device}));
std::ostringstream oss;
oss <<
" void kernel RGB2YUV422(global const unsigned char rgbImg[" << m_height << "][" << m_width << "*4], global unsigned int yuvImg[" << m_height << "][" << m_width << "/2]){ \n"
" int x_idx = get_global_id(0); \n"
" int y_idx = get_global_id(1)*8; \n"
" int alpha1 = rgbImg[x_idx][y_idx+3]; \n"
" int alpha2 = rgbImg[x_idx][y_idx+7]; \n"
" unsigned char R1 = rgbImg[x_idx][y_idx+2] * (255 / alpha1); \n"
" unsigned char G1 = rgbImg[x_idx][y_idx+1] * (255 / alpha1); \n"
" unsigned char B1 = rgbImg[x_idx][y_idx] * (255 / alpha1); \n"
" unsigned char R2 = rgbImg[x_idx][y_idx+6] * (255 / alpha2); \n"
" unsigned char G2 = rgbImg[x_idx][y_idx+5] * (255 / alpha2); \n"
" unsigned char B2 = rgbImg[x_idx][y_idx+4] * (255 / alpha2); \n"
" unsigned char Y1 = (unsigned char)(0.299000*R1 + 0.587000*G1 + 0.114000*B1); \n"
" unsigned char Y2 = (unsigned char)(0.299000*R2 + 0.587000*G2 + 0.114000*B2); \n"
" unsigned char U = (unsigned char)(-0.168736*R1-0.331264*G1+0.500000*B1+128);//(0.492*(B1-Y1)); \n"
" unsigned char V = (unsigned char)(0.500000*R1-0.418688*G1-0.081312*B1+128);//(0.877*(R1-Y1)); \n"
" yuvImg[get_global_id(0)][get_global_id(1)] = (unsigned int)(Y2 << 24 | V << 16 | Y1 << 8 | U); \n"
" } ";
m_kernel_code = oss.str();
m_sources.push_back({m_kernel_code.c_str(),m_kernel_code.length()});
m_program = *(new cl::Program(m_context,m_sources));
if(m_program.build({m_default_device})!=CL_SUCCESS){
std::cout<<" Error building: "<<m_program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(m_default_device)<<"\n";
exit(1);
}
// create buffers on the device
m_buffer_yuv = *(new cl::Buffer(m_context,CL_MEM_READ_WRITE,sizeof(unsigned int)*(m_width*m_height/2))); //each cell is int so it is 4 times the mem nedded, but each pixel is represented by 16 bits
m_buffer_rgb = *(new cl::Buffer(m_context,CL_MEM_READ_WRITE,sizeof(unsigned char)*(m_width*m_height*4))); // each pixel is represented by 4 bytes (alpha, RGB)
}
OpenClWrapper::~OpenClWrapper(){
free(&m_buffer_rgb);
free(&m_buffer_yuv);
}
void OpenClWrapper::RGB2YUV422(unsigned int * yuvImg, unsigned char * rgbImg){
cl::CommandQueue queue(m_context,m_default_device);
//write rgb image to the OpenCl buffer
queue.enqueueWriteBuffer(m_buffer_rgb,CL_TRUE,0,sizeof(unsigned char)*(m_width*m_height*4),rgbImg);
//run the kernel
cl::Kernel kernel_yuv2rgb=cl::Kernel(m_program,"RGB2YUV422");
kernel_yuv2rgb.setArg(0,m_buffer_rgb);
kernel_yuv2rgb.setArg(1,m_buffer_yuv);
queue.enqueueNDRangeKernel(kernel_yuv2rgb,cl::NullRange,cl::NDRange(m_height,(m_width/2)),cl::NullRange); //range is divided by 2 because we have width is represented in integers instead of 16bit (as needed in yuv422).
queue.finish();
//read result yuv Image from the device to yuv Image pointer
queue.enqueueReadBuffer(m_buffer_yuv,CL_TRUE,0,sizeof(unsigned int)*(m_width*m_height/2),yuvImg);
}
#包括“openclwrapper.h”
#包括
#包括
OpenClWrapper:：OpenClWrapper（大小\宽度、大小\高度）：
m_高度（高度），
m_宽度（宽度）
{
//获取所有平台（驱动程序）
cl:：平台：：获取（&m_所有平台）；
如果（m_所有平台.size（）==0）{
std:：coutOpenCV的一个简单实现：
void rgb_to_yuv422_uyvy(const cv::Mat& rgb, cv::Mat& yuv) {
assert(rgb.size() == yuv.size() &&
rgb.depth() == CV_8U &&
rgb.channels() == 3 &&
yuv.depth() == CV_8U &&
yuv.channels() == 2);
for (int ih = 0; ih < rgb.rows; ih++) {
const uint8_t* rgbRowPtr = rgb.ptr<uint8_t>(ih);
uint8_t* yuvRowPtr = yuv.ptr<uint8_t>(ih);
for (int iw = 0; iw < rgb.cols; iw = iw + 2) {
const int rgbColIdxBytes = iw * rgb.elemSize();
const int yuvColIdxBytes = iw * yuv.elemSize();
const uint8_t R1 = rgbRowPtr[rgbColIdxBytes + 0];
const uint8_t G1 = rgbRowPtr[rgbColIdxBytes + 1];
const uint8_t B1 = rgbRowPtr[rgbColIdxBytes + 2];
const uint8_t R2 = rgbRowPtr[rgbColIdxBytes + 3];
const uint8_t G2 = rgbRowPtr[rgbColIdxBytes + 4];
const uint8_t B2 = rgbRowPtr[rgbColIdxBytes + 5];
const int Y = (0.257f * R1) + (0.504f * G1) + (0.098f * B1) + 16.0f ;
const int U = -(0.148f * R1) - (0.291f * G1) + (0.439f * B1) + 128.0f;
const int V = (0.439f * R1) - (0.368f * G1) - (0.071f * B1) + 128.0f;
const int Y2 = (0.257f * R2) + (0.504f * G2) + (0.098f * B2) + 16.0f ;
yuvRowPtr[yuvColIdxBytes + 0] = cv::saturate_cast<uint8_t>(U );
yuvRowPtr[yuvColIdxBytes + 1] = cv::saturate_cast<uint8_t>(Y );
yuvRowPtr[yuvColIdxBytes + 2] = cv::saturate_cast<uint8_t>(V );
yuvRowPtr[yuvColIdxBytes + 3] = cv::saturate_cast<uint8_t>(Y2);
}
}
}
void rgb_to_yuv422_uyvy（常数cv:：Mat&rgb，cv:：Mat&yuv）{
断言（rgb.size（）==yuv.size（）&&
rgb.depth（）==CV_8U&&
rgb.channels（）==3&&
yuv.depth（）==CV_8U&&
yuv.channels（）==2）；
对于（int ih=0；ih
注意：假设（并检查）RGB以及YUV422 UYVY味道。我发现这相当快，但显然是。项目也有一个。我认为这将是最好的方式，因为它在GPU上执行。你可能可以从现有的cvtColor
实现中获得灵感。这是一个相当大的挑战。一般来说，它是可以接受的首先尝试使用OpenCL（如果有）并为给定的转换实现，然后尝试HAL版本（对于Tegra等极少数特定的东西）如果可用并实现，那么它可能会尝试IPP（如果对给定转换可用并实现），最后是基线实现，对于YUV转换，它似乎使用cv:：ParallelLoopImpl
和cv:：parallel\u for
。关于我如何实现这种颜色转换，我还没有计时，但这是一个问题去掉了现有的OpenCV代码（基线）。结果看起来很合理。我明天会继续玩。还没有看过Cuda的东西。|这可能是OpenCV的一个很好的补丁…它显然缺失了。。这在我的i4930k上用NVIdia GTX 760生成。使用12个线程（这是OpenCV默认的功能）我需要7毫秒的时间来转换256 x 65536 BGR图像。这对你来说够快吗？我仍在尝试OpenCL实现。CUDA版本的cvtColor
在两个方向上似乎都不支持YUV 4:2:2。这应该够快。我会尽快测试它，然后回复你。据我所知，Thanksip不是免费的.我更喜欢使用开源软件