CUDA小白 - NPP(3) 图像处理 Color and Sampling Conversion

这篇具有很好参考价值的文章主要介绍了CUDA小白 - NPP(3) 图像处理 Color and Sampling Conversion。希望对大家有所帮助。如果存在错误或未考虑完全的地方,请大家不吝赐教,您也可以点击"举报违法"按钮提交疑问。

cuda小白
原始API链接 NPP

GPU架构近些年也有不少的变化,具体的可以参考别的博主的介绍,都比较详细。还有一些cuda中的专有名词的含义,可以参考《详解CUDA的Context、Stream、Warp、SM、SP、Kernel、Block、Grid》

常见的NppStatus,可以看这里。

如有问题,请指出,谢谢

本节内容主要涉及到使用NPP进行一些图像色域变换等,类似功能的就不细说,仅针对一些常见的。由于Debayer,不同yuv之间的转换(Color Sampling Format Conversion),Gamma校正,Complement Color Key(一直不清楚这个啥)以及Color Processing(调色相关)的接触不对,后面就不介绍了,按需。

Color Model Conversion

当前模块主要涉及到的RGB,BGR,YUV,YUV420,YUV422,NV12,YCbCr,YCbCr422和YCbCr420之间的相互转换,以及批处理(分为同样ROI和不同ROI两个接口)的接口。其他还有一些未接触过的XYZ,LUV,Lab,YCC,CMYK_JPEG,HLS,HSV等与RGB或者BGR之间的转换。
本文就那比较简单的转换,以为RGB和YUV之间的相互转换为例子。

// 此处介绍的RGB和YUV都是三通道的,如果是packed的相互转换,则后缀是C3R,如果都是planar的,则后缀是P3R,反之则是P3C3R或者C3P3R。
// =========== RGB2YUV原理 ===========
// Npp32f nY =  0.299F * R + 0.587F * G + 0.114F * B; 
// Npp32f nU = (0.492F * ((Npp32f)nB - nY)) + 128.0F;
// Npp32f nV = (0.877F * ((Npp32f)nR - nY)) + 128.0F;
// if (nV > 255.0F) 
//     nV = 255.0F; 
NppStatus nppiRGBToYUV_8u_C3R(const Npp8u *pSrc,
							  int nSrcStep,
							  Npp8u *pDst,
							  int nDstStep,
							  NppiSize oSizeROI);

// =========== RGB2YUV原理 ===========
// Npp32f nY = (Npp32f)Y;
// Npp32f nU = (Npp32f)U - 128.0F;
// Npp32f nV = (Npp32f)V - 128.0F;
// Npp32f nR = nY + 1.140F * nV; 
// if (nR < 0.0F)    nR = 0.0F;
// if (nR > 255.0F)  nR = 255.0F;    
// Npp32f nG = nY - 0.394F * nU - 0.581F * nV;
// if (nG < 0.0F)    nG = 0.0F;
// if (nG > 255.0F)  nG = 255.0F;    
// Npp32f nB = nY + 2.032F * nU;
// if (nB < 0.0F)    nB = 0.0F;
// if (nB > 255.0F)  nB = 255.0F;    
NppStatus nppiYUVToRGB_8u_C3R(const Npp8u *pSrc,
						  	  int nSrcStep,
						 	  Npp8u *pDst,
							  int nDstStep,
							  NppiSize oSizeROI);	
code
#include <iostream>
#include <cuda_runtime.h>
#include <npp.h>
#include <opencv2/opencv.hpp>

#define CUDA_FREE(ptr) { if (ptr != nullptr) { cudaFree(ptr); ptr = nullptr; } }

int main() {
  std::string directory = "../";

  // =============== load image ===============
  cv::Mat image_dog = cv::imread(directory + "dog.png");
  int image_width = image_dog.cols;
  int image_height = image_dog.rows;
  int image_size = image_width * image_height * 3;

  // =============== device memory ===============
  uint8_t *in_ptr, *yuv, *out_ptr;
  cudaMalloc((void**)&in_ptr, image_size * sizeof(uint8_t));
  cudaMalloc((void**)&yuv, image_size * sizeof(uint8_t));
  cudaMalloc((void**)&out_ptr, image_size * sizeof(uint8_t));
  cudaMemcpy(in_ptr, image_dog.data, image_size, cudaMemcpyHostToDevice);
  
  NppiSize roi1, roi2;
  roi1.width = image_width;
  roi1.height = image_height;
  roi2.width = image_width / 2;
  roi2.height = image_height / 2;

  // =============== nppiRGBToYUV_8u_C3R ===============
  uint8_t *host_yuv = (uint8_t *)malloc(image_size);
  NppStatus status;
  status = nppiRGBToYUV_8u_C3R(in_ptr, image_width * 3, yuv, image_width * 3, roi1);
  if (status != NPP_SUCCESS) {
    std::cout << "[GPU] ERROR nppiRGBToYUV_8u_C3R failed, status = " << status << std::endl;
    return false;
  }
  cudaMemcpy(host_yuv, yuv, image_size, cudaMemcpyDeviceToHost);
  std::string out_yuv_file = directory + "yuv.bin";
  FILE *fp = fopen(out_yuv_file.c_str(), "wb");
  fwrite(host_yuv, image_size, 1, fp);
  fclose(fp);

  // =============== nppiYUVToRGB_8u_C3R ===============
  cv::Mat out_image = cv::Mat::zeros(image_height, image_width, CV_8UC3);
  status = nppiYUVToRGB_8u_C3R(yuv, image_width * 3, out_ptr, image_width * 3, roi2);
  if (status != NPP_SUCCESS) {
    std::cout << "[GPU] ERROR nppiYUVToRGB_8u_C3R failed, status = " << status << std::endl;
    return false;
  }
  cudaMemcpy(out_image.data, out_ptr, image_size, cudaMemcpyDeviceToHost);
  cv::imwrite(directory + "out_rgb.jpg", out_image);

  // free
  CUDA_FREE(in_ptr)
  CUDA_FREE(yuv)
  CUDA_FREE(out_ptr)
  if (host_yuv != nullptr) { free(host_yuv); host_yuv = nullptr; }
}
make
cmake_minimum_required(VERSION 3.20)
project(test)

find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})

find_package(CUDA REQUIRED)
include_directories(${CUDA_INCLUDE_DIRS})
file(GLOB CUDA_LIBS "/usr/local/cuda/lib64/*.so")

add_executable(test test.cpp)
target_link_libraries(test
                      ${OpenCV_LIBS}
                      ${CUDA_LIBS}
)
result

yuv的结果,使用7yuv工具查看
CUDA小白 - NPP(3) 图像处理 Color and Sampling Conversion,c++,CUDA,NPP
使用转换之后的yuv,再转回RGB,在限定ROI之后的结果:
CUDA小白 - NPP(3) 图像处理 Color and Sampling Conversion,c++,CUDA,NPP
注意点:

  1. 可以使用7yuv进行yuv图像的查看,7yuv,下载对应的版本即可,可以体验。
  2. 还有一个接口是基于JPEG的,主要介绍的是JPEG的Color Conversion,由于平时接触较少,这里暂时不介绍了,按需取《JPEG Color Conversion》.
ColorToGray Conversion

当前模块也主要分为三个大方向:RGBToGray,ColorToGray和GradientColorToGray。三个大方向都只选取一个比较常见的进行介绍:

// nGray =  0.299F * R + 0.587F * G + 0.114F * B; 
NppStatus nppiRGBToGray_8u_C3C1R(const Npp8u *pSrc,
								 int nSrcStep,
								 Npp8u *pDst,
								 int nDstStep,
								 NppiSize oSizeROI);
// nGray =  aCoeffs[0] * R + aCoeffs[1] * G + aCoeffs[2] * B;
NppStatus nppiColorToGray_8u_C3C1R(const Npp8u *pSrc,
								   int nSrcStep,
								   Npp8u *pDst,
								   int nDstStep,
								   NppiSize oSizeROI,
								   const Npp32f aCoeffs[3]);							 
// NppiNorm { max(inf) / sum(L1) / square root of sum of squares(L2) }
NppStatus nppiGradientColorToGray_8u_C3C1R(const Npp8u *pSrc,
										   int nSrcStep,
										   Npp8u *pDst,
										   int nDstStep,
										   NppiSize oSizeROI,
										   NppiNorm eNorm);
code
#include <iostream>
#include <cuda_runtime.h>
#include <npp.h>
#include <opencv2/opencv.hpp>

#define CUDA_FREE(ptr) { if (ptr != nullptr) { cudaFree(ptr); ptr = nullptr; } }

int main() {
  std::string directory = "../";

  // =============== load image ===============
  cv::Mat image_dog = cv::imread(directory + "dog.png");
  cv::cvtColor(image_dog, image_dog, CV_RGB2BGR);
  int image_width = image_dog.cols;
  int image_height = image_dog.rows;
  int image_size = image_width * image_height;

  // =============== device memory ===============
  uint8_t *in_ptr;
  cudaMalloc((void**)&in_ptr, image_size * 3 * sizeof(uint8_t));
  cudaMemcpy(in_ptr, image_dog.data, image_size * 3 * sizeof(uint8_t), cudaMemcpyHostToDevice);
  
  uint8_t *out_ptr1, *out_ptr2, *out_ptr3;
  cudaMalloc((void**)&out_ptr1, image_size * sizeof(uint8_t));
  cudaMalloc((void**)&out_ptr2, image_size * sizeof(uint8_t));
  cudaMalloc((void**)&out_ptr3, image_size * sizeof(uint8_t));

  NppiSize roi1, roi2;
  roi1.width = image_width;
  roi1.height = image_height;
  roi2.width = image_width / 2;
  roi2.height = image_height / 2;

  // =============== nppiRGBToGray_8u_C3C1R ===============
  cv::Mat out_image_gray = cv::Mat::zeros(image_height, image_width, CV_8UC1);
  NppStatus status;
  status = nppiRGBToGray_8u_C3C1R(in_ptr, image_width * 3, out_ptr1, image_width, roi1);
  if (status != NPP_SUCCESS) {
    std::cout << "[GPU] ERROR nppiRGBToGray_8u_C3C1R failed, status = " << status << std::endl;
    return false;
  }
  cudaMemcpy(out_image_gray.data, out_ptr1, image_size, cudaMemcpyDeviceToHost);
  cv::imwrite(directory + "rgb2gray.jpg", out_image_gray);

  // =============== nppiColorToGray_8u_C3C1R ===============
  float coeffs[3] = { 0.1f, 0.8f, 0.1f};
  status = nppiColorToGray_8u_C3C1R(in_ptr, image_width * 3, out_ptr2, image_width, roi1, coeffs);
  if (status != NPP_SUCCESS) {
    std::cout << "[GPU] ERROR nppiColorToGray_8u_C3C1R failed, status = " << status << std::endl;
    return false;
  }
  cudaMemcpy(out_image_gray.data, out_ptr2, image_size, cudaMemcpyDeviceToHost);
  cv::imwrite(directory + "color2rgb.jpg", out_image_gray);

  // =============== nppiGradientColorToGray_8u_C3C1R_Ctx ===============
  status = nppiGradientColorToGray_8u_C3C1R(in_ptr, image_width * 3, out_ptr3, image_width, roi1, nppiNormInf);
  if (status != NPP_SUCCESS) {
    std::cout << "[GPU] ERROR nppiGradientColorToGray_8u_C3C1R failed, status = " << status << std::endl;
    return false;
  }
  cudaMemcpy(out_image_gray.data, out_ptr3, image_size, cudaMemcpyDeviceToHost);
  cv::imwrite(directory + "gradient.jpg", out_image_gray);

  // free
  CUDA_FREE(in_ptr)
  CUDA_FREE(out_ptr1)
  CUDA_FREE(out_ptr2)
  CUDA_FREE(out_ptr3)
}
make
cmake_minimum_required(VERSION 3.20)
project(test)

find_package(OpenCV REQUIRED)
include_directories(${OpenCV_INCLUDE_DIRS})

find_package(CUDA REQUIRED)
include_directories(${CUDA_INCLUDE_DIRS})
file(GLOB CUDA_LIBS "/usr/local/cuda/lib64/*.so")

add_executable(test test.cpp)
target_link_libraries(test
                      ${OpenCV_LIBS}
                      ${CUDA_LIBS}
)
result

CUDA小白 - NPP(3) 图像处理 Color and Sampling Conversion,c++,CUDA,NPP
注意点:文章来源地址https://www.toymoban.com/news/detail-691962.html

  1. Color2Gray的接口与RGB2Gray的区别主要在于三个通道是不是比例关系。由于都是灰度图的,ColorGray的变化与直接rgb的转换结果差异性不易从肉眼看出来,但是通过不同的比例组合明显看出图像是有区别的。

到了这里,关于CUDA小白 - NPP(3) 图像处理 Color and Sampling Conversion的文章就介绍完了。如果您还想了解更多内容,请在右上角搜索TOY模板网以前的文章或继续浏览下面的相关文章,希望大家以后多多支持TOY模板网!

本文来自互联网用户投稿,该文观点仅代表作者本人,不代表本站立场。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如若转载,请注明出处: 如若内容造成侵权/违法违规/事实不符,请点击违法举报进行投诉反馈,一经查实,立即删除!

领支付宝红包 赞助服务器费用

相关文章

觉得文章有用就打赏一下文章作者

支付宝扫一扫打赏

博客赞助

微信扫一扫打赏

请作者喝杯咖啡吧~博客赞助

支付宝扫一扫领取红包,优惠每天领

二维码1

领取红包

二维码2

领红包