使用OpenCL+OpenCV实现图像旋转（二）

来源：互联网发布：openstack 源码分析编辑：程序博客网时间：2024/05/16 19:01

4、host端程序代码

Host端程序处理流程就是按照前面“程序设计”一节编写的。除了调用OpenCL+OpenCV的API函数，其他的地方都是按照C/C++语法编写的。

具体代码如下：

1.// ImageRotate.cpp : 定义控制台应用程序的入口点。  2.//  3.  4.#include "stdafx.h"  5.#include <iostream>  6.#include <fstream>  7.#include <sstream>  8.  9.#include <opencv2/opencv.hpp>  10.  11.#ifdef __APPLE__  12.#include <OpenCL/cl.h>  13.#else  14.#include <CL/cl.h>  15.#endif  16.  17.using namespace cv;  18.  19.int _tmain(int argc, _TCHAR* argv[])  20.{  21.    cl_int ciErrNum;  22.    const char *fileName = "ImageRotate.cl";  23.    int width = 0, height = 0;  24.    float cos_theta = 0.7071067811865476, sin_theta = 0.7071067811865475;  //for degree 45  25.    //float cos_theta = 0.5, sin_theta = 0.5;  26.    const char* imageName = "F:\\code\\pic\\test01.jpg";  27.    char *bufInput = NULL, *bufOutput = NULL;  28.  29.  30.    //read one jpeg pic and store it in a Mat variable.  31.    Mat img = imread(imageName);  32.    if (!img.data) {  33.        std::cout << "fail to open the file:" << imageName << std::endl;  34.    }  35.  36.    //the type of img is RGB, convert to gray image.  37.    Mat imgGray;  38.    cvtColor(img, imgGray, CV_BGR2GRAY);  39.    width = imgGray.cols;  40.    height = imgGray.rows;  41.    std::cout << "picture width: " << width << ", height: " << height << std::endl;  42.      43.    //save the source data of original gray image.  44.    FILE *yuvFileOrg = NULL;  45.    fopen_s(&yuvFileOrg, "gray_org.yuv", "wb");  46.    fwrite(imgGray.data, width * height * sizeof(unsigned char), 1, yuvFileOrg);  47.    fclose(yuvFileOrg);  48.    yuvFileOrg = NULL;    49.  50.    //display the original gray image in a window.  51.    namedWindow( imageName, CV_WINDOW_AUTOSIZE );  52.    imshow(imageName, imgGray);  53.    //waitKey(0);  54.  55.    //allocate the input buffer to store the original gray image  56.    if (NULL == (bufInput = (char *)malloc(width * height * sizeof(char)))) {  57.        std::cerr << "Failed to malloc buffer for input image. " << std::endl;  58.        return NULL;  59.    }  60.  61.    //allocate the output buffer to store the image rotated.  62.    if (NULL == (bufOutput = (char *)malloc(width * height * sizeof(char)))) {  63.        std::cerr << "Failed to malloc buffer for output image. " << std::endl;  64.        return NULL;  65.    }  66.  67.    //copy the data of gray image to the input buffer. initialize the output buffer by zero.   68.    memcpy(bufInput, imgGray.data, width * height * sizeof(unsigned char));  69.    memset(bufOutput, 0, width * height * sizeof(unsigned char));  70.      71.    //use the first platform  72.    cl_platform_id platform;  73.    ciErrNum = clGetPlatformIDs(1, &platform, NULL);  74.  75.    //use the first device  76.    cl_device_id device;  77.    ciErrNum = clGetDeviceIDs(  78.        platform,  79.        CL_DEVICE_TYPE_ALL,  80.        1,  81.        &device,  82.        NULL);  83.  84.    cl_context_properties cps[3] = {  85.        CL_CONTEXT_PLATFORM, (cl_context_properties)platform, 0  86.    };  87.    //create the context  88.    cl_context ctx = clCreateContext(  89.        cps,  90.        1,  91.        &device,  92.        NULL,  93.        NULL,  94.        &ciErrNum);  95.  96.    //create the command queue  97.    cl_command_queue myqueue = clCreateCommandQueue(  98.        ctx,  99.        device,  100.        0,  101.        &ciErrNum);  102.          103.    //allocate space for original image on the device  104.    cl_mem bufferA = clCreateBuffer(  105.        ctx,  106.        CL_MEM_READ_ONLY,  107.        width * height * sizeof(unsigned char),  108.        NULL,  109.        &ciErrNum);  110.    //copy input buffer to the device  111.    ciErrNum = clEnqueueWriteBuffer(  112.        myqueue,  113.        bufferA,  114.        CL_TRUE,  115.        0,  116.        width * height * sizeof(unsigned char),  117.        (void *)bufInput,  118.        0,  119.        NULL,  120.        NULL);    121.  122.    //allocate space for rotated image on the device  123.    cl_mem bufferC = clCreateBuffer(  124.        ctx,  125.        CL_MEM_READ_WRITE,  126.        width * height * sizeof(unsigned char),  127.        NULL,  128.        &ciErrNum);  129.  130.    //open kernel file and read the content to a string variable.  131.    std::ifstream kernelFile("ImageRotate.cl", std::ios::in);  132.    if (!kernelFile.is_open()) {  133.        std::cerr << "Failed to open file for reading: " << fileName << std::endl;  134.        return NULL;  135.    }  136.    std::ostringstream oss;  137.    oss << kernelFile.rdbuf();  138.    std::string srcStdStr = oss.str();  139.    const char *srcStr = srcStdStr.c_str();  140.    kernelFile.close();  141.  142.    //create the program with source code of kernel.  143.    cl_program myprog = clCreateProgramWithSource(  144.        ctx,  145.        1,  146.        (const char**)&srcStr,  147.        NULL,  148.        &ciErrNum);  149.  150.    //compile the program. passing NULL for the 'device_list' argument targets all devices in the context  151.    ciErrNum = clBuildProgram(myprog, 0, NULL, NULL, NULL, NULL);  152.  153.    //create the kernel  154.    cl_kernel mykernel = clCreateKernel(  155.        myprog,  156.        "img_rotate",  157.        &ciErrNum);  158.  159.    //set the kernel arguments  160.    clSetKernelArg(mykernel, 0, sizeof(cl_mem), (void *)&bufferC);  161.    clSetKernelArg(mykernel, 1, sizeof(cl_mem), (void *)&bufferA);  162.    clSetKernelArg(mykernel, 2, sizeof(cl_int), (void *)&width);  163.    clSetKernelArg(mykernel, 3, sizeof(cl_int), (void *)&height);  164.    clSetKernelArg(mykernel, 4, sizeof(cl_float), (void *)&cos_theta);  165.    clSetKernelArg(mykernel, 5, sizeof(cl_float), (void *)&sin_theta);  166.  167.    //set local and global workgroup sizes  168.    size_t localws[2] = {1, 1};  169.    size_t globalws[2] = {width, height};  170.  171.    //execute the kernel  172.    ciErrNum = clEnqueueNDRangeKernel(  173.        myqueue,  174.        mykernel,  175.        2,  176.        NULL,  177.        globalws,  178.        localws,  179.        0,  180.        NULL,  181.        NULL);  182.  183.    //read the output data back to the host  184.    ciErrNum = clEnqueueReadBuffer(  185.        myqueue,  186.        bufferC,  187.        CL_TRUE,  188.        0,  189.        width * height * sizeof(unsigned char),  190.        bufOutput,  191.        0,  192.        NULL,  193.        NULL);  194.  195.    //copy the output data from output buffer to Mat variable.   196.    memcpy(imgGray.data, bufOutput, width * height * sizeof(unsigned char));  197.  198.    //save the source data for gray image rotated  199.    FILE *yuvFile = NULL;  200.    fopen_s(&yuvFile, "gray.yuv", "wb");  201.    fwrite(imgGray.data, width * height * sizeof(unsigned char), 1, yuvFile);  202.    fclose(yuvFile);  203.    yuvFile = NULL;  204.  205.    //save the gray image rotated.  206.    imwrite("test_gray.jpg", imgGray);  207.  208.    //show the gray image rotated.  209.    const char *winName = "gray_image_rotated";  210.    namedWindow(winName, CV_WINDOW_AUTOSIZE );  211.    imshow(winName, imgGray);     212.    waitKey(0);  213.    destroyAllWindows();  214.  215.    //release all resource  216.    if (bufInput != NULL)  217.        free(bufInput);  218.  219.    if (bufOutput != NULL)  220.        free(bufOutput);  221.  222.    if (bufferA != 0)  223.        clReleaseMemObject(bufferA);  224.  225.    if (bufferC != 0)  226.        clReleaseMemObject(bufferC);  227.  228.    if (myqueue != 0)  229.        clReleaseCommandQueue(myqueue);  230.  231.    if (mykernel != 0)  232.        clReleaseKernel(mykernel);  233.  234.    if (myprog != 0)  235.        clReleaseProgram(myprog);  236.  237.    if (ctx != 0)  238.        clReleaseContext(ctx);  239.  240.    return 0;  241.}

5、程序处理结果

原始的灰度图像如下所示：

经过45度旋转的图像如下：

将sin、cos值都设置为0.5时的处理结果如下：

经过45度旋转的图像上面有很多暗点，那些暗点是在设备端分配的buffer中原始的点。意味着那些区域并非每个点都有对应的点旋转过来。

在kernel程序中计算坐标点时，使用的是float类型，最后获取图像时是将float转换为int了。应该是因为精度损失导致某些坐标点上没有对应的像素点，所以保留了buffer中原有的数据。

而当sin、cos设置为0.5时，图像看着就没有暗纹。按照0.5计算很少有精度的损失，所以就不存在上面的问题。

是因为旋转算法精度不够，还是程序哪里实现错了，这是个问题，将在后面的学习中去寻找问题的答案。

（完）

0 0