cuda(1)-imageBlur

　　以前我用CUDA 写过简单的图形模糊的例子，时间长了就忘了，现在看的时候不得不再花时间学习，

所以就想要记录在博客园上，方便自己复习和记忆。一旦自己用到这放面的程序的时候就可以随时下载。

　　简单的均值模糊就是将像素点周围的像素值加起来求均值。具体的原理我就不在这里赘述了直接上

CUDA 代码：

#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include <stdio.h>
#include<iostream>

#include <iostream>  
#include <opencv2/core/core.hpp>  
#include <opencv2/highgui/highgui.hpp> 

#define CHANNELS 3
#define BLUR_SIZE 3



__global__ void Img2Blurincuda(unsigned char *in, unsigned char*out, int imgwidth, int imgheight, int channels) {
    int ty = blockIdx.y*blockDim.y + threadIdx.y;
    int tx = blockIdx.x*blockDim.x + threadIdx.x;

    if (ty < imgheight&&tx < imgwidth) {
        int PixVal = 0;
        int Pixels = 0;
        for (int k = 0; k < channels; ++k) {
            for (int blurRow = -BLUR_SIZE; blurRow < BLUR_SIZE + 1; ++blurRow) {
                for (int blurCol = -BLUR_SIZE; blurCol < BLUR_SIZE + 1; ++blurCol) {

                    int Cur_ty = ty + blurRow;
                    int Cur_tx = tx + blurCol;

                    if (Cur_ty > -1 && Cur_ty<imgheight && Cur_tx>-1 && Cur_tx < imgwidth) {
                        PixVal += in[(Cur_ty*imgwidth + Cur_tx)*channels + k];
                        Pixels++;
                    }
                }
            }
            out[(ty*imgwidth + tx)*channels + k] = (unsigned char)(PixVal / Pixels);
        }

    }
}

using namespace cv;
int main(void) {

    // 读入一张图片（lena）
    Mat img = imread("E:\\opencv\\lena512.bmp");  //lena512color.tiff;lena512.bmp
    // 创建一个名为 "lena"窗口
    namedWindow("lena");
    // 在窗口中显示游戏原画
    imshow("lena", img);
    // 等待3000 ms后窗口自动关闭
    waitKey(3000);

    const int imgheight = img.rows;
    const int imgwidth = img.cols;
    const int imgchannel = img.channels();

    printf("imgchannel=%d\n", img.channels());

    //Mat blurImage(imgheight, imgwidth, CV_8UC3, Scalar(0));
    Mat blurImage(imgheight, imgwidth, CV_8UC3);

    //uchar3 *d_in;
    unsigned char *d_in;
    unsigned char*d_out;

    //printf("%d\n", sizeof(uchar3));

    cudaMalloc((void**)&d_in, imgheight*imgwidth *imgchannel * sizeof(unsigned char));
    cudaMalloc((void**)&d_out, imgheight*imgwidth *imgchannel * sizeof(unsigned char));

    cudaMemcpy(d_in, img.data, imgheight*imgwidth * imgchannel * sizeof(unsigned char), cudaMemcpyHostToDevice);

    dim3 BlockDim(16, 16);
    dim3 GridDim((imgwidth - 1) / BlockDim.x + 1, (imgheight - 1) / BlockDim.y + 1);

    clock_t start, end;
    start = clock();

    Img2Blurincuda << <GridDim, BlockDim >> > (d_in, d_out, imgheight, imgwidth, imgchannel);

    cudaDeviceSynchronize();

    end = clock();
    printf("cuda exec time is %.8f\n", (double)(end - start) / CLOCKS_PER_SEC);
    cudaMemcpy(blurImage.data, d_out, imgheight*imgwidth * imgchannel * sizeof(unsigned char), cudaMemcpyDeviceToHost);

    cudaFree(d_in);
    cudaFree(d_out);

    


    imshow("blurImage", blurImage);
    waitKey(6000);
    return 0;

}

运行的结果如下：

原始图片：

模糊处理后：