RGB2Gray cpp file

rgb2gray_main.cpp

#include <opencv2/core/core.hpp>
#include <opencv2/core/utility.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <iostream>
#include <stdio.h>
#include <cuda.h>
#include <cuda_runtime.h>

using namespace std;
using namespace cv;

extern "C" void gray_parallel(unsigned char* h_in, unsigned char* h_out, int elems, int rows, int cols);

// My serial implementation, works fine
Mat gray_serial(Mat img){
  int rows = img.rows;
	int cols = img.cols;
	Mat gray(rows, cols, CV_8UC1);
	for(int r=0; r<rows; r++){
		for(int c=0; c<cols; c++){
			Vec3b bgr = img.at<Vec3b>(r,c);
			double gray_val = 0.144*bgr.val[0] + 0.587*bgr.val[1] + 0.299*bgr.val[2];
			gray.at<unsigned char>(r,c) = (unsigned char)gray_val;
		}
	}
	return gray;
}

// One more serial code, just for testing, works fine
Mat gray_test(Mat img){
	cout << "running test" << endl;
	uint rows = img.rows;
	uint cols = img.cols;
	unsigned char* test = img.data;
	unsigned char* op = new unsigned char[rows*cols];
	for (uint i=0; i<rows*cols; i++){
		uint index = 3*i;
		double temp = 0.144*test[index]+0.587*test[index+1]+0.299*test[index+2];
		op[i] = (unsigned char)temp;
	}
	Mat gray = Mat(rows, cols, CV_8UC1, op);
	return gray;
}

//      Main Function

int main( int argc, char** argv )
{	
    Mat image;
    image = imread("baboon2.jpg");	// Read the file, it is simple 64x64 image. 

//           First trying serial code      
	int64 t1 = cv::getTickCount();
	Mat gray = gray_test(image);
	int64 t2 = cv::getTickCount();
	cout << double(t2 - t1)/getTickCount() << " " << endl;


//         Now trying GPU code
    int64 t3 = cv::getTickCount();
	const int rows = image.rows;
	const int cols = image.cols;
	int elems = rows*cols*3;
	unsigned char *h_in = image.data;
	unsigned char *h_out = new unsigned char[rows*cols];
    
	gray_parallel(h_in, h_out, elems, rows, cols);
	
	Mat gray2 = Mat(rows,cols,CV_8UC1,h_out);
	int64 t4 = cv::getTickCount();
	cout << double(t4 - t3)/getTickCount() << " " << sizeof(unsigned char) <<  endl;

    // Finally display result of GPU code (CPU code works fine, but not GPU code)
	imshow("image.jpg",gray2);
	waitKey(0);
	return 0;
}