Opencv中Mat高效率 左右上下平移一定的像素

Posted 2020-11-15 sofard

工作项目中需要完成图像平移的操作，由于是突然加入的需求，其他部分已经占用大量CPU时间，因此为了完成视频的同步，平移操作的效率要求很高
最开始采用仿射变换的API，时间很长，740*440分辨率大概需要个5ms左右，考虑到Mat中存在部分行复制的方法，就试了下

// TestMatSubCopy.cpp: 定义控制台应用程序的入口点。
#include "stdafx.h"
#include <opencvcv.h>
#include <opencv2/core/core.hpp>
#include <opencv2/imgcodecs.hpp>
#include <opencv2/highgui.hpp>
#include <opencv2/opencv.hpp>
#include <chrono>
using namespace std;
using namespace cv ;

inline void shiftLeftRightUpDown(Mat & src, Mat & dst,Mat & temp , int w, int h, int x, int y) {

	int x1, x2, y1, y2;
	int X1, X2, Y1, Y2;
	if (x > 0) {
		x1 = x; x2 = w;
		X1 = 0; X2 = w - x;
	}
	else{
		x1 = 0; x2 = w + x;
		X1 = -x; X2 = w;
	}
	if (y > 0) {
		y1 = y; y2 = h;
		Y1 = 0; Y2 = h - y;
	}
	else {
		y1 = 0; y2 = h + y;
		Y1 = -y; Y2 = h;
	}
	src.colRange(X1, X2).copyTo(dst.colRange(x1, x2));
	dst.rowRange(Y1, Y2).copyTo(temp.rowRange(0, Y2 - Y1));
	temp.rowRange(0, Y2 - Y1).copyTo(dst.rowRange(y1, y2));
}
int main()
{
	Mat inImage = imread("pic1.jpg", 1);//= Mat::zeros(1920, 1080, CV_8UC3);
	Mat src;
	resize(inImage,src,cv::Size(740, 440));
	Mat dst = Mat::zeros(src.rows, src.cols, CV_8UC3);
	Mat temp = Mat::zeros(src.rows, src.cols, CV_8UC3);
	imshow("src", src);
	cv::waitKey(0);
	auto t1 = std::chrono::steady_clock::now();
	auto t2 = std::chrono::steady_clock::now();
	auto duration = std::chrono::duration_cast<std::chrono::microseconds>(t2 - t1).count();
	
	int time = 25000;
	auto total_duration = duration - duration;
	int count = time;
	t1 = std::chrono::steady_clock::now();
	while (time > 0) {
		--time;
		shiftLeftRightUpDown(src, dst,temp, src.cols, src.rows, 10, 10);
	}
	t2 = std::chrono::steady_clock::now();
	total_duration = std::chrono::duration_cast<std::chrono::microseconds>(t2 - t1).count();
	cout << "average cost time :" << total_duration / count << endl;
	imshow("dst", dst);
	cv::waitKey(0);
    return 0;
}

在3.5GHZ的cpu上740*440分辨率图片实现速度大概在175us左右，平移像素越多，速度越快，大概能提高5-10us左右的速度，将主要函数中变量声明为静态变量可以提高3us左右
对于1080P视频，运行时间在3-4ms.