＜C++＞ 基于SSE实现图像二值化

基于SSE实现图像二值化

SSE介绍及使用可见：https://blog.csdn.net/thisiszdy/article/details/132512686
本文使用SSE指令集来实现图像二值化算法，同时对比OpenCV二值化算子及for循环求解二值化的效果及性能。

// opencvTest.cpp : 此文件包含 "main" 函数。程序执行将在此处开始并结束。
//

#include <iostream>
#include "opencv2/core/core.hpp"
#include "opencv2/imgproc.hpp"
#include "opencv2/core.hpp"
#include <emmintrin.h>
#include <opencv2/imgcodecs.hpp>

using namespace cv;
using namespace std;

int main()
{
	clock_t start, end;
	const char* imageName = "test.bmp";
	cv::Mat src = imread(imageName, IMREAD_GRAYSCALE);

	//OpenCV二值化
	start = clock();
	cv::Mat dst;
	cv::threshold(src, dst, 50, 255, cv::THRESH_BINARY);
	end = clock();
	double time_cost = (double)(end - start);
	std::cout << "OpenCV Threshold耗时：" << time_cost << std::endl;

	//for循环二值化
	start = clock();
	cv::Mat dstfor(src.rows, src.cols, CV_8UC1);
	uchar* ptrSrc = src.data;
	uchar* ptrdst = dstfor.data;
	int lenght = src.rows * src.cols * src.channels();
#pragma omp parallel
	for (int i = 0;i < lenght;i++)
	{
		if (*(ptrSrc + i) <= 50)
		{
			*(ptrdst + i) = 0;
		}
		else
		{
			*(ptrdst + i) = 255;
		}
	}
	end = clock();
	double time_cost1 = (double)(end - start);
	std::cout << "for Threshold耗时：" << time_cost1 << std::endl;

	//SSE加速二值化
	start = clock();
	uchar thre = 50;
	uchar maxV = 255;
	cv::Mat dts3(src.rows, src.cols, CV_8UC1);
	uchar* ptrSrc1 = src.ptr();
	uchar* ptrdst1 = dts3.data;
	size_t srcstep = src.step;
	__m128i smask = _mm_set1_epi8((char)128);
	__m128i mThre = _mm_set1_epi8(50);
#pragma omp parallel
	for (size_t i = 0; i < src.rows; i++, ptrSrc1 += srcstep, ptrdst1 += srcstep)
	{
		for (size_t j = 0; j <= src.cols * src.channels() - 16; j += 16)
		{
			__m128i mValue = _mm_loadu_si128((__m128i*)(ptrSrc1 + j)); //加载内存到寄存器
			auto sub1 = _mm_sub_epi8(mValue, smask); //无符号0~255范围转为有符号-128~127范围
			auto sub2 = _mm_sub_epi8(mThre, smask);
			mValue = _mm_cmpgt_epi8(sub1, sub2); //与阈值比较，大于阈值返回255，小于阈值返回0
			_mm_storeu_si128((__m128i*)(ptrdst1 + j), mValue); //寄存器加载到内存
		}
	}
	end = clock();
	double time_cost2 = (double)(end - start);
	std::cout << "SSE Threshold耗时：" << time_cost2 << std::endl;

	// 效果比较
	Mat findzero1 = dst != dstfor; //OpenCV和for循环比较效果
	Mat findzero2 = dst != dts3; //Opencv和SEE指针比较效果
	vector<cv::Point> veczero1;
	vector<cv::Point> veczero2;
	cv::findNonZero(findzero1, veczero1);
	cv::findNonZero(findzero2, veczero2);
	int num1 = veczero1.size();
	int num2 = veczero2.size();
	std::cout << "OpenCV和for循环两种方法逐像素比较，像素不同个数：" << num1 << std::endl;
	std::cout << "Opencv和SEE指针两种方法逐像素比较，像素不同个数：" << num2 << std::endl;

	system("pause");
	return 0;
}

运算结果显示，SSE指令集加速阈值分割更快，检测效果与OpenCV一致。