SSE优化系列九：小半径中值滤波的SSE极速优化（比OpenCV3.1.0快9-10倍）

前言

在优化系列六:https://blog.youkuaiyun.com/just_sort/article/details/97280807 中提出了一个直方图算法处理框架，可以处理中值滤波，最大值滤波等，算法复杂度是和半径大小无关的。但是这里忽略了一个点在半径很小的时候，这个算法实际上是比直接暴力处理更慢的。而我们在日常图像处理过程中，小半径的中值滤波是最常用的例如3 * 3, 5 * 5 在PS中都是十分常用的。受到ImageShop的这篇博客：https://www.cnblogs.com/Imageshop/p/11087804.html 启发，所以有了这篇文章。

朴素的3*3中值模糊

这个没有什么说的，就是在半径范围里面遍历即可，这里获取中值直接使用了c语言的qsort。

int ComparisonFunction(const void *X, const void *Y) {
	unsigned char Dx = *(unsigned char *)X;
	unsigned char Dy = *(unsigned char *)Y;
	if (Dx < Dy) return -1;
	else if (Dx > Dy) return 1;
	else return 0;
}

void MedianBlur3X3_Ori(unsigned char *Src, unsigned char *Dest, int Width, int Height, int Stride) {
	int Channel = Stride / Width;
	if (Channel == 1) {
		unsigned char Array[9];
		for (int Y = 1; Y < Height - 1; Y++) {
			unsigned char *LineP0 = Src + (Y - 1) * Stride + 1;
			unsigned char *LineP1 = LineP0 + Stride;
			unsigned char *LineP2 = LineP1 + Stride;
			unsigned char *LinePD = Dest + Y * Stride + 1;
			for (int X = 1; X < Width - 1; X++) {
				Array[0] = LineP0[X - 1];        Array[1] = LineP0[X];    Array[2] = LineP0[X + 1];
				Array[3] = LineP1[X - 1];        Array[4] = LineP1[X];    Array[5] = LineP2[X + 1];
				Array[6] = LineP2[X - 1];        Array[7] = LineP2[X];    Array[8] = LineP2[X + 1];
				qsort(Array, 9, sizeof(unsigned char), &ComparisonFunction);
				LinePD[X] = Array[4];
			}
		}
	}
	else {
		unsigned char ArrayB[9], ArrayG[9], ArrayR[9];
		for (int Y = 1; Y < Height - 1; Y++) {
			unsigned char *LineP0 = Src + (Y - 1) * Stride + 3;
			unsigned char *LineP1 = LineP0 + Stride;
			unsigned char *LineP2 = LineP1 + Stride;
			unsigned char *LinePD = Dest + Y * Stride + 3;
			for (int X = 1; X < Width - 1; X++){
				ArrayB[0] = LineP0[-3];       ArrayG[0] = LineP0[-2];       ArrayR[0] = LineP0[-1];
				ArrayB[1] = LineP0[0];        ArrayG[1] = LineP0[1];        ArrayR[1] = LineP0[2];
				ArrayB[2] = LineP0[3];        ArrayG[2] = LineP0[4];        ArrayR[2] = LineP0[5];

				ArrayB[3] = LineP1[-3];       ArrayG[3] = LineP1[-2];       ArrayR[3] = LineP1[-1];
				ArrayB[4] = LineP1[0];        ArrayG[4] = LineP1[1];        ArrayR[4] = LineP1[2];
				ArrayB[5] = LineP1[3];        ArrayG[5] = LineP1[4];        ArrayR[5] = LineP1[5];

				ArrayB[6] = LineP2[-3];       ArrayG[6] = LineP2[-2];       ArrayR[6] = LineP2[-1];