Image processing with SIMD optimization

Some image processing method with SSE/AVX optimization.

Functions:

int FastBlur(unsigned char *img, unsigned char *dst, int width, int height, int roi_x, int roi_y, int roi_w, int roi_h, int radius);
int DownSampling2X(unsigned char *src, unsigned char *dst, int width, int height);
int DownSampling4X(unsigned char *src, unsigned char *dst, int width, int height);
int DownSampling8X(unsigned char *src, unsigned char *dst, int width, int height);
void SplitUV(unsigned char *src, unsigned char *dst, int size);
void MergeUV(unsigned char *src, unsigned char *dst, int size);
void Average(unsigned char *src1, unsigned char *src2, unsigned char *dst, int size);

Test:

Test environment: Intel Core i3-6100 @ 3.70GHz

The input image has the resolution of 1920*1090 and only 1 channel. The function is executed in single thread.

Function	Time
FastBlur	995 us
DownSampling2X	241 us
DownSampling4X	102 us
DownSampling8X	92 us
SplitUV	50 us
MergeUV	40 us
Average	235 us

Part of the algorithm and code refer to Imageshop (http://www.cnblogs.com/Imageshop/) , and use AVX/AVX2 instruction set to optimize the code.