domainTransformFilter

The newest implimentation of the domainTransformFilter is moved to OpenCP. (https://github.com/norishigefukushima/OpenCP)

Introduction

The code is OpenCV implimentation of domain transform filter, which is fast edge keep filter for smoothing.
The code is parallelized by Intel TBB or OpenMP over c cv::ParallelLoopBody class in OpenCV, thus requirement of OpenCV version is 2.45 or later.
Also, the code is vectorized by SIMD SSE4.2.

Currentry, two demos, which are smoothing and detail enhancement demo, are called in the program.
Both demos are quited by 'q' or ESC key.

*Reference
Eduardo S. L. Gastal and Manuel M. Oliveira. "Domain Transform for Edge-Aware Image and Video Processing". ACM Transactions on Graphics. Volume 30 (2011), Number 4, Proceedings of SIGGRAPH 2011, Article 69. http://inf.ufrgs.br/~eslgastal/DomainTransform/

Usage

void domainTransformFilter(InputArray srcImage, OutputArray destImage, float sigma_r, float sigma_s, int maxiter, int norm, int convolutionType, int implementation);
srcImage: src image
destImage: dest image
sigma_s: sigma of space  
sigma_r: sigma of range[0:255]  
maxiter: number of maximam iteration  
norm: cost norm: L1 is default, and it is the same as original paper.
convolution type: switching Recursive Filtering, Normalized Convolution and Interpolated Convolution.
inplimentation: switching parallel implimentation (Recursive Filtering is carefully optimized).

//Joint filtering version. The 2nd argment of guideImage is added.
void domainTransformFilter(InputArray srcImage, InputArray guideImage, OutputArray destImage, float sigma_r, float sigma_s, int maxiter, int norm, int convolutionType, int implementation);

typedef enum
{
	DTF_L1=1,
	DTF_L2=2
}DTF_NORM;

typedef enum
{
	DTF_RF=0,//Recursive Filtering
	DTF_NC=1,//Normalized Convolution
	DTF_IC=2,//Interpolated Convolution

}DTF_METHOD;

typedef enum
{
	DTF_BGRA_SSE=0,
	DTF_BGRA_SSE_PARALLEL,
	DTF_SLOWEST
}DTF_IMPLEMENTATION;

Results

Exsample of test results on core i5-2450M 2.5GHz, Windows 7 64 bit with Visual Studio 2012 are as follows;

Computational time is about 20 ms for 505x757 RGB image.
Filtered image are as follows;

Input image

Smooth image

Detail enhanced image

---

This code is forked from https://github.com/tatsy/ImageProcessing .
pow operations are optimized by fmath.hpp (https://github.com/herumi/fmath) .