Skip to content

Sundrops/fast-guided-filter

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

28 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Fast Guided filter for OpenCV

Fast Guided filter is an edge-preserving smoothing filter like the bilateral filter. It is straightforward to implement and has linear complexity independent of the kernel size. For more details about this filter see[Guided filte] [Fast Guided filte].

Usage

The interface consists of one simple function fastGuidedFilter and a class FastGuidedFilter. If you have multiple images to filter with the same guidance image then use FastGuidedFilter class to avoid extra computations on initialization stage. The code supports single-channel and 3-channel (color) guidance images and CV_8U, CV_8S, CV_16U, CV_16S, CV_32S, CV_32F and CV_64F data types.

$ mkdir build && cd build
$ cmake ..
$ make
$ ./fast-guided-filter

Examples

These examples are adapted from the original MATLAB implementation.

Smoothing

P: GrayScale I: GrayScale

int R[]={2,4,8};
double EPS[]={0.1,0.2,0.4};
Mat result;
for(int s=1;s<=2;++s) {
	for (int i = 0; i < 3; ++i) {
		for (int j = 0; j < 3; ++j) {
			Mat P =  cv::imread("./imgs/cat.png", CV_LOAD_IMAGE_GRAYSCALE);
			Mat I;
			I = P;
			int r = R[i];
			float eps = EPS[j]*EPS[j];
			eps *= 255 * 255;
			clock_t start_time1 = clock();
			result = fastGuidedFilter(I, P, r, eps, s);
			clock_t end_time1 = clock();
			cout << "fastguidedfilter Running time is: "
			     << static_cast<double>(end_time1 - start_time1) / CLOCKS_PER_SEC * 1000 << "ms" << endl;
			string name = "result_s:" + to_string(s) + "_r:" + to_string(r) + "_eps:" + to_string(EPS[j]) + "^2.png";
			imwrite(name, result);
		}
	}
}

 

Cat

  • r=2,eps=0.1^2 -0.4^2,s=1 time: 1.7ms

r=2, eps=0.1^2 r=2, eps=0.2^2 r=2, eps=0.4^2

  • r=2,eps=0.1^2 -0.4^2,s=2 time: 0.6ms

r=2, eps=0.1^2 r=2, eps=0.2^2 r=2, eps=0.4^2

  • r=4,eps=0.1^2 -0.4^2,s=1 time: 1.7ms

r=4, eps=0.1^2 r=4, eps=0.2^2 r=4, eps=0.4^2

  • r=4,eps=0.1^2 -0.4^2,s=2 time: 0.6ms

r=4, eps=0.1^2 r=4, eps=0.2^2 r=4, eps=0.4^2

  • r=8,eps=0.1^2 -0.4^2,s=1 time: 1.7ms

r=8, eps=0.1^2 r=8, eps=0.2^2 r=8, eps=0.4^2

  • r=8,eps=0.1^2 -0.4^2,s=2 time: 0.6ms

r=8, eps=0.1^2 r=8, eps=0.2^2 r=8, eps=0.4^2

P: RGB(3 channels) I: GrayScale or RGB

int R[]={2,4,8};
double EPS[]={0.1,0.2,0.4};
Mat result;
for(int s=1;s<=2;++s) {
	for (int i = 0; i < 3; ++i) {
		for (int j = 0; j < 3; ++j) {
			Mat P = imread("../imgs/people.png", CV_LOAD_IMAGE_ANYCOLOR);
			Mat I;
			//cvtColor(P,I,CV_BGR2GRAY);
			I = P;
			int r = R[i];
			float eps = EPS[j]*EPS[j];
			eps *= 255 * 255;
			clock_t start_time1 = clock();
			result = fastGuidedFilter(I, P, r, eps, s);
			clock_t end_time1 = clock();
			cout << "fastguidedfilter Running time is: "
			     << static_cast<double>(end_time1 - start_time1) / CLOCKS_PER_SEC * 1000 << "ms" << endl;
			string name = "I:color_result_s:" + to_string(s) + "_r:" + to_string(r) + "_eps:" + to_string(EPS[j]) + "^2.png";
			imwrite(name, result);
		}
	}
}

People

  • I=grayscale r=2,eps=0.1^2 -0.4^2,s=1 time: 6ms

I=grayscale r=2, eps=0.1^2 I=grayscaler=2, eps=0.2^2 I=grayscaler=2, eps=0.4^2

  • I=RGB r=2,eps=0.1^2 -0.4^2,s=1 time: 16ms

I=RGB r=2, eps=0.1^2 I=RGB r=2, eps=0.2^2 I=RGB r=2, eps=0.4^2

  • I=RGB r=2,eps=0.1^2 -0.4^2,s=2 time: 6ms

I=RGB r=2, eps=0.1^2 I=RGB r=2, eps=0.2^2 I=RGB r=2, eps=0.4^2

  • I=grayscale r=4,eps=0.1^2 -0.4^2,s=1 time: 6ms

I=grayscale r=4, eps=0.1^2 I=grayscaler=4, eps=0.2^2 I=grayscaler=4, eps=0.4^2

  • I=RGB r=4,eps=0.1^2 -0.4^2,s=1 time: 16ms

I=RGB r=4, eps=0.1^2 I=RGB r=4, eps=0.2^2 I=RGB r=4, eps=0.4^2

  • I=RGB r=4,eps=0.1^2 -0.4^2,s=2 time: 6ms

I=RGB r=4, eps=0.1^2 I=RGB r=4, eps=0.2^2 I=RGB r=4, eps=0.4^2

I=grayscale r=8,eps=0.1^2 -0.4^2,s=1 time: 6ms

I=grayscale r=8, eps=0.1^2 I=grayscaler=8, eps=0.2^2 I=grayscaler=8, eps=0.4^2

  • I=RGB r=8,eps=0.1^2 -0.4^2,s=1 time: 16ms

I=RGB r=8, eps=0.1^2 I=RGB r=8, eps=0.2^2 I=RGB r=8, eps=0.4^2

  • I=RGB r=8,eps=0.1^2 -0.4^2,s=2 time: 6ms

I=RGB r=8, eps=0.1^2 I=RGB r=8, eps=0.2^2 I=RGB r=8, eps=0.4^2

Flash/no-flash denoising

cv::Mat I = cv::imread("./img_flash/cave-flash.bmp", CV_LOAD_IMAGE_COLOR);
cv::Mat p = cv::imread("./img_flash/cave-noflash.bmp", CV_LOAD_IMAGE_COLOR);

int r = 8;
double eps = 0.02 * 0.02;

eps *= 255 * 255;   // Because the intensity range of our images is [0, 255]

cv::Mat q = guidedFilter(I, p, r, eps);

Cave Flash Cave No Flash Cave Denoised

Feathering

cv::Mat I = cv::imread("./img_feathering/toy.bmp", CV_LOAD_IMAGE_COLOR);
cv::Mat p = cv::imread("./img_feathering/toy-mask.bmp", CV_LOAD_IMAGE_GRAYSCALE);

int r = 60;
double eps = 1e-6;

eps *= 255 * 255;   // Because the intensity range of our images is [0, 255]

cv::Mat q = guidedFilter(I, p, r, eps);

Mask Guidance Feathering

Enhancement

cv::Mat I = cv::imread("./img_enhancement/tulips.bmp", CV_LOAD_IMAGE_COLOR);
I.convertTo(I, CV_32F, 1.0 / 255.0);

cv::Mat p = I;

int r = 16;
double eps = 0.1 * 0.1;

cv::Mat q = guidedFilter(I, p, r, eps);

cv::Mat I_enhanced = (I - q) * 5 + q;

Tulip Smoothed Enhanced

License

MIT license.

About

fast guided filter implementation for OpenCV

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C++ 95.1%
  • CMake 4.9%