Gabor filtering

This sample demonstrates the use of multiple Gabor filter convolutions to get Fractalius-like image effect.

Sources:

https://github.com/opencv/opencv/blob/3.2.0/samples/python/gabor_threads.py

load image

img = imread(fullfile(mexopencv.root(),'test','img001.jpg'));

build an array of Gabor filters with different orientations

ksize = 31;
thetas = 0:pi/16:pi-pi/16;
filters = cell(size(thetas));
for i=1:numel(thetas)
    kern = cv.getGaborKernel('KSize',[ksize ksize], 'Theta',thetas(i), ...
        'Sigma',4, 'Lambda',10, 'Gamma',0.5, 'Psi',0, 'KType','single');
    kern = kern / (1.5 * sum(kern(:)));
    filters{i} = kern;
end

visualize Gabor filters

if mexopencv.require('images')
    montage(cat(4,filters{:}), 'DisplayRange',[])
end

process images using Gabor filters and accumulate results (Note: you can use PARFOR instead of FOR to run this in parallel)

tic
res = zeros(size(img), class(img));
for i=1:numel(filters)
    fimg = cv.filter2D(img, filters{i});
    res = max(res, fimg);
end
toc

Elapsed time is 0.550728 seconds.

show result

figure, imshow(img), title('image')
figure, imshow(res), title('result')