Fast Hough transform demo

This program demonstrates line finding with the Fast Hough transform.

Sources:

Contents

Image

img = cv.imread(fullfile(mexopencv.root(),'test','building.jpg'), 'Grayscale',true);
assert(~isempty(img), 'Failed to load image');

scale = min(min([500 1000] ./ size(img)), 1.0);
rescale = @(im) cv.resize(im, scale, scale, 'Interpolation','Area');

imshow(rescale(img)), title('Image')

Edges

canny = cv.Canny(img, [50 200], 'ApertureSize',3);
imshow(rescale(canny)), title('Canny Edges')

Fast Hough Transform

houghDepth = 'int32';
houghAngleRange = 'ARO_315_135';
houghOperator = 'Average';
houghSkew = 'Deskew';

tic
hough = cv.FastHoughTransform(canny, 'DDepth',houghDepth, ...
    'AngleRange',houghAngleRange, 'Op',houghOperator, 'MakeSkew',houghSkew);
toc

Elapsed time is 0.061526 seconds.

Find local maximums on FHT image

minWeight = 255 * 0.3 * min(size(img));
maxCount = 50;

if true
    points = houghpeaks(double(hough), maxCount, 'Threshold',minWeight);
else
    localExtr = imextendedmax(hough, minWeight);  % local extremum (logical)
    [rows,cols,vals] = find(immultiply(hough, localExtr));
    [~,idx] = sort(vals, 'descend');  % sort by strongest hough values
    idx(maxCount+1:end) = [];         % keep best N
    points = [rows(idx) cols(idx)];
end

Build line segments

lines = zeros(size(points,1), 4, 'int32');
for i=1:size(points,1)
    lines(i,:) = cv.HoughPoint2Line(points(i,[2 1])-1, canny, ...
        'AngleRange',houghAngleRange, 'MakeSkew',houghSkew);
end

Show FHT

imagesc(hough)
line(points(:,2), points(:,1), 'Color','r', 'LineStyle','none', 'Marker','s')
title('Fast Hough Transform')

Show lines

out = cv.cvtColor(canny, 'GRAY2RGB');
for i=1:size(lines,1)
    out = cv.line(out, lines(i,[1 2]), lines(i,[3 4]), ...
        'Color',[255 0 0], 'Thickness',1, 'LineType','AA');
end
imshow(rescale(out)), title('Lines')