To reduce the feature length of Local Binary Pattern (LBP) for content-based image retrieval, you can use uniform patterns. Here's a simple implementation:
function lbp_features = uniform_lbp(image, radius, neighbors)
% Convert image to grayscale if it's not already
if size(image, 3) == 3
image = rgb2gray(image);
end
image = double(image);
% Initialize output
lbp = zeros(size(image));
% Calculate weights
weights = 2.^(0:neighbors-1);
% Generate circular neighbor coordinates
[y, x] = meshgrid(-radius:radius, -radius:radius);
angle = atan2(y, x);
r = sqrt(x.^2 + y.^2);
circle_mask = (r <= radius+0.5) & (r >= radius-0.5);
angles = linspace(0, 2*pi, neighbors+1);
angles = angles(1:end-1);
% Compute LBP
for i = 1:neighbors
y_offset = round(-radius * sin(angles(i)));
x_offset = round(radius * cos(angles(i)));
neighbor = circshift(image, [y_offset, x_offset]);
lbp = lbp + (neighbor > image) .* weights(i);
end
% Uniform pattern mapping
uniform_map = zeros(256, 1);
uniform_patterns = 0;
for i = 0:255
pattern = bitget(i, 1:8);
transitions = sum(abs(diff([pattern(8), pattern, pattern(1)])));
if transitions <= 2
uniform_map(i+1) = uniform_patterns;
uniform_patterns = uniform_patterns + 1;
else
uniform_map(i+1) = neighbors + 1;
end
end
% Apply uniform mapping
lbp = uniform_map(lbp + 1);
% Compute histogram
lbp_features = histcounts(lbp, 0:(neighbors+2));
% Normalize histogram
lbp_features = lbp_features / sum(lbp_features);
end
To use the above function use the following code :
% Load an image (replace with your image filename)
image = imread('input_image.jpg');
% Extract LBP features
radius = 1;
neighbors = 8;
lbp_features = uniform_lbp(image, radius, neighbors);
% Display the features
figure;
bar(lbp_features);
xlabel('LBP Uniform Pattern');
ylabel('Normalized Frequency');
title('Uniform LBP Histogram');
Following is the output generated when the functions are invoked with different inputs.
Please refer to the following related documentation that to help you further :
Hope this helps. Thanks.
