get_features.m

function x = get_features(im, features, cell_size, cos_window,w2c)
%GET_FEATURES
%   Extracts dense features from image.
%
%   X = GET_FEATURES(IM, FEATURES, CELL_SIZE)
%   Extracts features specified in struct FEATURES, from image IM. The
%   features should be densely sampled, in cells or intervals of CELL_SIZE.
%   The output has size [height in cells, width in cells, features].
%
%   To specify HOG features, set field 'hog' to true, and
%   'hog_orientations' to the number of bins.
%
%   To experiment with other features simply add them to this function
%   and include any needed parameters in the FEATURES struct. To allow
%   combinations of features, stack them with x = cat(3, x, new_feat).
%
%   Joao F. Henriques, 2014
%   http://www.isr.uc.pt/~henriques/
%
%   revised by: Yang Li, August, 2014
%   http://ihpdep.github.io

    if ~isstruct(features)
        if strcmp(features,'hog')
            feature.hog = true;
            feature.hogcolor = false;
            feature.gray = false;
            feature.hog_orientations = 9;
        elseif strcmp(features,'hogcolor')
            feature.hog = false;
            feature.hogcolor = true;
            feature.gray = false;
            feature.hog_orientations = 9;
        elseif strcmp(features,'gray')
            feature.hog = false;
            feature.hogcolor = false;
            feature.gray = true;
        end
        features = feature;
    end
            
	if features.hog
        %HOG features, from Piotr's Toolbox
        x = double(fhog(single(im) / 255, cell_size, features.hog_orientations));
        x(:,:,end) = [];  %remove all-zeros channel ("truncation feature")
    end
	
    if features.hogcolor
        %HOG features, from Piotr's Toolbox
        x = double(fhog(single(im) / 255, cell_size, features.hog_orientations));
        x(:,:,end) = [];  %remove all-zeros channel ("truncation feature")
        sz = size(x);
        im_patch = imresize(im, [sz(1) sz(2)]);
        out_npca = get_feature_map(im_patch, 'gray', w2c);
        out_pca = get_feature_map(im_patch, 'cn', w2c);
        % 		out_pca = reshape(temp_pca, [prod(sz), size(temp_pca, 3)]);
        x = cat(3,x,out_npca);
        x = cat(3,x,out_pca);
	end
	
	if features.gray
		%gray-level (scalar feature)
		x = double(im) / 255;
		
		x = x - mean(x(:));
	end
	
	%process with cosine window if needed
	if ~isempty(cos_window)
		x = bsxfun(@times, x, cos_window);
	end
	
end