the functions to generate density maps

Author: shepnerd

genDensityMap.m

@@ -0,0 +1,17 @@
+function [ density ] = genDensityMap( loc_map, hsize, sigma)
+% loc_map : an image marks the locations of object, where 0 means none and
+% 1 (or > 1) means the location.
+% hsize : the influential range of the guassian kernel, usually 4.0*6
+% sigma : the standard error of the guassian kernel, usually 4.0
+% density : the density map of the given location map. Usually, the
+% integral of the whole density map is equal to or smaller than that of the
+% location map.
+if size(loc_map,3) == 3
+ loc_map = rgb2gray(loc_map);
+end
+m = double(max(loc_map(:)));
+loc_map = double(loc_map)./m;
+density = imfilter(loc_map, fspecial('gaussian', hsize, sigma));
+
+end
+

geometry_adaptive_kernel.m

@@ -0,0 +1,49 @@
+function [ density ] = geometry_adaptive_kernel( coordinates, img_sz, beta )
+% author: sheperd, wygamle@pku.edu.cn
+
+if ~exist('beta','var')
+ beta = 0.3;
+end
+hei = img_sz(1); wid = img_sz(2);
+coordinates = floor(abs(coordinates));
+
+dist2 = @(i,j)((coordinates(i,1)-coordinates(j,1)).^2+(coordinates(i,2)-coordinates(j,2)).^2);
+dist = @(i,j)(dist2(i,j).^0.5);
+n = size(coordinates,1);
+matDist = zeros(n,n);
+for i = 1 : n
+ for j = i+1 : n
+ matDist(i,j) = dist(i,j);
+ end
+end
+matDist = matDist + matDist';
+
+coordinates = int32(coordinates);
+
+density = zeros(hei,wid);
+k = 5;
+for i = 1 : n
+ delta = zeros(hei,wid);
+ r = max(min(coordinates(i,2),hei),1);
+ c = max(min(coordinates(i,1),wid),1);
+ delta(r,c) = 1;
+ 
+ dv = matDist(i,:);
+ dv(i) = inf;
+ 
+ d = 0;
+ for j = 1 : k
+ [tmp, idx] = min(dv);
+ d = d + tmp;
+ dv(idx) = inf;
+ end
+ d = d ./ k;
+ sigma = beta*d;
+ hsize = max(floor(4 * sigma),3);
+ f = fspecial('gaussian',hsize,sigma);
+ delta = imfilter(delta, f);
+ density = density + delta;
+end
+
+end
+