http://www.mathworks.com/matlabcentral/newsreader/view_thread/263453

%% Data

img = peaks(512);

map_img = linspace(min(img(:)),max(img(:)),256);

winx = 20;

winy = 20;

[ny nx] = size(img);

nbin = length(map_img);

%% histc

tic

for i = 1:ny-winy

ii = i+(0:winy);

for j = 1:nx-winx;

win = img(ii,j+(0:winx));

wh = histc(win(:),map_img);

% if i==200 && j==100; wh1=wh; end

end

end

toc % Elapsed time is 5.196251 seconds.

%% sliding accumarray

tic

[trash bin] = histc(img,map_img);

for i = 1:ny-winy

if i==1

b1 = bin(1:winy+1,:);

h = zeros(nbin, nx);

for j = 1:nx;

h(:,j) = accumarray(b1(:,j),1,[nbin 1]);

end

ioff = (0:nx-1)*nbin;

z = zeros(nbin,1);

else

ip = bin(i+winy,:)+ioff;

h(ip) = h(ip)+1;

im = bin(i-1,:)+ioff;

h(im) = h(im)-1;

end

csh = [z cumsum(h, 2)];

for j = 1:nx-winx

wh = csh(:,j+winx+1)-csh(:,j);

% if i==200 && j==100; wh2=wh; end

end

end

toc % Elapsed time is 0.705944 seconds.

% isequal(wh1, wh2

The paper 'Mikhail Sizintsev, Konstantinos G. Derpanis and Andrew Hogue. Histogram-based search: A comparative study' presents a multiscale histogram-based search algorithm 'the distributive histogram' that can be evaluated exhaustively in a fast and memory efficient manner.