aux_resize_3d.m

Standard model object recognition matlab code

function y = aux_resize_3d (x, new_size)

% FUNCTION y = aux_resize_3d (x, new_size)
% 
% This funciton resizes a 3d matrix "x", into a "new_size".  It is
% intended for the case when different scaled responses are to be
% combined (mixing scales).  Here the resizing is done with
% nearest-neighbor rule (ie. the values are replicated, not
% interpolated, to the nearest neighbor values).  Note that, for
% convenience, we assume the third dimension of x and y are to be
% the same, and we just resize the first two dimensions
% (corresponding to x and y in the image).  This should be used
% only when the prod(new_size) is not too big (due to memory
% shortage), and because of the for-loops this is to be used in the
% later layers (when the receptive field sizes are big or the x-y
% matrix sizes are small).

old_size = size(x);
y = zeros(new_size);
sf = new_size(1:2)./old_size(1:2); % scale factor

% For each index, find the range of new resized indices.  Then,
% use repmat to copy the values over.  This is more or less like
% imresize('nearest') function.
for i = 1:old_size(1)
  i_r = [round((i-1)*sf(1))+1:round(i*sf(1))];
  for j = 1:old_size(2)
    j_r = [round((j-1)*sf(2))+1:round(j*sf(2))];
    y(i_r,j_r,:) = repmat(x(i,j,:), [length(i_r) length(j_r) 1]);
  end
end