function [wI,wIx,wIy] = vl_imwbackward(varargin)
% VL_IMWBACKWARD  Image backward warping
%   J = VL_IMWBACKWARD(I, X, Y) returns the values of image I at
%   locations X,Y. X and Y are real matrices of arbitrary but
%   identical dimensions. I is bilinearly interpolated between samples
%   and extended with NaNs to the whole real plane.
%
%   [J,JX,JY] = VL_IMWBACKWARD(...) returns the warped derivatives JX and
%   JY too.
%
%   By default, VL_IMWBACKWARD() assumes that the image I uses the standard
%   coordinate system. VL_IMWBACKWARD(XR,YR,I,X,Y) assumes instead that I
%   is defined on a rectangular grid specified by the vectors XR and
%   YR.
%
%   VL_IMWBACKWARD() is less general than the MATLAB native function
%   INTERP2(), but it is significantly faster.
%
%   See also: IMWFORWARD(), INTERP2(), VL_HELP().

% Copyright (C) 2007-12 Andrea Vedaldi and Brian Fulkerson.
% All rights reserved.
%
% This file is part of the VLFeat library and is made available under
% the terms of the BSD license (see the COPYING file).

if nargin < 5
  I = varargin{1} ;
  [M,N,K] = size(I) ;
  xr = 1:N ;
  yr = 1:M ;
  varargin = { varargin{2:end} } ;
else
  xr = varargin{1} ;
  yr = varargin{2} ;
  I  = varargin{3} ;
  [M,N,K] = size(I) ;
  varargin = { varargin{4:end} } ;
end

if K == 1
	if nargout == 1
		wI = vl_imwbackwardmx(xr, yr, I, varargin{:}) ;
	else
		[wI,wIx,wIy] = vl_imwbackwardmx(xr, yr, I, varargin{:}) ;
	end
else
  [M,N] = size(varargin{1}) ;
	if nargout == 1
		wI = zeros(M,N,K) ;
		for k=1:K
			wI(:,:,k) = vl_imwbackwardmx(xr, yr, squeeze(I(:,:,k)),  varargin{:}) ;
		end
	else
		wI  = zeros(M,N,K) ;
		wIx = zeros(M,N,K) ;
		wIy = zeros(M,N,K) ;
		for k=1:K
			[tmp1, tmp2, tmp3] = vl_imwbackwardmx(xr, yr, squeeze(I(:,:,k)),  varargin{:}) ;
			wI(:,:,k)  = tmp1;
			wIx(:,:,k) = tmp2;
			wIy(:,:,k) = tmp3 ;
		end
	end
end