function out=randpoisson(inarray,thresh);
% function out=randpoisson(inarray,thresh);
% outputs an array of poisson-distributed numbers with mean equal to inarray
% For inarray values above the threshold thresh (default=32),
%  use a quick-and-dirty version of the gaussian method,
%  but with negatives clipped to zero
% J.R. Fienup 10/22/99


if nargin <  1, 
    error('Requires at least one input argument.'); 
end

if exist('thresh')~=1, thresh=32; end

out=inarray;
% High-count pixels - use Gaussian approach
gtthresh=find(inarray>thresh);
if ~isempty(gtthresh),
	out(gtthresh)=inarray(gtthresh) + sqrt(inarray(gtthresh)).*randn(size(inarray(gtthresh)));
	out(gtthresh)=round(max(0,out(gtthresh)));
end
% Low-count pixels - this goes into the counting-experiment method

ltthresh=find(inarray<=thresh);
if ~isempty(ltthresh)
	lamda=inarray(ltthresh); % segregate low-value pixels to speed computation
	% Now dealing with a 1-D column vector that will merge into n-D array out later on
	%Initialize r to zero.
	r = zeros(size(lamda));  % output array for ltthresh pixels
	p = zeros(size(lamda));
	ind = true(size(lamda));
	    
    while any(ind)
	    p(ind) = p(ind) - log(rand(length(ind),1));  % note, do repeatedly calculate over all of lamda
	    ind = find(p < lamda); % Q: does this k index over 
        r(ind) = r(ind) + 1;
    end

	    
	% Return NaN if lamda is not positive -- to do this, un-comment what follows (gives zero now).
	
% 	tmp = NaN;
% 	if any(any(any(lamda <= 0)));
% 	    if prod(size(lamda) == 1),   % i.e., a single pixel?
% 	        r = tmp(ones(size(lamda)));
% 	    else
% 	        k = find(lamda <= 0);
% 	        r(k) = tmp(ones(size(k)));
% 	    end
% 	end

	out(ltthresh)=r;  % Merge low-value-pixel results with large-value-pixel results
	
end; % of if length(ltthresh)>0

end