You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
77 lines
1.6 KiB
Matlab
77 lines
1.6 KiB
Matlab
function [X]=preprocess_V1_input(I,X)
|
|
GP=global_parameters;
|
|
|
|
[a,b,z]=size(I);
|
|
I=single(I);
|
|
|
|
if nargin<2 || isempty(X)
|
|
nChannelsInitial=0;
|
|
else
|
|
nChannelsInitial=length(X)
|
|
end
|
|
k=0;
|
|
for sigma=GP.LGNsigma
|
|
k=k+1;
|
|
LoG=filter_definitions_LGN(sigma);
|
|
|
|
for t=1:z %at each time step
|
|
%calculate LGN neuron responses to input image
|
|
Xonoff=conv2(I(:,:,t),LoG,'same');
|
|
|
|
%the edge of the image often contains strong discontinuities, so don't calculate
|
|
%responses here. Would be better the use the 'valid' paramenter in the
|
|
%convolution, but this produces problems when image is being analysed at
|
|
%different scales.
|
|
Xonoff=suppress_response_near_edges(Xonoff,odd(2.5*sigma));
|
|
|
|
%apply gain to response
|
|
Xonoff=GP.LGNresponseGain.*Xonoff;
|
|
|
|
%apply saturation to response
|
|
Xonoff=tanh(Xonoff);
|
|
|
|
%split into ON and OFF channels
|
|
Xon=Xonoff;
|
|
Xon(find(Xon<0))=0;
|
|
Xoff=-Xonoff;
|
|
Xoff(find(Xoff<0))=0;
|
|
|
|
X{nChannelsInitial+2*k-1}(:,:,t)=Xon;
|
|
X{nChannelsInitial+2*k}(:,:,t)=Xoff;
|
|
end
|
|
end
|
|
|
|
|
|
function y=suppress_response_near_edges(y,crop)
|
|
|
|
if ~iscell(y)
|
|
y=suppress_edges(y,crop);
|
|
else
|
|
[nMasks]=length(y);
|
|
for i=1:nMasks
|
|
y{i}=suppress_edges(y{i},crop);
|
|
end
|
|
end
|
|
|
|
function y=suppress_edges(y,crop)
|
|
[a,b]=size(y);
|
|
|
|
y(1:crop,:)=0;
|
|
y(:,1:crop)=0;
|
|
y(a-crop+1:a,:)=0;
|
|
y(:,b-crop+1:b)=0;
|
|
|
|
if 0
|
|
%linear reduction in response to zero at edges
|
|
for c=1:crop
|
|
scale=(c/(crop+1)).^2;
|
|
if c<=a
|
|
y(c,:)=y(c,:).*scale;
|
|
y(a-c+1,:)=y(a-c+1,:).*scale;
|
|
end
|
|
if c<=b
|
|
y(:,c)=y(:,c).*scale;
|
|
y(:,b-c+1)=y(:,b-c+1).*scale;
|
|
end
|
|
end
|
|
end |