templateAffineTransform

 templateAffineTranform - Affine transform of vertices in the template.

 tScale: change of log-scale for the whole template
 rScale: change of log-scale for row index (small change)
 cScale: change of log-scale for column index (small change)
 rotation: rotation of template
 inX, inY, inO, inS: selected position (relative to the origin), orientation and scale of elements
 nOri: number of quantized orientations inside 0 ~ pi

0001 function [outRow, outCol, outO, outS] =...
0002     templateAffineTransform(tScale, rScale,cScale,rotation,inRow,inCol,inO,inS,nOri)
0003 % templateAffineTranform - Affine transform of vertices in the template.
0004 %
0005 % tScale: change of log-scale for the whole template
0006 % rScale: change of log-scale for row index (small change)
0007 % cScale: change of log-scale for column index (small change)
0008 % rotation: rotation of template
0009 % inX, inY, inO, inS: selected position (relative to the origin), orientation and scale of elements
0010 % nOri: number of quantized orientations inside 0 ~ pi
0011 
0012 A1 = eye(3); A1(1,1) = cScale * (2^(tScale/2)); A1(2,2) = rScale * (2^(tScale/2));
0013 angle = rotation * pi/nOri;
0014 A2 = eye(3); A2(1,1) = cos(angle); A2(1,2) = sin(angle); A2(2,1) = -A2(1,2); A2(2,2) = A2(1,1);
0015 outRow = zeros(length(inRow),1);
0016 outCol = zeros(length(inCol),1);
0017 outO = zeros(length(inO),1);
0018 outS = zeros(length(inS),1);
0019 
0020 for iF = 1:length(inRow)
0021     pt = [inCol(iF), -inRow(iF), 1] * A1;
0022     pt = pt * A2;
0023     
0024     outCol(iF) = floor( .5 + pt(1) );
0025     outRow(iF) = floor( .5 - pt(2) );
0026     outO(iF) = inO(iF) + rotation;
0027     outS(iF) = inS(iF) + tScale;
0028     while outO(iF) < 0
0029         outO(iF) = outO(iF) + nOri;
0030     end
0031     while outO(iF) >= nOri
0032         outO(iF) = outO(iF) - nOri;
0033     end
0034 end
0035 
0036 
0037 % note: during squeezing, the orientations of some elements are changed. But
0038 %    we assume the squeezing effect is negligible on filter orientations.
0039 
0040 % Because of scaling, some Gabors are not applicable.
0041 % But we still keep all Gabor elements, and will filter out these elements when
0042 % computing S2 scores.
0043