findfaces.asv

eye detection sample in matlab. Eyetrackers can analyze a driver’s level of attentiveness while driv

function [pos,facesamples] = findfaces(im)
pic = getpic(double(im),1,1);
pos=[];% pos=[y x scale]
facesamples=[];
binpic = ones(size(pic));

% finds "facelike" pictures and saves their positions along with scale in pos

for i= [1.9 2.3 2.8 3.5 4.3 5.1 6 7 8]
    nom=10; den=10*i;
    cur_pic=getpic(pic,nom,den);
    cur_bin = getpic(binpic,nom,den);
    cur_bin = cur_bin>0.5;
    if (i<5)
    cur_bin(:,2:2:size(cur_bin,2))=0;
    cur_bin(2:2:size(cur_bin,1),:)=0;
    end;
    % if picture is too small go to bigger size (next loop)
    if ((size(cur_pic,1)<20)|(size(cur_pic,1)<20))
        continue;
    end
    
    % identifie "small face" with facepos]
    [p,s]=facepos(cur_pic,cur_bin);
    
    % choosing the most likly to be faces (we can fet positive results from facepos for images 2 pixels apart 
    % for example, will pick only one from them
    if (size(p,1) == 0)
        continue;
    end

    [p,s]=facelocation(p,s);
    facesamples = [facesamples s];
    for xx=1:size(p,1)
        starty = (p(xx,1)-10)*i;
        if (starty<1)  starty=1;end;
        endy = p(xx,1)*i+15*i;
        startx = (p(xx,2)-5)*i;
        if (startx<1) startx=1; end;
        endx = (p(xx,2)+15)*i;
      %  binpic(starty:endy,startx:endx)=0;
    end
    
    if size(p,1) > 0
        tmp = [p i*ones(size(p,1),1)];
        pos = [pos;tmp];
    end
end
% a face  can be recognized in several scales, order to get the picture in the "best" scale 
% we list them in a descending order by the SVM
posScore=[];
for j=1:size(pos,1)
    load FC;
    [l,sc] = osuSVMclass(facesamples(:,j),FC.n, FC.alpha, FC.sv,FC.b, FC.params);
    posScore=[posScore;sc];
    clear FC;
end
[dummy,newPos]=sort(posScore);
newPos=newPos(size(newPos,1):-1:1,:);
newPos=newPos';
pos=pos(newPos,:);
facesamples=facesamples(:,newPos);