esSampleProtoMap.m

function [M_tMap, protoMap protoMap_raw normSal] = esSampleProtoMap(sMap,...
                                                          currFrame,...
                                                          nBestProto)
%esSampleProtoMap - Generates the patch map M(t)
%
% Synopsis
%          [protoMap protoMap_raw normSal] = esSampleProtoMap(sMap,currFrame, nBestProto)
%
% Description
%     In a first step generates the raw patch map by thresholding  the normalized salience map
%     so as to achieve 95% significance level for deciding whether the given saliency values are 
%     in the extreme tails
%     In a second step get the N_V best patches ranked through their size and returns the actual
%     M(t) map
%
%
% Inputs ([]s are optional)
%   (matrix) sMap         the salience map, 0/1 overlay representation
%   (matrix) currFrame    the current frame
%   (integer) nBestProto  the N_V most valuable patches
%   
%
% Outputs ([]s are optional)
%   (matrix) M_tMap       the patch map M(t)
%   (matrix) protoMap     the object layer representation of patch map M(t)
%   (matrix) protoMap_raw   the raw patch map
%   (matrix) normSal      the normalized salience map
%
%
%
% Requirements
%   Image Processing toolbox
%
% References
%   [1] G. Boccignone and M. Ferraro, Ecological Sampling of Gaze Shifts
%       IEEE Trans. Systems Man Cybernetics - Part B (to appear)                      
%   
%
% Author
%   Giuseppe Boccignone <Giuseppe.Boccignone(at)unimi.it>
%
%
% Changes
%   12/12/2012  First Edition
%

protoMap_raw= false(size(sMap)); % allocating space for the raw map

% Normalizing salience
normSal=sMap;
maxsal=max(max(normSal));
minsal=min(min(normSal));
normSal= (normSal-minsal)./(maxsal-minsal);
normSal=normSal*100;

% Salience thresholding
% Method 1: adaptive on mu
%mu=mean2(sal);
%stdv=std2(sal);
%ind=find(sal>=4*mu); %standard measure 3*mu
% Method 2: percentile based
ind=find(normSal >= prctile(normSal(:),90));
protoMap_raw(ind)=1;
protoMap_raw(ind)= currFrame(ind);


% Samples the N_V best patches 
openingwindow=7;
M_tMap = esSampleNbestProto(protoMap_raw,nBestProto,openingwindow);


protoMap= M_tMap;
fbw = M_tMap==0;
protoMap(fbw) = 1;
fbw = M_tMap==1;
protoMap(fbw) = 0;
end


% -------------------------------------------------
function M_tMap = esSampleNbestProto(protoMap_raw, nBest, win)
%esSampleNbestProto - Samples the N_V best patches ranked through their size and returns the actual
%                     M(t) map

% Use morphological operations
se = strel('disk',win);
protoMap_raw = imopen(protoMap_raw,se);
       
% Calculating 8 connected components via Rosenfeld and Pfaltz
[L,obj] = bwlabel(protoMap_raw,8);  
		
% Returns the foreground connected component in the binary image
% supplied that have the specified ranked size(s).        
maxL = max(L(:));					      % max number of connected components
h = hist(L(find(protoMap_raw)),[1:maxL]); % find indexes of labeled elements, by column scanning,  
                                          %    and compute occurence of labels from 1 to maxL
[sh,sr] = sort(-h);                       % sorting with respect to the number of occurrence  
                                          %    sh is the num occurrences (negative: -8   -6  -4  -2), 
                                          %    sr = corresponding labels
M_tMap = protoMap_raw & 0;
if nBest > maxL
     nBest = maxL;
end
for i=1:nBest    
    M_tMap = M_tMap | (L==sr(i)) ;      % returns the nBest by  dimension
end
end