Initial commit

.gitattributes

@@ -0,0 +1,2 @@
+# Auto detect text files and perform LF normalization
+* text=auto

CiVol.m

@@ -0,0 +1 @@
+function [Volumes, N] = CiVol(mask, D)% CIVOL Intrinsic volumes of the search region in pixel.%	CiVol(mask, D) returns the intrinsic Volumes (N=Volumes(D))%	If mask is a scalar, it specifies the number of pixels in the search region% 		E.g., [Volumes, N] = CiVol(pi*128^2, 2);%	If mask is a binary matrix, where the 1's define the search region% 		E.g.,% 		mask = double(imread('faceMask.tif'));% 		mask = (mask - min(mask(:))) / (max(mask(:)) - min(mask(:)));% 		[Volumes, N] = CiVol(mask, length(size(mask)));%	D (D<=3) is the dimensionality of the search region% % From K. J. Worsley's ([email protected]) stat_threshold help section:%	Volumes is a vector of D+1 components, the intrinsic volumes of the search %	region = [Euler characteristic, 2*caliper diameter, 1/2 surface area, volume],%	e.g. [1, 4*r, 2*pi*r^2, 4/3*pi*r^3] for a sphere of radius r in 3D. For a 2D %	search region, it is [1, 1/2 perimeter length, area]. The random field theory %	threshold is based on the assumption that the search region is a sphere, which %	is a very tight lower bound for any non-spherical region.% % See also DEMOSTAT4CI% % The Stat4Ci toolbox is free (http://mapageweb.umontreal.ca/gosselif/stat4ci.html); if you use % it in your research, please, cite us:%	Chauvin, A., Worsley, K. J., Schyns, P. G., Arguin, M. & Gosselin, F. (2004).  A sensitive %	statistical test for smooth classification images.% % Alan Chauvin & FrŽdŽric Gosselin ([email protected]), 20/08/2004if length(mask) == 1,		N = mask;else	% the mask have to be 1 in the region of interest and 0 elsewhere	minMask = min(mask(:));	maxMask = max(mask(:));		N = sum( (mask(:) - minMask)./  (maxMask - minMask) );	end;EC = 1;switch Dcase 1	Volumes(1) = EC; 			% Euler characteristic			Volumes(2) = N;			% length	case 2		% N = pi*r^2	r = sqrt(N/pi);	Volumes(1) = EC; 			% Euler characteristic		Volumes(2) = pi * sqrt(N/pi)	% caliper diameter	Volumes(2) = pi * r	% caliper diameter		Volumes(3) = N;			% surface area	case 3	% N = (4/3)*pi*r^3	r = (3/4/pi*N)^(1/3);	% r is the radius of the sphere		Volumes(1) = EC; 			% Euler characteristic		Volumes(2) =  4 * r;		% 2 * caliper diameter		Volumes(3) = 2 * pi * r;	% 1/2 surface area			Volumes(4) = N;				% volumeotherwise	error('Not implemented for D > 3.')end;

DemoStat4Ci.m

@@ -0,0 +1 @@
+% DEMOSTAT4CI illustrates the usage of the Stat2Ci toolbox on two classification images from:%		Gosselin, F. & Schyns, P. G. (2001).  Bubbles: A technique to reveal the use of information % 		in recognition. Vision Research, 41, 2261-2271.% 	The Stat4Ci toolbox allows to perform the Pixel and the Cluster tests, both based on Random Field % 	Theory. These tests are easy to apply, requiring a mere four pieces of information; and they % 	typically produce statistical thresholds (or p-values) lower than the standard Bonferroni correction.% %	An excellent non-technical reference is:% 		K. J. Worsley (1996) the geometry of random image. Chance, 9, 27-40.% %	We borrowed from several sources: the STAT_THRESHOLD function was written by Keith Worsley for the fmristat toolbox% 	(http://www.math.mcgill.ca/~keith/fmristat); and our DISPLAYCI function calls many functions from the %	Image Processing toolbox.% % The Stat4Ci toolbox is free (http://mapageweb.umontreal.ca/gosselif/stat4ci.html); if you use % it in your research, please, cite us:%	Chauvin, A., Worsley, K. J., Schyns, P. G., Arguin, M. & Gosselin, F. (2004).  A sensitive %	statistical test for smooth classification images.% % Alan Chauvin & FrŽdŽric Gosselin ([email protected]), 20/08/2004% % Loads a classification imageSCi = double(imread('GenderFG.tiff')); % from Gosselin & Schyns, 2001, Experiment 1, GENDER, subject FG SCi = double(imread('exnexFG.tiff')); % from Gosselin & Schyns, 2001, Experiment 1, GENDER, subject FG SCi = cIm2hrCi(SCi);sigma_b = 20;	%std of smoothing filter% % Specifies a search space; here: the face areaS_r = double(imread('faceMask.tiff'));S_r = (S_r - min(S_r(:))) / (max(S_r(:))-min(S_r(:)));% % Extracts the area outside the mask; used for the estimation of expected mean and variance of the CivecSCi = SCi(eq(S_r,0));% % Z-transforms the SCiRes = ZTransSCi(SCi,mean(vecSCi(:)),std(vecSCi(:)));Res.ZSCi = Res.ZSCi.*S_r;% % Performs the Pixel or Cluster testp = .05;	%p-valuetC = 2.7; 	%threshold (for Cluster test)Res = StatThresh(Res,p,sigma_b,tC,S_r);% % Displays the resultsbackground = double(imread('w1H.JPG'));  % a face used by Gosselin & Schyns, 2001, Experiment 1tCi = DiplayRes(Res,background); 		% the results of one of the tests with a background	

DiplayRes.m

@@ -0,0 +1 @@
+function Res = DiplayRes(Res,background); 	% DISPLAYCI displays the results of the Pixel and Cluster tests.%	tCi = DisplayCi(Res,[background]) returns, for the Pixel test % 	(typeTest = 'P'), the Z-scores (tCi) above threshold (tP); and, for the Cluster test % 	(typeTest = 'C'), the Z-scores (tCi) above threshold (tC) and belonging to clusters % 	greater than the minimum size (k).% %		Res.ZCi	= Z-transformed smooth classification image (see ZTRANSCI) % 		Res.tC = Cluster test threshold (set by the user)% 		Res.k = Minimun cluster size (see STAT_THRESHOLD)% 		Res.tP = Pixel test threshold (see STAT_THRESHOLD)% 		Res.FWHM = full width half maximum of the Gaussian smoothing filter (see  HALFMAX)% 		Res.p = p-value for Pixel and Cluster tests (set by user)% 		[background] = image overlaid to the classification image to help interpretation.% 			With red pixels indicating the regions that attained statistical significance. % % 			E.g.,% 			SCi = double(imread('GenderCi.tiff'));  % 			sCi = cIm2hrCi(SCi); % 			sigma_b = 20;	% 			S_r = double(imread('faceMask.tif'));% 			vecSCi = SCi(eq(S_r,0));% 			ZSCi = ZTransSCi(SCi,[mean(vecSCi(:)),std(vecSCi(:))]);% 			ZSCi = ZSCi.*S_r;% 			P = .05;	% 			tC = 2.7; 	% 			Res = StatThresh(ZSCi,p,sigma_b,tC,S_r);% 			background = double(imread('w1H.JPG'));  % 			tCi = DiplayRes(Res,background); 	% %	For both tests, DISPLAYCI also returns a table with all relevant statistics (tC, tP, % 	p, sizes of significant clusters, and so on).%% Requires the ImageProcessing Toolbox.% % See also STATRESH, ZTRANSCI, DEMOSTAT4CI% % The Stat4Ci toolbox is free (http://mapageweb.umontreal.ca/gosselif/stat4ci.html); if you use % it in your research, please, cite us:%	Chauvin, A., Worsley, K. J., Schyns, P. G., Arguin, M. & Gosselin, F. (2005).  Accurate %	statistical tests for smooth classification images.% % Alan Chauvin & FrŽdŽric Gosselin ([email protected]), 01//03/2005% ------------------------------------------------------------% 1. displays cluster result    cCi = ( Res.ZSCi >= Res.tC);[LabelClC ,nbClusterC] = BWLABEL(cCi);    % labels all clusters in Res.ZSCiif nbClusterC ~=0        for ii = 1 : nbClusterC        cm(ii) = length(find(LabelClC == ii));	end;	%dbstop at 65    cm = [[1:nbClusterC];cm];    if nbClusterC > 1		%cm = sortrows(cm,2);       % sorts clusters per size		cm = sortrows(cm',1)';       % sorts clusters per size	end;    ii = 0;c= 0;    while ii < nbClusterC        ii = ii + 1;        if cm(2,ii) < Res.k            LabelClC(find(LabelClC == cm(1,ii))) = 0;		else            c = c +1;            maxZ(c) = max(max(Res.ZSCi(find(LabelClC == cm(1,ii)))));            sizeZ(c) = cm(2,ii);            [I,J]=find(LabelClC == cm(1,ii));            yZ(c) = round((max(I)+min(I))/2);            xZ(c) = round((max(J)+min(J))/2);            LabelClC(find(LabelClC == cm(1,ii))) = 1;		end;	end;    for i = 1 : c        Results(i,2) = Res.tC;        Results(i,3) = sizeZ(i);        Results(i,4) = sizeZ(i)/Res.FWHM/Res.FWHM';        Results(i,5) = maxZ(i);	        Results(i,6) = xZ(i);		        Results(i,7) = yZ(i);		if i == c					Results(i,1) = 67; % ascii code for 'C'		else	        Results(i,1) = 32; % ascii code for 'space'		end;	end;	if c, clear maxZ sizeZ yZ xZ;end;					if c > 0		Results = flipud(sortrows(Results,3));					fprintf('\n');		fprintf('\tt\t\tsize\tresels\tZmax\tx\ty\n');		fprintf('\t-----------------------------------------\n');		fprintf('%c\t[%.2f]\t%d\t\t%.2f\t%.2f\t%d\t%d\n',Results');			else		fprintf('\t-----------------------------------------\n');		fprintf('\tno cluster reach significance\n')			end;else	fprintf('\t-----------------------------------------\n');	fprintf('\tno cluster reach significance\n')end;% --------------------------------------------------------% 2. displays pixel resultspCi = ( Res.ZSCi >= Res.tP);nbClusterP = ( sum(pCi(:)) >= 1 );if nbClusterP	fprintf('\t-----------------------------------------\n');	fprintf('P\t%.2f\t-\t-\t-\t-\t-\n',Res.tP);end;fprintf('\n');fprintf('p-value = [%.2f]\n',Res.p);fprintf('FWHM  == [%.1f]\n',Res.FWHM);fprintf('Minimum cluster size = %.1f\n',max(Res.k,0));% --------------------------------------------------------% 3. displays the figureif nbClusterC == 0 & nbClusterP == 0;	Res.tCi = zeros(Res>ZSci);	return;end;figurefor ii = 1 : 2	if ii == 1		% 3.1 Pixel test		t = Res.tP;		LabelCl = ones(size(Res.ZSCi));		Res.PixTest = (Res.ZSCi >= t) .* LabelCl;		tCi = Res.PixTest;		ss = sprintf('Pixel Test');	else		% 3.1 Cluster  test		t = Res.tC;		LabelCl = LabelClC;		Res.ClusTest = (Res.ZSCi >= t) .* LabelCl;		tCi = Res.ClusTest;		ss = sprintf('Cluster Test');	end;	subplot(1,2,ii)		if ~exist('background')  				image(tCi);axis image;axis off		s = sprintf('t = %.2f, pval = %.3f',t,p);	title(s)			else	% thresholded classification image is overlaid on an image		background = 2*( ( background-min(background(:)) ) / ( max(background(:))-min(background(:)) ) ) - 1;				for ii = 1 : 3			colorIma(:,:,ii) = 2*background;  		end;			mmax = max(Res.ZSCi(:));		mmin = min(Res.ZSCi(:));				cIma = colorIma/2;     cIma(:,:,1) = colorIma(:,:,1)/2 + tCi;				cIma = ( cIma-min(cIma(:)) ) / ( max(cIma(:))-min(cIma(:))+0.0001 );				tCi = cIma;				% figure,imshow(cIma);axis on		imagesc(cIma); axis square;colormap(jet);axis off		s = sprintf('%s, t = %.2f, pval = %.3f',ss, t,Res.p);	title(s)		if ii == 1			% 3.1 Pixel test			Res.PixTest = cIma;		else			% 3.1 Cluster  test			Res.ClusTest = cIma;		end;		end;end;clear cm

HalfMax.m

@@ -0,0 +1 @@
+function FWHM = HalfMax(sigma);% HALFMAX Full width half maximum of a Gaussian filter.%	HalfMax returns FWHM, the full width half maximum of%	a Gaussian filter of standard deviation equals to sigma pixels.%		E.g., FWHM = HalfMax(20);% % See also DEMOSTAT4CI% % The Stat4Ci toolbox is free (http://mapageweb.umontreal.ca/gosselif/stat4ci.html); if you use % it in your research, please, cite us:%	Chauvin, A., Worsley, K. J., Schyns, P. G., Arguin, M. & Gosselin, F. (2004).  A sensitive %	statistical test for smooth classification images.% % Alan Chauvin & FrŽdŽric Gosselin ([email protected]), 20/08/2004FWHM = sigma * sqrt(8*log(2));

README.md

@@ -0,0 +1,2 @@
+# Stat4Ci-Toolbox
+

SmoothCi.m

@@ -0,0 +1 @@
+function [Sci, h] = SmoothCiConv(ci,sigma)% SMOOTHCI Smoothes a 2D image.%	SmoothCi(ci,sigma) convolves a 2D image (ci) with a 2D Gaussian %	filter of standard deviation equals to sigma pixels.%		E.g.,% 		ci = double(imread('GenderFG.tiff'));% 		Sci = SmoothCi(ci,20);% 		or [Sci, h] = SmoothCi(ci,20); if the smoothing filter is needed% % Requires the ImageProcessing Toolbox.% % See also DEMOSTAT4CI% % The Stat4Ci toolbox is free (http://mapageweb.umontreal.ca/gosselif/stat4ci.html); if you use % it in your research, please, cite us:%	Chauvin, A., Worsley, K. J., Schyns, P. G., Arguin, M. & Gosselin, F. (2004).  A sensitive %	statistical test for smooth classification images.% % Alan Chauvin & FrŽdŽric Gosselin ([email protected]), 20/08/2004TNoyau = 6*sigma;h = fspecial('gaussian',ceil(TNoyau),sigma);Sci = filter2(h,ci);

StatThresh.m

@@ -0,0 +1 @@
+function Res = StatThresh(Res,p,sigma_b,tC,S_r);FWHM = HalfMax(sigma_b);		% computes the full width half maximumif exist('S_r')	D = length(size(S_r)); 					% number of dimensions in the search space	[volumes, N] = CiVol(sum(S_r(:)), D) 	% computes the intrinsic volumeselse	D = length(size(ZSCi)); 					% number of dimensions in the search space	[volumes, N] = CiVol(sum(ZSCi(:)), D) 	% computes the intrinsic volumesend;	[tP, k] = stat_threshold(volumes, N, FWHM, Inf, p, tC);	 % the actual statistical testsRes.tP = tP;Res.p = p;Res.tC = tC;Res.k = k;Res.FWHM = FWHM;

ZTransSCi.m

@@ -0,0 +1 @@
+function Res = ZtransSCi(SCi,varargin)% ZTRANSCI Z-transforms a smooth classification image.% 	In ZtransCi(SCi, meanSCi, stdSCi) the mean (meanSCi) and the standard % 	deviation (stdSCi) of the classification image are provided by the user.% 		E.g.,% 		mask = double(imread('faceMask.tif'));% 		mask = (mask - min(mask(:))) / (max(mask(:)) - min(mask(:)));% 		Ci = double(imread('GenderFG.tiff'));% 		vecCi = Ci(eq(mask, 0));% 		ZCi = ZTransCi(Ci, mean(vecCi(:)), std(vecCi(:)));% % 	In ZtransCi(SCi, meanSCi) the mean (meanSCi) of the classification image % 	is provided by the user and the standard deviation is estimated from the data.% % 	In ZtransCi(SCi) the mean and the standard deviation of the classification image % 	are estimated from the data.% % 	In ZtransCi(Ci1, n1, Ci2, n2, sigmaNoise, smoothFilter),% 		Ci1 = sum of white noise fields that led to a type 1 response (e.g., correct)  % 		n1 = number of type 1 response% 		Ci2 = sum of white noise fields that led to a type 2 response (e.g., incorrect)  % 		n2 = number of type 2 response % 		sigmaNoise = standard deviation of white noise% 		smoothFilter = Gaussian filter used to smooth the classification image% 	For an explanation, see:% 		Chauvin, Worsley, Schyns, Arguin & Gosselin (2004). A sensitive statistical % 		test for smooth classification images.% % See also SMOOTHCI, DEMOSTAT4CI%  % The Stat4Ci toolbox is free (http://mapageweb.umontreal.ca/gosselif/stat4ci.html); if you use % it in your research, please, cite us:%	Chauvin, A., Worsley, K. J., Schyns, P. G., Arguin, M. & Gosselin, F. (2004).  A sensitive %	statistical test for smooth classification images.% % Alan Chauvin & FrŽdŽric Gosselin ([email protected]), 20/08/2004% the mean and the std are estimated if not providedswitch nargincase 6,	cCi = SCi; 	nc = varargin{1};	iCi = varargin{2};	ni = varargin{3};	sigmaNoise = varargin{4};	smoothFilter = varargin{5};		n = ni + nc;				% number of trial		rhoC = (cCi/nc-iCi/ni)*nc*ni/n;		NvarX = n*sigmaNoise^2*sum(smoothFilter(:).^2);	NvarY = nc*ni/n;		% Z transform	Res.ZSCi = sqrt(n)*rhoC/sqrt(NvarX)/sqrt(NvarY);case 5,	Ci = SCi; 	nc = varargin{1};	ni = varargin{2};	sigmaNoise = varargin{3};	smoothFilter = varargin{4};		n = ni + nc;				% number of trial		rhoC = Ci;		NvarX = n*sigmaNoise^2*sum(smoothFilter(:).^2);	NvarY = nc*ni/n;		% Z transform	Res.ZSCi = sqrt(n)*rhoC/sqrt(NvarX)/sqrt(NvarY);	case 3	muCi = varargin{1};	stdCi = varargin{2};	% Z transform	Res.ZSCi = (SCi - muCi) / stdCi;	case 2	stdCi = std(SCi(:));	muCi = varargin{1};		% Z transform	Res.ZSCi = (SCi - muCi) / stdCi;	case nargin == 1    muCi = mean(SCi(:));    stdCi = std(SCi(:));        % Z transform    Res.ZSCi = (SCi - muCi) / stdCi;otherwise	error('Incorrect number of arguments.')end

cIm2hrCi.m

@@ -0,0 +1 @@
+function hrCi = cIm2hrCi(cIm)% CIM2HRCI Transforms a uint8 color image (cIm) into a 2D high-res %	(24 bits) classification image (hrCi).%		E.g.,	im = rand(256,256,3);%				cIm = uint8((256 - 1) * im);%				hrCi = cIm2hrCi(cIm);%% See also DEMOSTAT4CI, HRCI2CIM%% The Stat4Ci toolbox is free (http://mapageweb.umontreal.ca/gosselif/stat4ci.html); if you use % it in your research, please, cite us:%	Chauvin, A., Worsley, K. J., Schyns, P. G., Arguin, M. & Gosselin, F. (2004).  A sensitive %	statistical test for smooth classification images.% % Alan Chauvin & Fr�d�ric Gosselin ([email protected]), 20/08/2004	cIm = double(cIm);hrCi = cIm(:,:,1) * 256^2 + cIm(:,:,2) * 256 + cIm(:,:,3);

hrCi2cIm.m

@@ -0,0 +1 @@
+function cIm = hrCi2cIm(hrCi)% HRCI2CIM Transforms a 2D high-res (24 bits) classification image (hrCi)%	into a uint8 color image (cIm). The color image can then be saved in % 	your favorite image format without loss of information (e.g., TIF). % 	E.g.,	Ci = rand(256,256);% 			hrCi = round((256^3 - 1) * Ci);% 			cIm = hrCi2cIm(hrCi);% % See also DEMOSTAT4CI, CIM2HRCI% % The Stat4Ci toolbox is free (http://mapageweb.umontreal.ca/gosselif/stat4ci.html); if you use % it in your research, please, cite us:%	Chauvin, A., Worsley, K. J., Schyns, P. G., Arguin, M. & Gosselin, F. (2004).  A sensitive %	statistical test for smooth classification images.% % Alan Chauvin & Fr�d�ric Gosselin ([email protected]), 20/08/2004tIm(:,:,3) = mod(hrCi, 256);tIm(:,:,2) = mod(hrCi - squeeze(tIm(:,:,3)), 256^2);tIm(:,:,1) = mod(hrCi - squeeze(tIm(:,:,3)) - squeeze(tIm(:,:,2)), 256^3);tIm(:,:,2) = tIm(:,:,2) / 256;tIm(:,:,1) = tIm(:,:,1) / 256^2;cIm = uint8(tIm);