code.txt

face recongnisation code

Code

Here is some Matlab code that you might find useful. All code is licensed under the GNU Lesser General Public License (LGPL) unless otherwise stated. I also have a Forge project which potentially has more up-to-date code, available here. Please could you contact me if you make any modifications to these files - I'd really like to hear from you!
Pre-process

    * normalise.m - normalise a matrix of examples so that each feature has unit norm.

                  function [normalisedX1, normalisedX2] = normalise(X1, X2)
% Normalise the features (columns) of matrices X1 (and optionally X2) such that 
% each feature of X1 has unit norm. X1 and X2 have examples as their rows. 
%
% Usage: [normalisedX1, normalisedX2] = normalise(X1, X2)
% Inputs/Outputs: 
%   X1 - an (l x n) matrix whose rows are examples
%   X2 (optional) - an (l2 x m) matrix whose rows are examples
%
%   normalisedX1 - normalised X1 
%   normalisedX2 (optional) - normalised X2 
%
% Copyright (C) 2006 Charanpal Dhanjal 

% This library is free software; you can redistribute it and/or
% modify it under the terms of the GNU Lesser General Public
% License as published by the Free Software Foundation; either
% version 2.1 of the License, or (at your option) any later version.
% 
% This library is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
% Lesser General Public License for more details.
% 
% You should have received a copy of the GNU Lesser General Public
% License along with this library; if not, write to the Free Software
% Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA

if (nargin < 1)
    fprintf('%s\n', help('normalise'));
    error('Incorrect number of inputs - see above usage instructions.');
end

numFeatures = size(X1, 2);

%Just make sure the training example features have unit norm 
featureNorms = sqrt(sum(X1.^2));

%Bit of cheat to make sure we don't divide by zero
zeros = featureNorms == 0;
featureNorms = 1./(featureNorms+zeros);

if ~issparse(X1)
    diagNorms = diag(featureNorms); 
else    
    diagNorms = speye(numFeatures); 

    for i=1:numFeatures
        diagNorms(i, i) = featureNorms(i);
    end 
end

normalisedX1 = X1*diagNorms;

if (nargin == 2)
    normalisedX2 = X2*diagNorms; 
end




    * centerData.m - center a matrix of examples so that each feature has zero mean.
   
                 function [cX1, cX2] = centerData(X1, X2) 
% Centers matrices X1 (and optionally X2) by taking the mean of each column 
% (feature) of X1 and subtracting it from the feature values.
%
% Usage: [cX1, cX2] = centerData(X1, X2) 
% Inputs/Outputs: 
%   X1 - an (l x n) matrix whose rows are examples
%   X2 (optional) - an (l2 x m) matrix whose rows are examples
%
%   cX1 - centered X1 
%   cX2 (optional) - centered X2 
%
% Copyright (C) 2006 Charanpal Dhanjal 

% This library is free software; you can redistribute it and/or
% modify it under the terms of the GNU Lesser General Public
% License as published by the Free Software Foundation; either
% version 2.1 of the License, or (at your option) any later version.
% 
% This library is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
% Lesser General Public License for more details.
% 
% You should have received a copy of the GNU Lesser General Public
% License along with this library; if not, write to the Free Software
% Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301
% USA

if (nargin < 1)
    fprintf('%s\n', help('centerData'));
    error('Incorrect number of inputs - see above usage instructions.');
end

numX1Examples = size(X1, 1);
meanX1 = mean(X1);

if (nargin == 1)
    cX1 = X1 - ones(numX1Examples, 1)*meanX1;
else 
    numX2Examples = size(X2,1);
    cX1 = X1 - ones(numX1Examples, 1)*meanX1;
    cX2 = X2 - ones(numX2Examples, 1)*meanX1;
end



 * normaliseExamples.m - scale each example so that it lies on a hyper-sphere of radius 1.

            function [normalisedX] = normaliseExamples(X)
% Normalise examples so they lie on a sphere of radius 1
%
% Usage: [normalisedX] = normaliseExamples(X)
% Inputs/Outputs: 
%   X - an (l x n) matrix whose rows are examples
%
%   normalisedX - normalised X 

% Copyright (C) 2006 Charanpal Dhanjal 

% This library is free software; you can redistribute it and/or
% modify it under the terms of the GNU Lesser General Public
% License as published by the Free Software Foundation; either
% version 2.1 of the License, or (at your option) any later version.
% 
% This library is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
% Lesser General Public License for more details.
% 
% You should have received a copy of the GNU Lesser General Public
% License along with this library; if not, write to the Free Software
% Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301
% USA

if (nargin ~= 1)
    fprintf('%s\n', help(sprintf('%s', mfilename)));
    error('Incorrect number of inputs - see above usage instructions.');
end

R =  max(sqrt(sum(X.^2, 2)));

normalisedX = X/R; 


Evaluate

    * precision.m - compute the precision on a set of predicted labels.
    * recall.m - compute the recall of a set of predicted labels.
    * fMeasure.m - compute the F-measure of a set of predicted labels.
    * averagePrecision.m - compute the average precision of a set of predicted labels.
    * truePositiveRate.m - compute the true positive rate of a set of predicted labels.
    * falsePositiveRate.m - compute the false positive rate of a set of predicted labels.
    * balancedErrorRate.m - compute the balanced error rate (BER) of a set of predicted labels.
    * rootMeanSqError.m - compute the root mean squared erorr of a set of predicted labels.

Feature extraction

    * primalGeneralFeatures.m - extract primal general features for a matrix of examples and predicted labels
    * maxVariance.m - find the projection vector of maximal variance for a matrix of examples.
    * maxCovariance.m - find the projection vector which maximises the covariance between a matrix of examples and corresponding labels.
    * dualPCATrain.m - train the Kernel Principal Components Analysis (KPCA) algorithm.
    * dualPCAProject.m - project test examples for the Kernel Principal Components Analysis (KPCA) algorithm.

Miscellaneous

    * data.zip - a data object which is efficient with memory usage. Operations to add and delete matrices to the object as well as permuting and partitioning data.
    * binaryLabels.m - check if a label matrix contains binary values.
    * vprintf.m - print strings to screen with optional relevancy parameter.
    * getSpaceNames.m - utility function to get name of X and Y spaces.
    * maxN.m - return indices for the maximum n elements of a matrix.