fdsc.m

模式识别的主要工具集合

%FDSC Feature based Dissimilarity Space Classification
%
%   W = FDSC(A,R,FEATMAP,TYPE,P,CLASSF)
%   W = A*FDSC([],R,FEATMAP,TYPE,P,CLASSF)
%
% INPUT
%   A       Dateset used for training
%   R       Dataset used for representation
%   FEATMAP Preprocessing in feature space (e.g. SCALEM)
%           Default: no preprocessing.
%   TYPE    Dissimilarity rule, see PROXM
%           Default 'DISTANCE'.
%   P       Parameter for dissimilarity rule, see PROXM
%           Default P = 1.
%   CLASSF  Classifier used in dissimilarity space
%           Default FISHERC.
%
% OUTPUT
%   W       Resulting, trained feature space classifier
%
% DESCRIPTION
% This routine builds a classifier in feature space based on a 
% dissimilarity representation defined by the representation set R
% and the dissimilarities found by A*FEATMAP*PROXM(R*FEATMAP,TYPE,P).
% FEATMAP is a preprocessing in feature space, e.g. scaling
% (SCALEM([],'variance') or pre-whitening (KLMS).
%
% New objects in feature space can be classified by D = B*W or by
% D = MAP(B,W). Labels can be found by LAB = D*LABELD or LAB = LABELD(D).
%
% EXAMPLE
% A = GENDATB([100 100]); % Training set of 200 objects
% R = GENDATB([10 10]);   % Representation set of 20 objects
% W = FDSC(A,R);          % Compute classifier
% SCATTERD(A);            % Scatterplot of trainingset
% HOLD ON; SCATTERD(R,'kBeste o'); % Add representation set to scatterplot
% PLOTC(W);               % Plot classifier
%
% SEE ALSO
% DATASETS, MAPPINGS, SCALEM, KLMS, PROXM, LABELD

% Copyright: R.P.W. Duin, [e-mail]
% Faculty EWI, Delft University of Technology
% P.O. Box 5031, 2600 GA Delft, The Netherlands

function w = fdsc(a,r,featmap,type,p,classf)
if nargin < 6, classf = fisherc; end
if nargin < 5, p = 1; end
if nargin < 4, type = 'distance'; end
if nargin < 3, featmap = []; end
if nargin < 2
	error(['Insufficient number of input parameters,' newline ...
			    'at least a representation set is needed'])
end

if isempty(a)
		w1 = mapping(mfilename,'untrained',{r,featmap,type,p,classf}); 
		w1 = setname(w1,'FeatDisSpaceC');
	return
end


isvaldset(a,1,2); % at least 1 object per class, 2 classes
if isdataset(r)
	isvaldset(r,1,1);
	if isempty(featmap)
		v = scalem(a); % shift mean to the origin
	elseif isuntrained(featmap)
		v = a*featmap;
	elseif ismapping(featmap)
		v = featmap;
	else
		error('Feature mapping expected')
	end
	w = proxm(r*v,type,p);
	if ~isuntrained(classf)
		error('Untrained classifier expected')
	end
	u = a*v*w*classf;
	w = v*w*u;
else
	error('Unexpected input')
end