www.gusucode.com > 用于人脸识别和人脸定位的主动形状模型（asm）matlab工具箱 > asm toolbox\asm.m
%%%%%%%%%%%%%%%%%%%%%%%%%%
%(c) Ghassan Hamarneh 1999
%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%% Based on (unless otherwise stated):
%%%
%%% Active Contour Models--Their Training & Application
%%% T.F.Cootes, C.J.Taylor, et.al.
%%% Computer Vision and Image Understanding,
%%% Vol.61 No.1,January pp.38-59, 1995
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%% and for the multiresolution part:
%%%
%%% Active Shape Models: Evaluation of a Multi-Resolution Method
%%% for Improving Image Search
%%%
%%% T.F.Cootes, C.J.Taylor, A.Lanitis
%%% Proc. British Machine Vision Conference, vol. 1 ,
%%% 1994, Ed.E.Hancock BMVA Press. pp327-336
%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%
%%% Programming by:
%%% Ghassan Hamarneh
%%% Image Analysis Group
%%% CHALMERS UNIVERSITY OF TECHNOLGY
%%%
%%% Currently at:
%%% Computing Science
%%% Simon Fraser University
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%
%%% On 991006 the following fixes where made:
%%% 1. using the function landmark(..) so you can see/delete landmarks when labeling
%%% 2. specify percentage of explained variance instead of numnber of eigen vectors
%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%
%%% July 19, 2004: added ExplainPercent to the saved training variables
%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
LABEL_DATA_EXIST=0;
%LABEL_DATA_EXIST=1; %for loading DTU labelled data
% ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
%Using uiwait and helpdlg
uiwait(helpdlg({'A C T I V E S H A P E M O D E L S' ...
' with Multi-Resolution statistics ' ''...
'By ... ' '' 'Ghassan Hamarneh' '' ['done: 11-Aug-1997 -- today: ',date] },'ASM'));
%help information
uiwait(helpdlg({'HELP INFORMATION:',...
'',...
' .I. This Program is divided into 3 main stages:',...
' you can choose to skip any one of them',...
' 1 TRAINING ',...
' 2 TRYING WEIGHTS ',...
' 3 APPLICATION ',...
'',...
' .II. Follow the instructions in the dialog boxes.',...
'',...
'.III. Click Cancel in any dialog box to quit.',...
'',...
},'ASM'));
ButtonName=questdlg('Do you want to train?','ASM');
DidTrain=ButtonName;
if(strcmp(ButtonName,'Cancel'))
msgbox('Terminating: cancelled request to train.','ASM');
return;
elseif (strcmp(ButtonName,'Yes'))
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% TRAINING STAGE %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%Number of shapes in training set = NumTrnSetImgs
%Number of landmark points = NumLandMarkPts
%Number of dimensions = D = 2
uiwait(helpdlg('THIS IS THE FIRST STAGE: TRAINING STAGE.','ASM'));
uiwait(helpdlg('Now, you will enter some initialization information used for training.','ASM'));
%Using the inputdlg
prompt={'Enter Contours Ending Points. (Last entry is no. of landmarks, ex. [8 16 27])',...
'Enter The number of training set images (2,3,4,..):',...
'Enter Percentage of explained vaiance (0-1):',...
'Enter Number of points along training profile above landmark (1,2,3,...):',...
'Enter Number of points along training profile below landmark (1,2,3,...):'};
def={'[76 93 110 157 222 239 253 261 269]','3','0.90','3','3'};
TheTitle='ASM';
lineNo=[1,1,1,1,1];
answer=inputdlg(prompt,TheTitle,lineNo,def);
if isempty(answer) msgbox('Terminating: cancelled inputing initialization info. for training','ASM'); return; end
ContoursEndingPoints=str2num(answer{1});
NumTrnSetImgs = str2num(answer{2});
ExplainPercent= str2num(answer{3});
TrnPntsAbove = str2num(answer{4});
TrnPntsBelow = str2num(answer{5});
NumLandMarkPts=ContoursEndingPoints(end);
%NOTE:
%------
%ContoursEndingPoints,NumTrnSetImgs,NumLandMarkPts
%will be overwritten if labelled data already exist and will be loaded
%as in DTU data
%STEP 1 -----------------
%Obtain landmark coordinates for each shape in the training set
%Result: 'unaligined training set shape coordinates' matrix Xu with rows=2*NumLandMarkPts & cols=NumTrnSetImgs
% remove remark of next 3 lines >>>>>>
uiwait(helpdlg('Now, you will select the landmark points in the training images.','ASM'));
if LABEL_DATA_EXIST==1,
[Xu,TrnImgFiles,ContoursEndingPoints,NumTrnSetImgs,NumLandMarkPts]=GetTrnSetCoor2;
else
[Xu,TrnImgFiles]=GetTrnSetCoor(NumTrnSetImgs,NumLandMarkPts);
end
if Xu==0 msgbox('Terminating: cancelled loading images','ASM'); return; end
% <<<<<<<< remove remark of prev 3 lines
figure
PlotShapes(Xu,'ASM: unaligined training set',ContoursEndingPoints);
%STEP 1.1 -----------------
%Calculate the Mean Normalized Derivative Profile for each landmark point
%RESULT: a 2D array of NumLandMarkPts Profiles
%Calculate the covariance matrix of the Mean Normalized Derivative Profiles for each landmark
%RESULT: a cell array of squares arrays of length (1+TrnPntsAbove+TrnPntsBelow)-1
MaxNumPyramidLevels=GetMaxNumPyramidLevels(TrnImgFiles);
uiwait(msgbox(['Using ',num2str(MaxNumPyramidLevels),' levels in image pyramid'],'ASM'));
uiwait(msgbox(['The image intensity training profile statistics will now be collected'],'ASM'));
[MnNrmDrvProfiles,ProfilesCov]=GetProfileStatistics(TrnImgFiles,Xu,TrnPntsAbove,TrnPntsBelow,ContoursEndingPoints,MaxNumPyramidLevels);
%STEP 2 -----------------
%Find the weighting matrix to give more significance to those points which tend to be stable
%RESULT: 'landmark points weighting matrix' square matrix W with rows=cols=NumLandMarkPts
uiwait(helpdlg({'Now, the weighting matrix will be calculated,'...
'the shapes will be aligned and' ...
'the training set statistics will be obtained.'},'ASM'));
W=GetWeights(Xu);
%STEP 3 -----------------
%Align the shapes of the training set using the obtained coordinates
%RESULT: 'aligned training set shape coordinates' matrix Xa
Xa=AlignTrnSetCoor(Xu,W,ContoursEndingPoints);
figure
PlotShapes(Xa,'ASM: aligined training set',ContoursEndingPoints);
%STEP 4 -----------------
%Obtaining the statistical description of the training set shape coordinates
%REUSLT: 'Statistical Model x=xm + P*b' where,
% xm: coordinates of mean shape -> 2*NumLandMarkPts x 1
% P: matrix of first t eigenvectors of covariance matrix -> 2*NumLandMarkPts x t
% b: vector of weights -> t x 1
[MeanShape,tEigenvectors,tEigenValues]=GetShapeStatistics(Xa,ExplainPercent);
ButtonName=questdlg('Do you want to save training results','ASM');
if(strcmp(ButtonName,'Cancel'))
msgbox('Terminating: cancelled request to save training results','ASM');
return;
elseif (strcmp(ButtonName,'Yes'))
ExplainPercent %useful to use in filename when saving
TrnPntsAbove %useful to use in filename when saving
TrnPntsBelow %useful to use in filename when saving
[newmatfile, newpath] = uiputfile('*.mat', 'ASM: Save as...');
if(newmatfile==0)msgbox('Terminating: cancelled entering .mat file name for save','ASM'); return; end
save([newpath,newmatfile],'MeanShape','tEigenvectors','tEigenValues','W','ContoursEndingPoints',...
'MnNrmDrvProfiles','ProfilesCov','TrnPntsBelow','TrnPntsAbove','MaxNumPyramidLevels','ExplainPercent');
end
end
if (~strcmp(DidTrain,'Yes')) %if did not do training
uiwait(helpdlg('Since you chose no training, you will have to load a previous data (.mat) file.','ASM'));
[fname,pname]=uigetfile('*.mat','ASM: choose *.mat file to load');
if fname==0; msgbox('Terminating: cancelled loading data (.mat) file','ASM'); return; end
load([pname,fname])
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% TRY WEIGHTS %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
ButtonName=questdlg('Do you want to animate modes of variation?','ASM');
if(strcmp(ButtonName,'Cancel'))
msgbox('Terminating: cancelled animating modes of variation.','ASM');
return;
elseif (strcmp(ButtonName,'Yes'))
%animating modes of variation b in X=M+Pb
uiwait(helpdlg('THIS IS THE SECOND STAGE: TRYING WEIGHTS: Animating.','ASM'));
uiwait(helpdlg('Now, you will choose a mode of variation and a range of weights in X = Xmean + P*b.','ASM'));
TerminateNotContinue = TryWeights2(MeanShape,tEigenvectors,tEigenValues,ContoursEndingPoints);
if TerminateNotContinue msgbox('Terminating: cancelled trying weights','ASM'); return; end
end
ButtonName=questdlg('Do you want to try weights?','ASM');
if(strcmp(ButtonName,'Cancel'))
msgbox('Terminating: cancelled request to try weights.','ASM');
return;
elseif (strcmp(ButtonName,'Yes'))
%trying different weights b in X=M+Pb
uiwait(helpdlg('THIS IS THE SECOND STAGE: TRYING WEIGHTS: Entering weights.','ASM'));
uiwait(helpdlg('Now, you will enter different weight values for b in X = Xmean + P*b.','ASM'));
TerminateNotContinue = TryWeights(MeanShape,tEigenvectors,tEigenValues,ContoursEndingPoints);
if TerminateNotContinue msgbox('Terminating: cancelled trying weights','ASM'); return; end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% APPLICATION STAGE %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
ButtonName=questdlg('Do you want to search for a shape in an image?','ASM');
if(strcmp(ButtonName,'Cancel'))
msgbox('Terminating: cancelled request to search for a shape.','ASM');
return;
elseif (strcmp(ButtonName,'Yes'))
uiwait(helpdlg('THIS IS THIRD STAGE: APPLICATION STAGE.','ASM'));
uiwait(helpdlg('Now, you will choose an image to search and enter some initialization information','ASM'));
TerminateNotContinue=FindShapeInImage(MeanShape,tEigenvectors,tEigenValues,W,ContoursEndingPoints,...
MnNrmDrvProfiles,ProfilesCov,TrnPntsBelow,TrnPntsAbove,MaxNumPyramidLevels);
if TerminateNotContinue msgbox('Terminating: cancelled searching image','ASM'); return; end
end
msgbox('Done!','ASM');