function varargout = image_dbr(varargin)
%IMAGE_DBR M-file for image_dbr.fig
% IMAGE_DBR, by itself, creates a new IMAGE_DBR or raises the existing
% singleton*.
%
% H = IMAGE_DBR returns the handle to a new IMAGE_DBR or the handle to
% the existing singleton*.
%
% IMAGE_DBR('Property','Value',...) creates a new IMAGE_DBR using the
% given property value pairs. Unrecognized properties are passed via
% varargin to image_dbr_OpeningFcn. This calling syntax produces a
% warning when there is an existing singleton*.
%
% IMAGE_DBR('CALLBACK') and IMAGE_DBR('CALLBACK',hObject,...) call the
% local function named CALLBACK in IMAGE_DBR.M with the given input
% arguments.
%
% *See GUI Options on GUIDE's Tools menu. Choose "GUI allows only one
% instance to run (singleton)".
%
% See also: GUIDE, GUIDATA, GUIHANDLES
% Edit the above text to modify the response to help image_dbr
% Last Modified by GUIDE v2.5 16-Aug-2008 23:50:43
% Begin initialization code - DO NOT EDIT
gui_Singleton = 1;
gui_State = struct('gui_Name', mfilename, ...
'gui_Singleton', gui_Singleton, ...
'gui_OpeningFcn', @image_dbr_OpeningFcn, ...
'gui_OutputFcn', @image_dbr_OutputFcn, ...
'gui_LayoutFcn', [], ...
'gui_Callback', []);
if nargin && ischar(varargin{1})
gui_State.gui_Callback = str2func(varargin{1});
end
if nargout
[varargout{1:nargout}] = gui_mainfcn(gui_State, varargin{:});
else
gui_mainfcn(gui_State, varargin{:});
end
% End initialization code - DO NOT EDIT
function image_dbr_OpeningFcn(hObject, eventdata, handles, varargin)
% Main routine, this starts after creation of the guide
% Handling of input parameters
handles.output = hObject;
h = my_handles(handles);
if length(varargin) < 4
cclasf = meanc; % default combiner
else
cclasf = varargin{4};
end
if length(varargin) < 3
clasf = knnc([],1); % default classifier
else
clasf = varargin{3};
end
featsets = varargin{2}; % arrange feature sets in cells
if ~iscell(featsets), featsets = {featsets}; end
if ~iscell(clasf)
clasf = repmat({clasf},1,length(featsets));
else
if length(clasf) ~= length(featsets)
error('Number of classifiers and number of feature sets do not match')
end
end
database = varargin{1}; % dataset or datafile, just used for image display
m = size(database,1);
labels = zeros(1,m); % user give labels: 0 (no), 1 (target, 2 (outlier)
S = randperm(m); % initial set of images to be shown
for j=1:4 % feature weight slider bars default invisible
set(h.feat_weight(j),'visible','off');
set(handles.(['text' num2str(j+2)]),'visible','off');
end
% optimize location sliders in case of 2 or 3 sliders
if length(featsets) == 1, feath = 2; end
if length(featsets) == 2, feath = [2 3]; end
if length(featsets) == 3, feath = [1 2 3]; end
if length(featsets) == 4, feath = [1 2 3 4]; end
for j=1:length(featsets) % initialize feature sets and related sliders
if size(featsets{j},1) ~= m
error('Feature set(s) have wrong size')
end
set(handles.(['text' num2str(feath(j)+2)]),'visible','on');
set(handles.(['text' num2str(feath(j)+2)]),'string',getname(featsets{j}));
% re-arrange datasets with features, use 'target' and 'outlier' as labels
featsets{j} = prdataset(featsets{j});
featsets{j} = featsets{j}*scalem(featsets{j},'variance');
featsets{j} = prdataset(+featsets{j});
featsets{j} = setlablist(featsets{j},char('target','outlier'));
featsets{j} = setnlab(featsets{j},labels);
set(h.feat_weight(feath(j)),'value',1); % initial weights: 1
set(h.feat_weight(feath(j)),'visible','on');
end
if length(featsets) == 1 % no feature slider bar in case of one feature set
set(h.feat_weight(feath(1)),'visible','off');
set(handles.(['text' num2str(feath(j)+2)]),'visible','off');
end
% store useful data in figure user data field
guidata(hObject,{h,database,featsets,clasf,cclasf});
set(h.classify,'userdata','reset'); % initial procedure: reset
set(h.all,'value',0); % initial ranking:
set(h.unlabeled,'value',2); % unlabeled objects
labels_im = zeros(1,10); % labels of shown images
weights = zeros(1,4); % weights of feature sets
Targets = [];
Outliers= [];
show_targ_out(database,Targets,1,h,'Targets'); % initialise target /
show_targ_out(database,Outliers,1,h,'Outliers');% outlier images
S = randperm(m);
waitstate = prwaitbar;
prwaitbar off
% main loop
while(1) % S, images to be shown
show_images(database,S,1,h); % show images on axes
% wait for the user
% ..........
uiwait % uiresume is activated by one of the
% CLASSIFY / LABEL / RESET / QUIT buttonS.
% get labels and weights
Targets = get(h.targets,'userdata'); % get Tagerts
Outliers = get(h.outliers,'userdata');%abd Outliers
if isempty(guidata(hObject)) % we need to stop (QUIT button is pressed)
guidata(hObject,{S,Targets,Outliers});%
prwaitbar(waitstate);
return % return ranking S, Targets and Outliers.
end
proch = proc(h); % classify / label / reset
if ~strcmp(proch,'reset');
for j=1:10 % labels are 1-2, convert from 0-1 check boxes
labels_im(j) = 2 - get(h.handlab(j),'value');
end
for j=1:length(featsets) % get feature weights from sliders
weights(j) = get(h.feat_weight(feath(j)),'value');
end
T = [1:m]; % potential trainingset
% Targets = get(h.targets,'userdata');
% Outliers = get(h.outliers,'userdata');
t = get(h.next,'userdata'); n = t{2};
SS = []; % training set indices in case of 'label' procedure
for j=1:10 % update labeling
SAct = S(j-1+n); % object number j in S
if labels_im(j) == 1
if isempty(find(Targets==SAct)), Targets = [Targets SAct]; end
else
if isempty(find(Outliers==SAct)), Outliers = [Outliers SAct]; end
end
SS = [SS SAct];
end
set(h.targets,'userdata',Targets); % store indices of Targets
set(h.outliers,'userdata',Outliers); % and outliers
labels = zeros(1,m); % construct labels training set
labels(Targets) = ones(1,length(Targets));
labels(Outliers) = 2*ones(1,length(Outliers));
if strcmp(proch,'label') % Label: learn from present set only
if all(labels_im == labels_im(1)) % present set is uniformly labeled
if labels_im(1) == 1 % no outliers, use remaining unlabeled dataset as outlier
labels_train = ones(size(labels))*2; % Label entire set as outlier
labels_train(SS) = ones(size(SS)); % Label present set as target
else % no targets, use remaining dataset as target
labels_train = ones(size(labels)); % Label entire set as target
labels_train(SS) = 2*ones(size(SS)); % Label present set as outlier
end
else
T = SS;
labels_train = labels_im;
end
[S,W] = train_classify(featsets,T,labels_train,clasf,cclasf,weights);
else % Classify procedure:
T = find(labels > 0); % learn from all Targets and Outliers
labels_train = labels(T);
[S,W] = train_classify(featsets,T,labels_train,clasf,cclasf,weights);
end
U = find(labels(S) == 0);
set(h.all,'userdata',{S,U});
if (get(h.all,'value') == 0)
S = S(U);
end
else % reset
S = randperm(m);
end
% S is the ranked set of indices, either for all objects, or for just
% the unlabeled objects
end
return
function h = my_handles(handles)
% construct meaningfull handle names / numbers
T = [1 2 3 4 7 8 9 10 11 12]; % correct checkbox numbering
h.image = zeros(1,10);
h.handlab = zeros(1,10);
h.feat_weight = zeros(1,4);
for j=1:10
h.image(j) = handles.(['axes' num2str(j)]);
h.imnum(j) = handles.(['text' num2str(j+8)]);
h.obnum(j) = handles.(['text' num2str(j+18)]);
h.handlab(j) = handles.(['checkbox' num2str(T(j))]);
end
for j=1:4
h.feat_weight(j) = handles.(['slider' num2str(j)]);
end
h.all = handles.radiobutton4;
h.unlabeled = handles.radiobutton5;
h.classify = handles.pushbutton1;
h.targets = handles.axes11;
h.outliers = handles.axes12;
h.target_slider = handles.slider5;
h.outliert_slider = handles.slider6;
h.target_title = handles.text1;
h.outlier_title = handles.text2;
h.target_obnum = handles.text29;
h.outlier_obnum = handles.text30;
h.target_delete = handles.pushbutton2;
h.target_move = handles.pushbutton3;
h.outlier_delete = handles.pushbutton5;
h.outlier_move = handles.pushbutton4;
h.next = handles.pushbutton8;
h.previous = handles.pushbutton9;
return
function varargout = image_dbr_OutputFcn(hObject, eventdata, handles)
% takes care of final return of main return
varargout = guidata(hObject);
delete(hObject);
return
function pushbutton1_Callback(hObject, eventdata, handles)
% CLASSIFY button, this reactivates the UIWAIT function in the main loop
s = guidata(hObject);
h = s{1};
set(h.classify,'userdata','classify');
uiresume
return
function show_images(database,S,n,h)
% show images on axes
Targets = get(h.targets,'userdata');
Outliers = get(h.outliers,'userdata');
im = data2im(database,S(n:n+9));
% the new images
for j=1:10
show_im(im,j,h.image(j));
set(h.handlab(j),'value',1); % set target default
set(h.imnum(j),'string',num2str(n+j-1));
set(h.obnum(j),'string',num2str(S(n+j-1)));
end
set(h.next,'userdata',{S,n});
% set visibility of next / previous buttons
if n > 1
set(h.previous,'visible','on');
else
set(h.previous,'visible','off');
end
if n < (length(S)-9)
set(h.next,'visible','on');
else
set(h.next,'visible','off');
end
% the targets, show most recent one
if ~isempty(Targets)
show_targ_out(database,Targets,length(Targets),h,'Targets');
set(h.target_obnum,'string',num2str(Targets(end)));
end
% the outliers, show most recent one
if ~isempty(Outliers)
show_targ_out(database,Outliers,length(Outliers),h,'Outliers');
set(h.outlier_obnum,'string',num2str(Outliers(end)));
end
return
function show_targ_out(database,Pointers,n,h,name)
% show targets or outliers and update properties
if strcmp(name,'Targets') % create single handle variable for targets and outliers
h_axes = h.targets;
h_slider = h.target_slider;
h_title = h.target_title;
h_delete = h.target_delete;
h_move = h.target_move;
h_obnum = h.target_obnum;
else % Outliers
h_axes = h.outliers;
h_slider = h.outliert_slider;
h_title = h.outlier_title;
h_delete = h.outlier_delete;
h_move = h.outlier_move;
h_obnum = h.outlier_obnum;
end
if isempty(Pointers) % all Targets / Outliers deleted, make all related objects invisible
set(h_title,'String',['No ' name ' defined'])
set(get(h_axes,'children'),'visible','off')
set(h_axes,'visible','off');
set(h_slider,'visible','off');
set(h_delete,'visible','off');
set(h_move,'visible','off');
set(h_obnum,'visible','off');
set(h_axes,'userdata',[]);
else
show_im(data2im(database,Pointers(n)),1,h_axes); % show image
set(h_axes,'userdata',Pointers);
num_pointers = length(Pointers); % number of images
set(h_title,'String',[num2str(num_pointers) ' ' name]); % set title
set(h_obnum,'String',num2str(Pointers(n)));
set(h_slider,'Min',0.99999/num_pointers); % min (about 1/n)
set(h_slider,'Max',1); % max
if num_pointers > 1 % step size
set(h_slider,'SliderStep',[1/(num_pointers-1) 1/(num_pointers-1)]);
set(h_slider,'value',n/num_pointers);
set(h_slider,'visible','on');
else % special case: one target/outlier image
set(h_slider,'SliderStep',[1 1]);
set(h_slider,'value',n/num_pointers);
set(h_slider,'visible','off');
end
% make image and buttons visible
set(get(h_axes,'children'),'visible','on')
set(h_delete,'visible','on');
set(h_move,'visible','on');
set(h_obnum,'visible','on');
end
return
function proch = proc(h)
% get desired procedure
proch = get(h.classify,'userdata'); % classify / label / reset
return
function show_im(im,n,h_axes)
% low level display routine
if iscell(im), y = im{n};
else, y = squeeze(im(:,:,:,n));
end
if size(y,3) == 1 % reset gray images to color
y = cat(3,y,y,y);
end
axes(h_axes); % activate the right axes
image(y); % display
% prepare for callback
set(get(gca,'children'),'ButtonDownFcn',{@resetlab,n})
axis off % make axes invisible
axis equal
return
function resetlab(hObject, eventdata,n)
% Reset the image labels (target/outlier) by clicking in the image
s = guidata(hObject);
h = s{1};
get(h.handlab(n),'value');
set(h.handlab(n),'value',1 - get(h.handlab(n),'value'));
get(h.handlab(n),'value');
return
function [S,W] = train_classify(featsets,T,labels_train,clasf,cclasf,weights);
% train, combine, classify and rank
% S will be the ranked object indices of all or unlabeled objects
d = [];
W = [];
for j=1:length(featsets)
b = featsets{j};
trainset = setnlab(b(T,:),labels_train);
%trainset = setprior(trainset,getprior(trainset,0));
trainset = setprior(trainset,0);
if ~isvaldset(trainset,2)
v = trainset*knnc([],1);
else
v = trainset*clasf{j}*classc;
end
d = [d featsets{j}*v*weights(j)];
W = [W;v];
end
d = d*cclasf;
d = +d(:,'target');
[dd,S] = sort(-d);
% W = v*affine(weights)*cclasf; % to be corrected
W = [];
return
function slider5_Callback(hObject, eventdata, handles)
% target slider
s = guidata(hObject);
h = s{1};
database = s{2};
Targets = get(h.targets,'userdata');
n = round(get(h.target_slider,'value')*length(Targets));
show_im(data2im(database,Targets(n)),1,h.targets);
set(h.targets,'userdata',Targets); % image() destroys userdata, restore it
set(h.target_title,'String',[num2str(length(Targets)) ' Targets']); % needed???
set(h.target_obnum,'String',num2str(Targets(n)));
return
function slider6_Callback(hObject, eventdata, handles)
% outlier slider
s = guidata(hObject);
h = s{1};
database = s{2};
Outliers = get(h.outliers,'userdata');
n = round(get(h.outliert_slider,'value')*length(Outliers));
show_im(data2im(database,Outliers(n)),1,h.outliers);
set(h.outliers,'userdata',Outliers); % image() destroys userdata, restore it
set(h.outlier_title,'String',[num2str(length(Outliers)) ' Outliers']); % needed???
set(h.outlier_obnum,'String',num2str(Outliers(n)));
return
function pushbutton2_Callback(hObject, eventdata, handles)
% delete target
s = guidata(hObject);
h = s{1};
database = s{2};
Targets = get(h.targets,'userdata');
n = round(get(h.target_slider,'value')*length(Targets));
Targets(n) = [];
n = max(n-1,1);
show_targ_out(database,Targets,n,h,'Targets');
return
function pushbutton3_Callback(hObject, eventdata, handles)
% move target to outlier
s = guidata(hObject);
h = s{1};
database = s{2};
Targets = get(h.targets,'userdata');
Outliers = get(h.outliers,'userdata');
n = round(get(h.target_slider,'value')*length(Targets));
Outliers = [Outliers Targets(n)];
Targets(n) = [];
n = max(n-1,1);
show_targ_out(database,Outliers,length(Outliers),h,'Outliers');
show_targ_out(database,Targets,n,h,'Targets');
return
function pushbutton4_Callback(hObject, eventdata, handles)
% move outlier to target
s = guidata(hObject);
h = s{1};
database = s{2};
Outliers = get(h.outliers,'userdata');
Targets = get(h.targets,'userdata');
n = round(get(h.outliert_slider,'value')*length(Outliers));
Targets = [Targets Outliers(n)];
Outliers(n) = [];
n = max(n-1,1);
show_targ_out(database,Targets,length(Targets),h,'Targets');
show_targ_out(database,Outliers,n,h,'Outliers');
return
function pushbutton5_Callback(hObject, eventdata, handles)
% delete outlier
s = guidata(hObject);
h = s{1};
database = s{2};
Outliers = get(h.outliers,'userdata');
n = round(get(h.outliert_slider,'value')*length(Outliers));
Outliers(n) = [];
n = max(n-1,1);
show_targ_out(database,Outliers,n,h,'Outliers');
return
function pushbutton6_Callback(hObject, eventdata, handles)
% Reset
s = guidata(hObject);
h = s{1};
database = s{2};
set(h.targets,'userdata',[]);
set(h.outliers,'userdata',[]);
show_targ_out(database,[],0,h,'Targets');
show_targ_out(database,[],0,h,'Outliers');
set(h.classify,'userdata','reset');
uiresume
return
function pushbutton7_Callback(hObject, eventdata, handles)
% Quit
guidata(hObject,[]);
uiresume
return
function radiobutton4_Callback(hObject, eventdata, handles)
% All
s = guidata(hObject);
h = s{1};
database = s{2};
if (get(h.all,'value') == 0)
set(h.unlabeled,'value',2);
else
set(h.unlabeled,'value',0);
end
t = get(h.all,'userdata');
S = t{1};
show_images(database,S,1,h)
return
function radiobutton5_Callback(hObject, eventdata, handles)
% unlabeled
s = guidata(hObject);
h = s{1};
database = s{2};
if (get(h.unlabeled,'value') == 0)
set(h.all,'value',2);
else
set(h.all,'value',0);
end
t = get(h.all,'userdata');
S = t{1};
U = t{2};
S = S(U);
show_images(database,S,1,h)
return
function pushbutton8_Callback(hObject, eventdata, handles)
% Next
s = guidata(hObject);
h = s{1};
database = s{2};
t = get(h.next,'userdata');
S = t{1};
n = t{2};
n = min(length(S)-9,n+10);
show_images(database,S,n,h);
return
function pushbutton9_Callback(hObject, eventdata, handles)
% Previous
s = guidata(hObject);
h = s{1};
database = s{2};
t = get(h.next,'userdata');
S = t{1};
n = t{2};
n = max(1,n-10);
show_images(database,S,n,h);
return
function pushbutton10_Callback(hObject, eventdata, handles)
% Label
s = guidata(hObject);
h = s{1};
set(h.classify,'userdata','label');
uiresume
return
function pushbutton11_Callback(hObject, eventdata, handles)
% All Target
s = guidata(hObject);
h = s{1};
for j=1:10
set(h.handlab(j),'value',1);
end
return
function pushbutton12_Callback(hObject, eventdata, handles)
% All Outlier
s = guidata(hObject);
h = s{1};
for j=1:10
set(h.handlab(j),'value',0);
end
return