cvLabMATLAB/CvtCMSM_revise.m at main · ComputerVisionLaboratory/cvLabMATLAB · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
classdef CvtCMSM_revise
    properties
        DicV
        DifDicV
        DifV
    end

    methods
        function obj = train(obj, train_data, subspace_dim, del_subspace_dim)
            % train_data:学習データ
            % subspace_dim:辞書部分空間の次元数
            % del_subspace_dim:差分部分空間を作るときに消す次元数


            [dim, numPic, class_num] = size(train_data);
%             主成分分析によって各クラスの部分空間を生成
			for I=1:class_num
                [obj.DicV(:,:,I), ~] = cvtBasisVectorSVD(train_data(:,:,I), subspace_dim);
            end


            % Generalizated difference subspace(Constraint Subspace)
            P = zeros(dim, dim);
            for I=1:class_num
                P = P + abs(obj.DicV(:,:,I)*obj.DicV(:,:,I)');
            end
            [B, C] = eig(P);
            C = diag(C);
            [~, index] = sort(C,'descend');
            B = B(:,index); C = C(index);

            obj.DifV = B(:,del_subspace_dim+1:rank(P))';

            for I=1:class_num
                obj.DifDicV(:,:,I) = obj.DifV*obj.DicV(:,:,I);
            end
        end

        function [sim] = predict(obj, data, subspace_dim)
            % data:入力データ size:データの次元x各データの個数xデータの個数
            % subpsace_dim:入力の部分空間の次元

            [data_dim, ~,data_num] = size(data);
            data = cvtNormalize(data);

            inpV = zeros(data_dim, subspace_dim, data_num);
            inpDifV = zeros(size(obj.DifV,1), subspace_dim, data_num);
            for di = 1:data_num
                inpV(:,:,di) = cvtBasisVectorSVD(data(:,:,di), subspace_dim);
                inpDifV(:,:,di) = orth(obj.DifV*inpV(:,:,di));
            end

            num_class = size(obj.DicV, 3);
            sim = zeros(num_class, data_num);
			sim = orzCanonicalAngles(obj.DifDicV, inpDifV);

%             for di = 1:data_num
%                 for ci = 1:num_class
%                     % cos( Canonical Angles )
%                     cos_angle = svd(obj.DifDicV(:,:,ci)'*inpDifV(:,:,di));
%                     %cos_angle = svd(inpV*obj.V(:,:,I)*inpV);
%
%                     % similarity = mean(cos^2(c_angles))
%                     sim(ci,di) = mean(cos_angle.^2);
%                 end
%             end
        end
    end


end