Add files via upload

DBD.m

@@ -0,0 +1,41 @@
+function [W] = DBD(X, sw, sb, params)
+% X：MDPDV数据
+% K：取特征向量的个数
+% Loop_num：循环次数
+
+lambda1 = 1e-3;       % 论文数值
+lambda2 = 1e-4;    % 论文数值
+lambda3 = 0.01;       % 论文数值
+opts.record = 0;
+opts.mxitr  = 3000;
+opts.xtol = 1e-5;
+opts.gtol = 1e-5;
+opts.ftol = 1e-8;
+K = params.K;            % 论文数值
+[n_Sample, n_Feature] = size(X);
+M = mean(X,1);
+data = (X - repmat(M,n_Sample,1));
+[eigenvector, eigenvalue] = eig(data'*data);  % 求特征向量、特征值
+[value, index]= sort(diag(eigenvalue),'descend');
+W0 = eigenvector(:, index(1:K));   % 取前k个特征向量初始化W
+
+Loop_num = 100;
+W = W0;
+for i = 1:Loop_num
+    % 固定W，更新B
+    B = double(X*W>0);
+
+    % 固定B，更新W
+    F1 = lambda1*(X'*X) + lambda2* sw - lambda3*sb;
+    F2 = -2*lambda1*X'*B;
+    [W, G]= OptStiefelGBB(W, @objectfunc,opts,F1,F2');
+end
+end
+
+function [F, G] = objectfunc(W, F1, F2)
+    F = trace(W'*F1*W) + trace(F2*W);
+    G = 2*F1*W + F2';
+end
+
+
+

Extract_feature.m

@@ -0,0 +1,58 @@
+function fea = Extract_feature(img,W,D,params)
+
+clusters = params.clusters;
+M = params.M;
+N = params.N;
+
+Tdata = double(img);
+[h, w] = size(img);
+
+R = params.R;
+spoints = [];
+for i=-R:R
+    spoints=[spoints;0,i];
+end
+
+for i=-R:R
+    spoints = [spoints;i,0];
+end
+for i=-R:R
+    spoints = [spoints;i,i];
+end
+for i=-R:R
+    spoints = [spoints;i,-i];
+end
+spoints(R+1,:)=[];
+spoints(3*R+1,:)=[];
+spoints(5*R+1,:)=[];
+spoints(7*R+1,:)=[];
+CG = Tdata((R+1):(h-R), (R+1):(w-R));
+Tcode1 = zeros(size(CG,1),size(CG,2),size(spoints,1));
+for ii=1:size(spoints,1)
+    Tmp = Tdata(((R+1):(h-R))+spoints(ii,1),((R+1):(w-R))+spoints(ii,2));
+    Tcode1(:, :, ii) = Tmp-CG;
+end
+
+[Th, Tw, ~] = size(Tcode1);
+delta_h = round(Th / M);
+delta_w = round(Tw / N);
+
+fea = zeros(1,M*N*clusters);
+cnt=0;
+for i = 1:M
+    s_h = delta_h * (i-1) + 1;
+    e_h = s_h + delta_h - 1;
+    e_h = min(e_h, Th);
+    for j = 1:N
+        s_w = delta_w * (j-1) + 1;
+        e_w = s_w + delta_w - 1;
+        e_w = min(e_w, Tw);
+        Tmp = Tcode1(s_h:e_h, s_w:e_w, :);
+        [t_h, t_w, t_d] = size(Tmp);
+        Tmp = reshape(Tmp, t_h*t_w, t_d);
+
+        fea((cnt+1):cnt+clusters)= ...
+          getHist(double(Tmp*W{i,j} > 0), D{i,j});
+        cnt = cnt+clusters;
+    end
+end

Loda_data.m

@@ -0,0 +1,75 @@
+clear all;
+clc;
+
+register_path='.\data\Register\';
+test_path='.\data\TestImage\';
+register_image_num = 360;   % 注册图片总数
+test_image_num = 525;   % 测试图片总数
+width = 96;
+height = 64;
+
+
+data = zeros( height, width, register_image_num);
+label = cell(1, register_image_num);
+register_fingerID = dir(register_path);   % 获取文件夹
+N=size(register_fingerID,1);  
+% CLASSES = N - 2;
+num = 0;
+for i = 3:N
+    fingerdir=strcat(register_path,register_fingerID(i,1).name,'\');
+    redister_IDname = register_fingerID(i, 1).name;
+    file=dir(fingerdir);    % 获取图片
+    N1=size(file,1);  
+
+    pre_class = N1 - 2;  % 每个ID 图片的个数
+    for j=3:N1
+        filename=strcat(fingerdir,file(j,1).name);
+        img= imread(filename);
+
+        img = rgb2gray(img);
+%         figure(1);
+%         imshow(img);
+%         img = imresize(img,[height, width], 'bicubic');
+        img = double(img);
+%         figure(2);
+%         imshow(img);
+        data(:, :, num+j-2) = img;
+        label{1, num+j-2} = redister_IDname;
+    end   
+    num = num + N1 - 2;
+end
+save train data label
+
+
+data = zeros( height, width, test_image_num);
+label = cell(1, test_image_num);
+test_fingerID = dir(test_path);   % 获取文件夹
+N=size(test_fingerID,1);  
+% CLASSES = N - 2;
+
+num_2 = 0;
+for i = 3:N
+    fingerdir=strcat(test_path,test_fingerID(i,1).name,'\');
+    test_IDname = test_fingerID(i, 1).name;
+    file=dir(fingerdir);    % 获取图片
+    N1=size(file,1);  
+
+    pre_class = N1 - 2;  % 每个ID 图片的个数
+    for j=3:N1
+        filename=strcat(fingerdir,file(j,1).name);
+        img= imread(filename);
+
+        img = rgb2gray(img);
+        img = double(img);
+        data(:, :, num_2+j-2) = img;
+        label{1, num_2+j-2} = test_IDname;
+    end   
+    num_2 = num_2 + N1 - 2;
+end
+save test data label
+
+
+
+
+
+

MDPDV_extract.m

@@ -0,0 +1,121 @@
+function [MDPDV, sw, sb] =MDPDV_extract(data, param)
+M = param.M;
+N = param.N;
+train_num = param.train_num;
+block_w = param.block_w;
+block_h = param.block_h;
+stride_h = param.stride_h;
+stride_w = param.stride_w;
+image_num = size(data,3);
+% 暂定窗口的R为3，则线上的像素点为24个
+R = param.R;
+Th = size(data(:,:,1), 1); % 图片的高
+Tw = size(data(:,:,1), 2); % 图片的宽
+spoints = [];
+for i=-R:R
+    spoints=[spoints;0,i];
+end
+
+for i=-R:R
+    spoints = [spoints;i,0];
+end
+for i=-R:R
+    spoints = [spoints;i,i];
+end
+for i=-R:R
+    spoints = [spoints;i,-i];
+end
+spoints(R+1,:)=[];
+spoints(3*R+1,:)=[];
+spoints(5*R+1,:)=[];
+spoints(7*R+1,:)=[];
+
+% spoints= [0,-8;0,-7;0,-6;0,-5;0,-4;0,-3;0,-2;0,-1;0,1;0,2;0,3;0,4;0,5;0,6;0,7;0,8;
+%           -8,0;-7,0;-6,0;-5,0;-4,0;-3,0;-2,0;-1,0;1,0;2,0;3,0;4,0;5,0;6,0;7,0;8,0;
+%           -8,-8;-7,-7;-6,-6;-5,-5;-4,-4;-3,-3;-2,-2;-1,-1;1,1;2,2;3,3;4,4;5,5;6,6;7,7;8,8;
+%           -8,8;-7,7;-6,6;-5,5;-4,4;-3,3;-2,2;-1,1;1,-1;2,-2;3,-3;4,-4;5,-5;6,-6;7,-7;8,-8];
+MDPDV = cell(M, N);  % M：行的分块数，N：列的分块数
+for i=1:M
+    for j=1:N
+        MDPDV{i,j} = zeros( block_w * block_h, size(spoints,1), size(data,3) ); % (窗宽*窗长*图片数， 64）
+    end
+end
+cnt = 0;
+
+% Tdata = zeros(Th+R*2, Tw+R*2);
+for d=1:size(data,3)
+    [h, w] = size(data(:,:,d));
+    Tdata = padarray(data(:,:,d), [R, R], 'replicate', 'both');
+    Tdata( (1+R):(h+R) , (1+R):(w+R) ) = double(data(:,:,d));  % 一张图片
+    %Tdata = preproc2(Tdata,0.2,1,2,[],[],10);
+
+    CG = Tdata((R+1):(h+R), (R+1):(w+R));  % 一张图片
+    Tcode1 = zeros(size(CG,1),size(CG,2),size(spoints,1)); % (图长，图宽，24)
+
+    for ii=1:size(spoints,1)   % 1:24
+        Tmp = Tdata(((R+1):(h+R))+spoints(ii,1),((R+1):(w+R))+spoints(ii,2));
+        Tcode1(:, :, ii) = Tmp-CG;
+    end
+
+    for i = 1:M
+        r = (i-1)*stride_h;
+        for j = 1:N
+            c = (j-1)*stride_w;
+            Tmp = Tcode1(r+1:r+block_h, c+1:c+block_w, :);
+%             MDPDV{i,j}(cnt+1:cnt+block_w * block_h,:) = ...
+%                 reshape(Tmp, block_w * block_h, size(spoints,1));
+            MDPDV{i,j}(:,:,d) = reshape(Tmp, block_w * block_h, size(spoints,1));
+        end
+    end
+end
+
+% 计算sw, sb
+sw = cell(M,N);  % 类内间距和
+sb = cell(M,N);  % 类间间距和
+m = cell(M,N);   % 所有样本的平均值
+% 计算所有样本的平均值
+for i=1:M
+    for j=1:N
+        m{i,j}=mean(MDPDV{i,j}, 3);
+    end
+end
+% 计算类间间距和
+
+for i=1:M
+    for j=1:N
+        sw{i,j} = zeros(size(spoints, 1),size(spoints, 1));
+        sb{i,j} = zeros(size(spoints, 1),size(spoints, 1));
+        for k = 1:train_num
+           sw{i,j} = sw{i,j} + (MDPDV{i,j}(:, :, k) - mean(MDPDV{i,j}(:,:, 1:train_num), 3))'*(MDPDV{i,j}(:, :, k) - mean(MDPDV{i,j}(:,:, 1:train_num), 3));
+        end
+        sb{i,j} =sb{i,j} + 6*(mean(MDPDV{i,j}(:,:,1:train_num), 3)-m{i,j})'*(mean(MDPDV{i,j}(:,:,1:train_num),3)-m{i,j});
+    end
+end
+
+for i=1:M
+    for j=1:N
+        for k = 2*train_num:train_num:image_num
+            for z = k-train_num+1:k
+                sw{i,j} = sw{i,j} + (MDPDV{i,j}(:, :, z) - mean(MDPDV{i,j}(:,:, k-train_num+1:k), 3))'*(MDPDV{i,j}(:, :, z) - mean(MDPDV{i,j}(:,:, k-train_num+1:k), 3));
+            end
+%             sw{i, j} = sw{i, j} + (bsxfun(@minus,MDPDV{i,j}(block_w*block_h*(k-2):block_w*block_h*k, :),mean(MDPDV{i,j}(block_w*block_h*(k-2):block_w*block_h*k, :))))'* (bsxfun(@minus,MDPDV{i,j}(block_w*block_h*(k-2):block_w*block_h*k, :),mean(MDPDV{i,j}(block_w*block_h*(k-2):block_w*block_h*k, :))));
+%             sw{i, j} = sw{i,j} + var(MDPDV{i, j}(block_w*block_h*(k-2):block_w*block_h*k, :), 0, 1);
+%             sb{i,j} =sb{i,j} + 3*(mean(MDPDV{i,j}(block_w*block_h*(k-2):block_w*block_h*k, :))-m{i,j}).^2;
+%             sb{i,j} = 3*(mean(MDPDV{i,j}(block_w*block_h*(k-2):block_w*block_h*k, :))-m{i,j})'*(mean(MDPDV{i,j}(block_w*block_h*(k-2):block_w*block_h*k, :))-m{i,j});
+            sb{i,j} = sb{i,j} + train_num*(mean(MDPDV{i,j}(:,:,k-train_num+1:k), 3)-m{i,j})'*(mean(MDPDV{i,j}(:,:,k-train_num+1:k),3)-m{i,j});
+        end
+    end
+end
+
+for i=1:M
+    for j=1:N
+        MDPDV{i,j} = permute(MDPDV{i,j},[2, 1, 3]);
+        MDPDV{i,j} = reshape(MDPDV{i,j},size(spoints,1),[])';
+    end
+end
+
+
+
+
+
+

MGramSchmidt.m

@@ -0,0 +1,16 @@
+function V = MGramSchmidt(V)
+[n,k] = size(V);
+
+for dj = 1:k
+    for di = 1:dj-1
+        V(:,dj) = V(:,dj) - proj(V(:,di), V(:,dj));
+    end
+    V(:,dj) = V(:,dj)/norm(V(:,dj));
+end
+end
+
+
+%project v onto u
+function v = proj(u,v)
+v = (dot(v,u)/dot(u,u))*u;
+end