【正文】 r s=1: for j=1:sizeTrain count = count + 1。 Xtr(count) = (*(s1))+imageNo(s,j)。 endend%=====================================================% Allocating images into training and testing datasets%=====================================================XX1 = allImages。 % attention!!!trainImages = allImages(:,Xtr)。 XX1(:,Xtr) = []。testImages = XX1。 clear XX1。%===========================================% Display the trainImages and testImages%===========================================visual_switch = 。if(visual_switch == 1) figure。 for s=1: for j=1:sizeTrain tempImage = trainImages(:,(s1)*sizeTrain+j)。 tempImage = uint8(reshape(tempImage,))。 subplot(3,4,j)。 imshow(tempImage)。 title([39。Training Face 39。,39。{39。,num2str(s),39。}39。])。 end for k=1: tempImage = testImages(:,(s1)*+k)。 tempImage = uint8(reshape(tempImage,))。 subplot(3,4,k+sizeTrain)。 imshow(tempImage)。 title([39。Testing Face 39。,39。{39。,num2str(s),39。}39。])。 end pause。 endendfunction [drImages,eigenInfo] = PCAdr(trainImages,dbaseInfo)% get pressed images (drImages) by PCA% Calculate the meanImage of trainImages[rw,trainImagesTotal] = size(trainImages)。 = sum(trainImages,2) / trainImagesTotal。imagesCentered = zeros(rw,trainImagesTotal)。for i=1: trainImagesTotal imagesCentered(:,i) = trainImages(:,i) 。end % Calculate the eigenvectors and eigenvalues cov = (imagesCentered39。 * imagesCentered) / trainImagesTotal。 [Vec,Vals,e] = svd(cov)。%Calaulate eigenface = imagesCentered * Vec(:,1:)。drImages = 39。 * imagesCentered。function [wopt] = LDA(dbaseInfo,drImages)%pute the project matrix of LDA[rw,cl] = size(drImages)。mean = sum(drImages,2) / cl。mean = repmat(mean,1,)。 c_mean = zeros(rw,)。for i = 1: head = (i1)* + 1。 tail = i*。 c_mean(:,i) = sum(drImages(:,head:tail),2) / 。 end SB = * (c_mean mean) * (c_mean mean)39。 drImagesCentered = zeros(rw,cl)。for i = 1:cl j = ceil(i / )。 drImagesCentered(:,i) = drImages(:,i) c_mean(:,j)。endSW = drImagesCentered * drImagesCentered39。 [Vector,Value] = eig( pinv( SW ) * SB )。Value = diag(Value)。[Value,index] = sort(abs(Value),39。descend39。)。Vector = Vector(:,index)。 wopt = Vector(:,1:)。function [membership] = FKNN(drImages,dbaseInfo)%pute the membership degree for every trainImage[rw,cl] = size(drImages)。distance = zeros(cl,cl)。for i = 1:cl for j = 1:cl distance(i,j) = norm(drImages(:,i) drImages(:,j))。 endendfor i = 1:cl distance(i,i) = Inf。end[distance,index] = sort(distance)。index = index(1:,:)。 counter = zeros(,cl)。for i = 1:cl class = ceil(index(:,i) / )。 for j = 1: counter(class(j),i) = counter(class(j),i) +1。 endendmembership = zeros(,cl)。for i = 1:cl temp = ceil(i/)。 for j = 1: if j==temp membership(j,i) = + * counter(j,i) / 。 else membership(j,i) = * counter(j,i) / 。 end endendfunction [wopt] = FFLD(dbaseInfo,drImages,membership)%pute the project matrix of FFLD[rw,cl] = size(drImages)。mean = sum(drImages,2) / cl。mean = repmat(mean,1,)。 c_mean = zeros(rw,)。for i = 1: temp = sum(membership(i,:))。 c_mean(:,i) = drImages * membership(i,:)39。 c_mean(:,i) = c_mean(:,i) / temp。endSFB = * (c_mean mean) * (c_mean mean)39。 drImagesCentered = zeros(rw,cl)。for i = 1:cl j = ceil(i / )。 drImagesCentered(:,i) = drImages(:,i) c_mean(:,j)。endSFW = drImagesCentered * drImagesCentered39。 [Vector,Value] = eig( pinv( SFW ) * SFB )。Value = diag(Value)。[Value,index] = sort(abs(Value),39。descend39。)。Vector = Vector(:,index)。 wopt = Vector(:,1:)。function [precision] = recognize(dbaseInfo,drImages,testImages,eigenInfo,wopt) % This function will calculate the precision statistics on a given database NCorrect = 0。projTrain = wopt39。 * drImages。%================================% Testing each of the test images%================================ [rw,testImagesTotal] = size(testImages)。[rw,trainImagesTotal] = size(projTrain)。for i=1: testImagesTotal % Center the test image and project into face space meanTest = testImages(:,i) 。 projTest = wopt39。 * (39。 * meanTest)。 dist = zeros(1,trainImagesTotal)。 for j = 1: trainImagesTotal test = projTrain(:,j) projTest。 dist(j) = norm(test)。 end [prevBest,match] = min(dist)。 counterTmp = ceil(i / )。 match2 = ceil(match / )。 if (match2 == counterTmp) NCorrect = NCorrect + 1。 endendprecision = NCorrect / testImagesTotal*100。在ORL人臉庫(kù)中：clear all。clc。 % The information on the training and testing datasets are specified as below:%================%ORL%================ = 39。39。 = 400。 = 40。 = 10。 = 112。 = 92。 %==============================% Training Database information%============================== = 0。 = 6。 = 。 = * 。 % Rank(Sw) = Nc。 = 1。 % Rank(Sb) = c1。 = 80。%================================================% Reading image%================================================allFid = fopen(,39。r39。)。display(39。Read All Images ....39。)。 allImages = zeros( * ,)。for j=1: imageName = fgetl(allFid)。 tempImage = double(imread(imageName))。 allImages(:,j) = reshape(tempImage,*,1)。endclear tempImage。display(39。Done39。)。fclose(allFid)。 %================================================% Simulation for k^th times%================================================simuTime = 50。precisionLDA = zeros(simuTime,1)。precisionFLDA = zeros(simuTime,1)。for t = 1:simuTime