Nearest-Neighbour-Search-Neural-Networks-Product-Quantization/MNIST/Euclidean NNS/EuclideanNNS.m

clear all
close all
clc

addpath('../Functions');

imgSz=28;           % Lateral Image pixel size
imgSzSq=imgSz^2;    % Square Image pixel size
Ntr=6e4;            % Number of pictures train set
Ntests=1e4;         % Number of pictures test set
fileOffsetPic=16;   % Bytes of pictures file offset
fileOffsetLab=8;    % Bytes of label file offset
ENNSon=0;           % 

%% Reading Binary Training Images and Labels

[XSet, XLab]=ReadingPicSetLabSet(imgSz, Ntr, fileOffsetPic, '../Data/train-images.idx3-ubyte',...
                                  fileOffsetLab, '../Data/train-labels.idx1-ubyte', 0);

%% Reading Binary Testing Images and Labels
[YSet, YLab]=ReadingPicSetLabSet(imgSz, Ntests, fileOffsetPic, '../Data/t10k-images.idx3-ubyte',...
                                  fileOffsetLab, '../Data/t10k-labels.idx1-ubyte', 0);

%% Computing NNS with Euclidean Distance

if(ENNSon)
    fprintf('Computing Euclidean NNS: ...    ');

    dEuc=zeros(1,Ntr);
    indmin=zeros(1,Ntests);

    for ii=1:Ntests
        for jj=1:Ntr
            Xjj=XSet(imgSzSq*(jj-1)+1:imgSzSq*jj);
            Yii=YSet(imgSzSq*(ii-1)+1:imgSzSq*ii);
            dEuc(jj)=sum((Xjj-Yii).^2);
        end
        [~,indmin(ii)]=min(dEuc);
        fprintf('\b\b\b\b\b\b%3.2f %%',ii/Ntests*100);
    end
else
    fprintf('Loading Euclidean NNS results ...');
    load('indminEuNNS.mat')
end

%% Plot some of the matches
%Nstart=9983;
%for kk=Nstart+1:Nstart+10%Ntests
    %subplot(2,10,kk-Nstart)
    %imshow255(0,imgSz,YSet')
    %xlabel(int2str(YLab(kk)));
    %title('Test Image')
    %subplot(2,10,kk-Nstart+10)
    %figure()
    %imshow255(0,imgSz,XSet(indmin(1))')
    %xlabel(int2str(XLab(indmin(kk))));
    %title('Match Image')
%end

NErrors=Ntests-sum(XLab(indmin)==YLab(1:length(indmin)));
Performances=100-NErrors/Ntests*100