[pattern recognition] PCA handwritten numeral recognition based on MATLAB [including Matlab source code 309]

1, Introduction

PCA algorithm is based on image reconstruction method for image feature recognition. There are training samples, multiple test pictures and document descriptions.
Identification steps:
① Select training samples
② Calculate sample average digital feature, digital feature space
③ Read the number to be identified, segment the connected components, and determine the number of numbers to be identified
④ Classification by discriminant

2, Source code

clear all
clc
close all

% Select training data and test data path (i.e. directory) TrainData and TestData)
TrainDatabasePath = uigetdir('D:\Program Files\MATLAB\R2007b\work', 'choice[Training data]route' );
TestDatabasePath = uigetdir('D:\Program Files\MATLAB\R2007b\work', 'choice[test data]route');

prompt = {'Enter test image name(1,2):'};
dlg_title = 'PCA Identification input';
num_lines= 1;
def = {'1'};

TestImage  = inputdlg(prompt,dlg_title,num_lines,def);
TestImage = strcat(TestDatabasePath,'\',char(TestImage),'.jpg');
im = imread(TestImage);

T = CreateDatabase(TrainDatabasePath);%Create test database
[m, A, EigenPos] = PCA(T);
OutputName = Recognition(TestImage, m, A, EigenPos);%Recognition output matching image

SelectedImage = strcat(TrainDatabasePath,'\',OutputName);
SelectedImage = imread(SelectedImage);

imshow(im)
title('Test image');
figure,imshow(SelectedImage);
title('Equivalent image');
unction OutputName = Recognition(TestImage, m, A, EigenPos)
% Recognition operation:Compare the two images, map the image to the feature space, and measure the Euclidean distance between them
% parameter:TestImage Enter test image path
%       m   (M*Nx1) Mean value of training database
%      EigenPos   (M*Nx(P-1)) Covariance matrix eigenvector of training data
%       A     (M*NxP) Centered image vector matrix 
% return: OutputName  The names in the training database are recognized            

% All centered images are multiplied by Eigenfaces To map, the mapping vector of each graph will be the corresponding feature vector
ProjectedImages = [];
Train_Number = size(EigenPos,2);
for i = 1 : Train_Number
    temp = EigenPos'*A(:,i); %Centered image mapping
    ProjectedImages = [ProjectedImages temp]; 
end

% Extract from the test image PCA features
InputImage = imread(TestImage);
temp = InputImage(:,:,1);

[irow icol] = size(temp);
Difference = double(InImage)-m; % Centered test image
ProjectedTestImage = EigenPos'*Difference; % Test image feature vector

% Calculate the Euclidean distance between the mapped test image and the projection of all centered training images
% Set the minimum distance between the measurement image and the corresponding image in the training database
Euc_dist = [];
for i = 1 : Train_Number
    q = ProjectedImages(:,i);
    temp = ( norm( ProjectedTestImage - q ) )^2;
    Euc_dist = [Euc_dist temp];
end

3, Operation results

4, Remarks

Tags: MATLAB image processing

Posted by Tarsonis21 on Tue, 19 Apr 2022 00:22:58 +0930

Programs Wiki

[pattern recognition] PCA handwritten numeral recognition based on MATLAB [including Matlab source code 309]

1, Introduction

2, Source code

3, Operation results

4, Remarks

Popular Tags