Artificial Intelligence Project: PCA for Face Recognition

This project implements a face recognition system using Principal Component Analysis (PCA) and machine learning techniques. The assignment begins with an introduction to PCA, explaining its function in reducing data dimensionality and identifying principal components. The core of the project focuses on face recognition, detailing the application of PCA for this purpose. The project utilizes MATLAB for code implementation, with screenshots provided to demonstrate the program's execution and results. Several datasets are selected and used, including the Bio ID Face Database and FaceScrub, and the reasons for choosing these datasets are explained. The project also includes visualizations after the transformation process. The conclusion summarizes the project's findings, emphasizing the effectiveness of PCA and machine learning in face recognition. References to the sources used are also included.

PRINCIPAL COMPONENT ANALYSIS

Table of Contents
1. Introduction............................................................................................................................2
2. PCA..........................................................................................................................................2
3. Face recognition......................................................................................................................3
4. Datasets....................................................................................................................................3
5. Screenshots of running your program..................................................................................3
6. Visualization after transform................................................................................................4
7. Conclusions.............................................................................................................................5
1

1. Introduction
Principal Component Analysis (PCA) which is a multivariate procedure that examines an
information table in which perceptions are portrayed by a few between connected quantitative
subordinate factors will be used to trace the face recognition schemes using the datasets.
Machine learning will also be used to aid PCA. Machine learning which has given us reasonable
discourse salutation, compelling web seek, and a limitlessly enhanced comprehension of the
social genome will be used along with PCA to implement the face recognition process
2. PCA
PCA is nothing but Principal Component Analysis ("Principal Component Analysis
explained visually", 2017). It will likely concentrate the essential data from the table, to speak to
it as an arrangement of new orthogonal factors called primary segments, and to show the
example of comparability of the perceptions and of the factors as focuses in maps. The nature of
the PCA model can be assessed utilizing cross-approval systems, for example, the bootstrap and
the folding blade. PCA can be summed up as correspondence investigation (CA) so as to deal
with subjective factors and as different factor examination (MFA) with a specific end goal to
deal with heterogeneous arrangements of factors.
To start with, consider a dataset in just two measurements, similar to (stature, weight).
This dataset can be plotted as focuses in a plane. In any case, in the event that we need to coax
out variety, PCA finds another facilitate framework in which each point has another (x,y)
esteem. The tomahawks don't really mean anything physical; they're blends of stature and weight
called "primary segments" that are given one tomahawks loads of variety. Drag the focuses
around in the accompanying perception to see PC facilitate framework changes. With three
measurements, PCA is more helpful, in light of the fact that it's difficult to see through a billow
of information. In the case underneath, the first information are plotted in 3D, yet you can extend
the information into 2D through a change the same than finding a camera point: pivot the
tomahawks to locate the best edge. To see the "authority" PCA change, tap the "Show PCA"
catch. The PCA change guarantees that the flat pivot PC1 has the most variety, the vertical hub
PC2 the second-most, and a third hub PC3 the slightest. Clearly, PC3 is the one we drop.
3. Face recognition
Face is a compound multidimensional assembly and needs great recording approaches for
acknowledgment. The face is our important and first awareness of deliberation in social life
pretentious a domineering part in appeal of person. Here Face recognition is done using PCA.
Entire work is done in MATLAB 2013B ("Principal component analysis of raw data - MATLAB
pca - MathWorks United Kingdom", 2017).
2

Face Detection Database
Selected images from https://archive.ics.uci.edu/ml/databases/faces/
http://archive.ics.uci.edu/ml/datasets/cmu+face+images
https://archive.ics.uci.edu/ml/datasets/Grammatical+Facial+Expressions
Reasons for chosing it
The Bio ID Face Database has been recorded and is distributed to give all analysts
working in the zone of face identification the likelihood to think about the nature of their face
discovery calculations with others. It might be utilized for such purposes without encourage
authorization. Amid the chronicle uncommon accentuation has been set on "certifiable"
conditions. Subsequently the test set highlights a substantial assortment of brightening,
foundation, and face measure
FaceScrub – A Dataset With Over 100,000 Face Images of 530 People
Downloaded the dataset from the university websites and tried to run full dataset. The code is
taking lot of time to give results. Hence decreased the images and created a small dataset.
Reasons for choosing it
The face detection is a well-known problem that can be investigated using PCA. PCA is
unsupervised approach where as another similar approach LDA is supervised. PCA gives better
classification results if the number of datasets is relatively small. PCA considers entire dataset as
a whole. If the dataset is big then the processing time and finding the results will take lot of time.
LDA will create classes within the data. It won’t consider the full dataset as a whole.
3

As shown above the PCA finds the way where maximum variance happens.
4

LDA does class severability.
Both PCA and LDA uses mean and co-variance matrix an as important parameters.
PCA is a theoretical matrix and practical statistics. Image processing is equal to
investigating multi variable data sets. Getting Eigen values and Eigen vectors will be easy if
most of the data are fully co-related. Finally few Eigen vector values are sufficient to represent
the full dataset.
5

4. Screenshots of running your program
5. Visualization after transform.
6

6. Conclusions
Face recognition is carried out with the help of Principal Component Analysis and machine
learning techniques. The algorithm for face recognition is written and code is formulated using
MATLAB. Popular datasets are chosen and are used for executing the code.
7. References
Principal Component Analysis explained visually. (2017). Explained Visually. Retrieved 2
October 2017, from http://setosa.io/ev/principal-component-analysis/
Principal component analysis of raw data - MATLAB pca - MathWorks United Kingdom.
(2017).In.mathworks.com. Retrieved 2 October 2017, from
https://in.mathworks.com/help/stats/pca.html?requestedDomain=www.mathworks.com
7