Face Recognition Using Principal Component Analysis With Support Vector Machine Classifier

Andy, Low Foo Hwa (2015) Face Recognition Using Principal Component Analysis With Support Vector Machine Classifier. Project Report. UTeM, Melaka, Malaysia. (Submitted)

Text (24 Pages) Face Recognition Using Principal Component Analysis With Support Vector Machine Classifier 24 Pages.pdf - Submitted Version Download (228kB)

Abstract

An efficient biometrics identification system is important to identify the identity of an individual. Biometrics includes human face, fingerprint, iris and others. In this project, human face recognition has been chosen to be used in the system. Principal Component Analysis (PCA) has been selected to be the algorithm working with the system. However, due to the fact that PCA itself is not a good classifier so a classifier is picked to classify the data which is Support Vector Machine (SVM) classifier. Basically, the main idea of PCA is to reduce the high dimensional image space to a lower one, as known as dimensionality reduction. During the process of reducing dimension, each of the images is being treated as an n*n matrix and the algorithm will calculate the mean face of all the individuals; subsequently create a lower dimensional face space. When a sample test image is feed into the system to perform matching, the test image will be projected onto the face space and perform matching. Then, SVM classifier will classify which class the individual belongs to. In order to test the system, an Adaboost classifier has been picked to compare the accuracy. As a result, SVM classifier outperforms the Adaboost classifier. In general, the graphical user interface is user friendly and user just need to train the system, browse a test image and then perform recognition. Although there are some weaknesses of this system, it can still be improved by hybrid with other efficient algorithm to achieve a more accurate result. Further research needs to be carry out in order to improve and enhance the efficiency and performance of the system.

Actions (login required)

View Item