Enhancement of hybrid face recognition by incorporating half face features

• Main Author: Munira Hamdan
• Format: Thesis
• Language: English
• Published: 2015
• Online Access: https://eprints.ums.edu.my/id/eprint/27034/1/Enhancement%20of%20hybrid%20face%20recognition%20by%20incorporating%20half%20face%20features.pdf; https://eprints.ums.edu.my/id/eprint/27034/

Face recognition is a fast growing research field because of its potential application as an important tool for security surveillance, human-computer interaction, biometric and other fields. Face recognition techniques can be categorized into 3 categories namely; global approach, local approach, and hybrid approach. While the global approach can be represented using the full face feature, there are many ways to select the local features. In this thesis, a comparison of different local features selection methods is carried out. In the first technique, the geometric facial features of the face such as the eyes, nose and mouth are used. In the second method, a grid that divides a full face image into smaller sub-components with no regard of the locations of the geometric features is used. While in the third method, the half face image is used for face recognition. For these three methods, the Linear Discriminant Analysis (LDA) is used in the classification process. Finally the Bunch Graph method is used for the fourth local approach. Since the full face image may not always be available for face recognition especially during real life surveillance and some parts of the face may become occluded. By using the half face feature, a new full face image can be constructed by making use of the symmetrical property of the human face. A fusion method that fuses the global feature and the local features has also been investigated. These approaches were tested on the FERET database. It was found that the Bunch Graph method gave the highest ECR of 97% while the fusion of geometric feature gave the lowest ECR with 36% on the mouth feature. The global approach of the original full face performed better than the newly constructed full face. This is due to the fact that the half face was obtained automatically from the full face image by dividing into half directly. It was also found that fusing the global and local approach gave higher recognition rate than each individual approach.