Informatics Publishing Limited

S. A. Naqvi ¹, Waqas Rehan ¹, F. M. Zafar ²

Affiliations
1 DEE, FET, IIUI, PK
2 ICCC, PK

Abstract

Face recognition has become a very important research area of image processing, which is used in a variety of fields and disciplines. The paper presents a new model for face recognition. The model uses a hybrid color space (HCS) RIQ, which has been constructed by combining color components of RGB color space and YIQ color space. For feature reduction the model uses Fisher Linear Discriminant (FLD) method along with Face Symmetry Property (FSP). When FLD was combined with FSP, the model was named as FLD-FSP. The performance of the FLD-FSP model was estimated using face images of 64×64 resolution. The face images were extracted by employing Viola John face detector. The results show that the system can perform considerably well in real world controlled environments.

Keywords

Face Recognition, Hybrid Color Space (HCS), Fisher Linear Dicriminant (FLD), Face Symmetry Property (FSP), Recognition Error Rate (RER).

Full Text

     

Waqas Rehan ¹, Ali Naqvi ¹, Faisal Zafar ², Mirza Zain ¹

Affiliations
1 DEE, IIUI, PK
2 ICCC, PK

Abstract

Face recognition has become an actively researched area due to its applications like security systems, criminal identification, and credit card verification. Unfortunately, expressing face images mathematically is quite difficult because faces are complex visual stimuli. In recognition systems there is a problem of minimizing the dimensionality of extracted features from face images. This paper presents the work done on a face recognition system, which uses Hybrid Color space (HCS) and Principal Component Analysis (PCA) method, along with proposed Face Symmetry Property (FSP) for recognition of human face. Color provides useful information for face recognition system so a RCbCr hybrid color space has been proposed. Less computation increases the efficiency of any recognition system, thus computational cost has been reduced half by using an interesting and simple innovation FSP. In addition to FSP, PCA method is employed for improving the generalization capability. Experiments were conducted on 64×64 resolution using various criteria of recognition. Experiments using PCA with and without FSP are conducted. The proposed model which used PCA with FSP was named as PCA-FSP. When using PCA-FSP, the resolution of the images became 64×32. It was seen that the concept of FSP is very effective for reduction of computational cost, storage space, and system complexity. Performance comparison of the proposed system(PCA-FSP) was made with simple PCA. Results suggest that the proposed system shows satisfactory performance.

Keywords

Face Symmetry Property (FSP), Principal Component Analysis (PCA), Face Recognition, Hybrid Color Space (HCS), Recognition Error Rate (RER).

Full Text