Effect of averaging techniques on PCA algorithm and its performance evaluation in face recognition applications

Biometric authentication is a process that validates the user's identity by measuring his intrinsic characteristics. This process is used to cater for the security-related issues. There are many techniques of biometric identification, e.g., face recognition, fingerprint identification, retina and iris scanning. Face recognition is a simplest and efficient technique. It can be implemented through different methods, e.g., Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Discrete Cosine Transform (DCT) and Independent Component Analysis (ICA). PCA is the most practical method because of its accuracy, time limitations and process speed. This paper analyzes the performance of PCA using different averaging techniques and proposes useful recommendations that, once incorporated, will improve the accuracy and computational complexity of the algorithm under certain constraints. The proposed system has been programmed in MATLAB and tested on Olivetti Research Laboratory (ORL) database of faces. The performance of the modified PCA has been compared with the conventional PCA and simulation results have been discussed in terms of accuracy and complexity analysis. The proposed system with recommended changes has also been evaluated through hardware implementation, discussed at the end of this paper.