Efficient facial recognition using vector quantization of 2D DWT features

A new approach for facial recognition employing Two Dimensional Discrete Wavelet Transform (2D DWT) and Vector Quantization (VQ) is proposed. preprocessing, feature extraction, and classification are the three main phases in this paper. A cropping technique and appropriate dimensions selection are employed in the preprocessing step. In the feature extraction step, 2D DWT is applied to the processed facial images for dimensionality reduction and feature extraction. Then the VQ algorithm using Kekre Fast Codebook Generation (KFCG) for codebook initialization is implemented to the resultant DWT features for further feature compression and better facial representation. The proposed algorithm is evaluated using four databases, namely, ORL, YALE, FERET, and FEI that have different facial variations, such as facial expressions, illumination, rotations, etc. Then the experimental results are analyzed using K-fold Cross Validation (CV). The results show that the proposed approach improves the recognition rates and reduces the storage requirements compared with existing methods.