الفهرس | Only 14 pages are availabe for public view |
Abstract Face recognition consists of two essential parts: (1) feature extraction from facial images; and (2) classification. Due to this importance of feature extraction, the main contributions of this thesis are in this direction. This thesis presents an efficient feature extraction method based on local statistics features of BDIP (Block Difference of Inverse Probabilities) and the wavelet transform is proposed for face recognition. A second aspect of this work involves the new feature extraction method based on the Embedded Zero-tree of the Discrete Cosine Transform (DCT) is proposed and named EZDCT. Two techniques are used to compute the recognition rates. The first one is to use the Euclidean Distance (EUD). The second is the Support Vector Machine (SVM) which is implemented on both authentication and identification applications. For authentication, the classifier is designed to maximize the verification performance then the results obtained by applying Receiver Operating Characteristic (ROC). The performance of the proposed features extraction methods are evaluated using two databases ORL and FERET. Experimental results show that the proposed methods achieve higher recognition accuracies with higher dimensionality reduction of the feature vector. It is also outperforms the other well known methods such as Principle Component Analysis (PCA) and DCT with the zigzag scan. |