Research Of Face Recognition Technology

Face recognition occurs as a new biometric identification technique with the rapid development of image processing, computer vision, pattern recognition and so on in recent years. The so-called biometric identification refers to automatic identification recognition by acquiring and analyzing the body's physical and behavioral characteristics (such as iris, fingerprint, voice, handwritten signature, gait and so on). The face recognition is a technique that extracts visual features, and distinguishes one face from another based on these features. Compared to other biometrics, face recognition technique is simpler, more intuitive, and having more hidden capability. Therefore, face recognition have great potential in a wide range of applications in information security, criminal investigation, public utilities and other fields .The thesis presents the framework of face recognition, and designs a face recognition system; implements main algorithms of face recognition, and validates the key algorithms (including feature extraction and classifier designing); compares the recognition results from different algorithms, and points out their strengths and weaknesses. Feature extraction is one of the most important step in the framework, which employs subspace algorithm and frequency domain feature extraction techniques. Subspace algorithms include PCA, 2DPCA, ICA, LDA, and NMF, etc, and frequency domain feature extraction includes WT and DCT. NNC and SVM are usually employed in this thesis for the task of classification. Subspace algorithms perform well in dimension reduction, which greatly decrease the memory space and increase recognition rate. Subspace methods narrow the training and testing time, moreover, with WT employed, better recognition rate is archived with rather lower training time. NNC is used due to the following advantages: simple, easy to implement, high efficiency, well performance. Comparably, SVM has especially better performance in small samples, non-linear cases and high dimension.