Study Of Face Recognition Technology

Human face recognition technology is a most challenging and hot subject in the field of pattern-recognition and vision of the machine. Up to now, there have many methods of recognition of face been proposed, experienced and even put into practical application. However, it can be found out from several respects of discerning rate, the complexity of calculation and resources taken up that the methods of recognition each have pluses and minuses. How to full play the advantage and avoid the shortcoming of the methods is our goal. The important contributions of the paper are displayed as following:1. A new human face recognition method based on eigenface analysis and wavelet transform was proposed. Firstly, some human face images were decomposed using wavelet transformation, and low frequency subbands and middle frequency subbands were extracted. For the low frequency subbands, an eigenface space and its feature-coefficient database were constructed. For the average middle frequency subbands, we construct another eigenface-space and its feature-coefficient database. Following, a face image was projected into the two spaces and two groups of feature-coefficients were calculated. They were compared with the two feature coefficient databases respectively to get two similarity matrixes. Finally, two matrixes with different weights for recognition were added. Eigenface analysis has many advantages, such as effectiveness and high recognition rate. Wavelet transform also has many advantages, such as multi-resolution analysis and multi-scale decomposition. This approach synthetically utilized these advantages and reasonably used twice addition to improve its performances. Experiment results indicate that this method can largely reduce computing complexity and has higher recognition rate. The results also show that this method has obvious potential in practical usage.2. This paper has proposed an method based on the discrete cosine transform(DCT) and neural networks, which extracts holistic and local DCT coefficients and feeds them to the multi-layer perception classifier. The experiment shows the method achieves a balanced tradeoff between recognition rate and speed.3. A new human face recognition based on ICA and improved BP neural networks was proposed,It uses improved BP networks of three layers to sort face database after lowered and links ICA. It has higher recognition rate and stupid and excellent of facerecognition system. ICA adopts more effective algorithm of FastICA, improved BP adopts momentum mode and adaptive adjustment of studying rate, it reduces sensitive to error of curved surface particulars, effectively controls the smallest rate that network is limited to part and improves disappearing speed of calculation and reliability. By experiments, it's confirmed an effective method, especially it improves recognition rate to face database changed by expression and posture.