Ear Recognition Based On Principle Component Analysis

Biometrics, because of using the proper living creature characteristic of human body, is the totally brand new technique different from traditional personal identification method. More and more people attach importance to biometrics, because it has the better safety, dependable with the usefulness. Ear recognition is one of the most important part in all kings of biometrics. It is a research area spanning several disciplines such as image processing, pattern recognition, computer vision, physiology. Now it is one of the key issues. However, the fact that the recognition results are very easy to be effected by the variation of angle, accouterment and illumination, how to improve this problem is the primary task in this paper.Firstly, the methods such as PCA, 2DPCA,2DFLD and PCA&FLD based on the multi-element statistic analysis was actualized in this paper. PCA is the ear recognition method based on whole feature. The result of experiment proves to be fast in computing speed and stabilization. But it costs huge computation amount and has the problem of computation complexity as a result of it based on image vectors. 2DPCA based on image matrix projection can overcome the problem of computation complexity in PCA and it has the preferable recognition effect. 2DFLD and PCA&FLD can overcome the problem of losing classification information when choosing eigenvectors in PCA method. The experimental results show that the recognition rate is far higher than PCA. Secondly, a method, based on the two main factors influencing the 2DFLD recognition effect, the discreet degree of samples in projection space and the similarity between sub-patterns, for evaluating the capability of sub-pattern is contrived. Experiments are carried out in the ear database, ORL face database and iris database which show that this evaluation method can effectively choose the preferable sub-pattern, and shorten the calculation time to about 1/4 of the conventional choosing method. Finally, to allay the influence on recognition effect of scale change and parallel shift change and reduce the feature extraction time and match time, a method, using Bilinear Interpolation to change the image resolutions, was proposed. The experimental results show that the recognition effect is remarked improved.