Human Verification Based On Ear Geometrical Features

Auricle has stable and rich structures. It can't be affected by countenance and it is difficult to be cheated, easier to be collected, more acceptable. Therefore, ear biometrics is becoming a new hotspot of biometrics. Now ear recognition just gets off the mark, and its feature extraction method and recognition rate is not satisfactory.In this thesis, an automatic ear recognition system is introduced. The research work is done in three stages, contour extraction, contour localization, feature extraction and classification. The main work of this paper concludes:(1)Image acquisition equipment with close light sources has been designed. A database with 63 individuals and 830 ear samples has been established.(2)An image segmentation method has been used in ear contour extraction. An optimal global and adaptive thresholding method is modified in the system. It performs better in efficiency than Snake models, Canny method, and Sobel method.(3)In this thesis, a new ear contour localization method is presented, i.e., using root points of helix and lobule. Then an anchor points detecting method based on contour directions is introduced and carried out in the system.(4)We use a grid extracting method in feature extraction, and present a new grid feature extracting method based on polar coordinates systems of images which performs good robust and has shorter feature vectors than traditional methods.Experimental results based on large datasets show the good robust and effectiveness on the proposed approach for ear recognition.