Study On Some Key Issues Of Ear Recognition

Biometrics is a kind of science and technology using individual physiological or behavioral characteristics to werify identity.It provides a highly reliable and robust approach to the identity recognition.Automatic ear detection and recognition is not only a new branch of biometrics but also one of the most challenging tasks for image processing, pattern recognition and computer vision at present.Due to the unique biological location and structural feature of human ears,the research has prospective potential application in public safety,information security field and so on,which has drawn increasing attention. Presently the research on ear biometrics is still at the experimental stage.There are many crucial problems to be solved for the emergence of a practical ear biometrics system.In this dissertation the ear recognition algorithms are researched,and the main works are as follows:1) The ear biometrics technology and its development,application were summarized. Not only the feasibility and content of the research were descripted,but also the advantages and disadvantages of the research were discussed.Then the dissertation conducted a detailed survey on previous methods used in the area.Furthermore the crucial problems of ear recognition were indicated.2) Ear image pre-processing technologys were researched.A method for auricle segmentation based on template matching and genetic algorithm in side face was proposed, which can segment ear from 2D grey side face,and obtain a minimum rectangular erea, only including one ear.3) The edge detection methods of ear images were researched.Firstly an ear contour extraction method according to the human ear position and the contour characteristic based on edge tracking was proposed.According to different tend of ear contour,the whole contour curve can be extracted accurately througth tracking different region.Then with regard to human ear image,an edge detection method based on main curvature of surface was proposed,in order to detect weak edges,ear structure feature extraction based on multi-scale Hessian matrix was presented,it can extract stable ear edge curve.The tesl results show these methods can all detect ear image edges effectively.4) The contour curve registration methods were researched.According to the characteristic of outer ear contour,an ear contour curve registration method based on improved Hausdorff distance was proposed.The method was robust to position change and plane rotation.For more recognition rate,a local feature point extraction method was used for final refined match.5) The application of invariant feature in ear image recognition was researched.Stable feature key points and robust feature descriptors were extracted from ear image using SIFT For Gabor's eminent characteristics in spatial local feature exaction and orientation selection,a new low-dimension point feature descriptor was presented in this dissertation, matching efficiency can be improved by using combing Gabor with SIFT.Because of the matching defect of local descriptors,a ear recognition method based on fusion SIFT and geometric feature was proposed,it can help disambiguate mistake matching when an image has multiple similar regions.This method is effective for illumination change and slight depth rotation ear images recognition.6) Force field transformation theory and its application in ear recognition were researched,an ear image feature extraction method based on force field transformation was proposed.According to human ear's characteristic,force field transformation theory was applied to human ear image and energy transformated image respectively,it extracted the structural feature points and contour feature points.The feature points extracted from ear image were stable,reliable,unique and discriminative,and robust to rigid transformation of ear image.They can effectively remove the effect of illumination changes for recognition and resolve the problems of lower recognition owing to ear posture shift and minute alteration caused by head depth rotation.