The Application Of Ear Recognition Based On Multi-directional And Multi-scale Analysis And Moment Features

In recent years, biometrics has received a lot of attention, the ear image recognition is one of the new biometrics hotspots. Ear recognition can be a beneficial supply for biometrics recognition because of ear's unique physiological characteristics and observation angle. Hitherto, it is at the primary stage of research in the world and there is few related study institutions on ear recognition in China. So the ear recognition has enormous potentials and immense space of application and development.In this paper, a human ear Automatic Identification System is established. It consists of a human ear image acquisition module, an image preprocessing module, a feature extraction module, a feature data processing module and a neural network classification module. In the human ear image acquisition module, based on our laboratory-designed human ear acquisition system, an ear collecting system is designed to construct a standard Chinese Ear Image Database (CEID), which includes 200 Chinese ear samples. During the acquisition process, it needs to acquire everyone's two ear images, and each has 4 kinds of circumstances (including 3 kinds of typical light changes and one kinds of occlusion) and 4 kinds of shooting angles. Totally there are 6400 ear images in CEID. In the image preprocessing module, the principle and characteristics of multi-scale geometric transformation have been studied, and with the combination of the characteristics of the ear image, curvelet transformation has been used to enhance and de-noise human ear images. Then extraction of human ear image's edge with the wavelet modulus maxima solves the problems of uneven light, illumination variation. In the feature extraction module, in order to rapidly and non-destructively calculate image eigenvalue, combining Mallat algorithm, a new invariant wavelet moment algorithm with non-destructive sampling is presented. On this basis, combining with the principle and characteristics of Fourier's Transformation, feature extraction based on amplitude spectrum and wavelet moment invariants is also presented, and a comparison is made among the improvement of wavelet moment invariants algorithm, the traditional cubic B-spline wavelet moments and Hu moment invariants. In the feature data processing module and neural network classification module, the reason of the error of image feature has been analyzed. The weights of eigenvectors are calculated with error processing method to reduce the effect of the error for the classification and 3 layers BP neural network is employed for classification.The experimental results are as follows.①Compared with other methods, the PSNR (Peak Signal to Noise Ratio) and visual effect of the curvelet de-noised image are greatly improved, especially on the restoration of the edge of image, and the effect of curvelet enhancement has better results than other methods as well.②The new invariant wavelet moment algorithm greatly accelerates the wavelet moment invariants calculation under the condition that there is few losses in performance. At the same time, the new wavelet moment invariants algorithm based on amplitude spectrum is verified to be insensitive to noises compared with other wavelet moment invariants.③There are 200 samples of the human ear images in this experiment, each sample has illumination changes, translation, scaling, noise, rotation, etc. totally 72 kinds of status. With the aforementioned automatic identification system to identify, various dimensions of the human ear recognition rate exceeds 97%. The recognition rate exceeds 95% when the human ear image is added with 10% random noise in the simulation.In sum, this human ear automatic identification system can not only solve the problems of uneven light, illumination variation and noise interference, but also the problems of translation , rotating and scaling during the process of human ear image acquisition. So it can chive high recognition rate. Therefore, this paper is expected to provide valuable ideas for the development and application of human ear recognition technology.