Ear Recognition Technology Based On The Contour Of The Auricle

Ear recognition is a new technology of biologic recognition, which has drawn more and more attention from the scientists because of its unique feature and applied direction. Ear recognition involves many domains such as feature extract, computer vision, image processing, pattern recognition and identity authentication etc. Ear recognition can be not only a beneficial supplement for other biometric recognition, but also solely used at some situation.There are many algorithms about ear recognition now, but ear recognition employing the spatial information of edge map associated with optimal information remains an unexplored direction. Based on present research status, the basic theory and method of edge information used in ear recognition is studied in this paper.Ear image denoising plays an important role in ear recognition system. In this thesis, an ear image denoising and inpainting approach based on skin color detection is proposed. This method employs skin color detection to segment images into skin color and non-skin color regions in HSI color space. Those non-skin color regions surround by skin color pixels will be considered as noise regions, then these noise regions are inpainted by a new image-inpainting algorithm. When the image-inpainting algorithm is carried out, non-skin color pixels in noise regions will be replace by skin color pixels around the non-skin noise region. The experiment results with ear images demonstrate that the proposed method is very simple and works well and fast for image denoising while some important information like edges is substantially retained.Most of the classical methods of edge detection are either based on various differential operators combined with the methods of threshold ranking or smoothing or template matching, or some optimized arithmetic operators based on the classic differential operators. The applications of the above methods have some disadvantages such as sensitivity to noise, loss of edge detection. In this paper, Gray-Scale morphology is applied to edge detection. An improved edge detection operator is proposed that uses morphological operations such as dilation, erosion, opening, closing and their combination. The method can accurately detect and catch the position of edge points and has satisfied filtering results. Experiments demonstrate that compared with traditional edge detectors, this edge detector has a good performance of noise reduction and requires fewer calculations, which enhances its practicality and feasibility. The Hausdorff distance between planar sets of points is known as a good method to compare binary images. In this paper, a new modified Hausdorff distance using standard deviation and difference in length of edge lines is proposed, which could gets more accurate measurement of the differences between the edge lines. It reduces the errors that induced by noise, pseudo edge segments and outlier points. On the last part of ear recognition, the author attempts to employs the method of SVM (Support Vector Machine), which has some superiority in small sample biometric recognition. The parameters of this method can be confirmed artificially, and it operates simply very much. Experimental results show that the modified algorithm can achieve recognition rate of 94.4% based on 320 ear images in database. It is obvious that this algorithm can achieve satisfied recognition rate.