Human ear recognition using voting of statistical and geometrical techniques

Ear recognition gained great importance in the field of pattern recognition due to its simple formulation, and rich characteristics compared to other used biometrics such as the face, iris, and voice and so on. In this paper, we presented a new algorithm for ear recognition based on a voting method between results obtained by three efficient techniques of 2nd level Haar wavelet transform, Histogram of oriented gradient descriptors and geometrical based technique. Firstly, enhancement procedure is applied on the images. Then we developed an ear segmentation technique that selects the resulted ear image from two methods one depends on active ear contour and the other depending on ear edges endpoints connection. After that for every image one feature vector is extracted using the three methods and a matching is tested between testing and registered images by using Euclidian distance. Such that any distance beyond a certain threshold value for each method is a candidate to be for the matched image. Voting between the candidates selects the correct match. The experimental results achieved overall accuracy of 99.6% when applied in images affected by illuminating changes and pose variations.