Study On Ear Detection In Dynamic Image Sequences

As a relatively new kind of biometrics, ear recognition aimes at determining or validating people's identity based on ear images. Ear recognition can be a beneficial supplement for other biologic recognition or be solely used at some occasion. As the first and also key step of a whole ear recognition system, robust real-time ear detection and tracking is of great research values. Nowadays, the research of ear detection and tracking is still at the exploration stage. It is a great challenge since there is no fully-developed theory.On the basis of mainstream moving detection and tracking technologies, a new method of detecting and tracking an ear in dynamic image sequences is presented. This new method includes three parts. Moving information of an image is used in the first part. Firstly, a simple background difference algorithm is applied to detect moving human from moving image, then a binary image is obtained with adaptive threshold segmentation algorithm. Finally, a binary edge image of moving human can be obtained by implement opening operation and morphological gradient operation. Skin-color information of an image is applied in the second part. Firstly, hue skin-color histogram in HSV color space is computed on line. Then skin-color probability distribution is calculated according to the skin-color histogram. The region of interest (ROI) of the side view of human face which contains human ear is located with CamShift algorithm. Finally, the ear is roughly located by applying ROI to a binary edge image. Gray-scale contour information is employed in the third part. According to the characteristic of an ear containing rich contour information, the contour of an ear edge image is extracted by ellipse curve fitting with the least square method based on certain rules. Finally, the size and position information of the ear in video sequence images is obtained. Therefore, an ear can be located and tracked successfully.According to the algorithm of ear detection and tracking which has been elaborated above, an ear detection system in dynamic image sequences has been designed and implemented. Experimental results show that an ear appearing in the video image sequences can be tracked accurately. The algorithm is fast, effective and robust, which can meet the real-time requirement and has certain tolerance of angle change of the ear towards to the camera and interferences. Furthermore, the probability of practical application increased since the algorithm is not very sensitive to the interference of accouterments such as the glasses.