It is known that criminal investigation assistance systems that use face images are unreliable when profile images are queried. Although frontal facial recognition has matured, profile facial recognition is not yet well developed. To compensate for such unreliability, the use of ear recognition is a promising direction, as ear shapes are known to be unique and visible in profile images. However, the ear has a complex, three-dimensional concave shape. Thus, accurate ear identification is challenging when the angle of an ear image from a surveillance camera is different from that of images in an image database. Some studies (including our past work) have addressed this issue and improved the robustness of ear recognition against off-angle ear rotation. However, there is room for improvement in this area. To improve on our earlier work on single view-based ear biometrics, another estimation method for the camera angle of an ear image based on principal component analysis is examined in this study. Experimental results show that the proposed method can improve the accuracy of the estimation of the camera angle. In conjunction with this enhancement, the improvement in robustness against off-angle ear rotation is examined. Therefore, the proposed method improves the robustness of ear recognition by accurately calculating the camera angle of the image of the ear.
展开▼
