基于SIFT特征的人耳识别

Human ear recognition based on SIFT feature

人耳的角度变化和遮挡是人耳识别中的难点问题,SIFT局部描述算子具有对图像尺度缩放、平移、旋转等的不变性,因此提出利用SIFT特征的人耳识别算法。该算法将人耳图像划分为相互重叠的网格区域,在每个子区域中计算SIFT的局部特征,再计算测试图像与训练图像的匹配相关度作为图像的全局特征,将SIFT的局部和全局特征相结合作为人耳识别的标准。通过在人耳库中的实验表明,此算法优于传统的全局方法,对于人耳角度变化和遮挡具有较好的鲁棒性,并且适用于单训练样本的情况。

The variety of ear angle and occlusion are the difficulties of ear recognition. The Scale Invariant Feature Transform （SIFT） is invariant to image scaling, translation and rotation. So the human ear recognition algorithm based on SIFT features was proposed. The SIFT features were computed from the ear image, and then image was divided into several overlapping grid regions, in which the local features of SIFT on each region are also computed. The matching similarity was computed between training image and test image, which was treated as global feature. The local feature and global feature were fused finally. The experiment results on ear database show that the algorithm works better than traditional global method, and is robust for the variety of ear angle and occlusion, and it is suitable for the recognition using the only one training sample.