The authors’ have previously reported single-view-based ear recognition rotated in-depth. Feature vectors of various poses have been estimated in advance and subjected to the training of discriminant analysis. Accuracies are compared between this method estimating various-poses from registered image subjected to discriminant analysis and a method based on pose-normalization of input image. Experimental evaluations indicate that our previously reported method outperforms the method based on pose-normalization of input image.
Performance comparison of ear biometrics between the method based on orientation-normalization of the input image and the method estimating various-poses from registered image.