Ear recognition based on discriminant multi-resolution image representation

Abstract

The multi-resolution analysis is more appropriate for extracting information from measured data, because it is generally multi-scale in nature. This paper proposes a new approach for ear representation, based on multi-resolution analysis framework. Such representation relies on significantly Gabor wavelet, local phase quantisation (LPQ) descriptor and spatial pyramid histogram (SPH) method. First, to capture the local structure in ear image, the Gabor wavelet function with two scales and four orientations is used. Second, to fully explore the blur invariant property and the texture information in different scales and directions spaces, the LPQ operator is applied on the image responses of Gabor filter to get label LPQ images. Third, the SPH of horizontal decomposition is applied for each of them, to obtain local ear feature descriptors. Next, the obtained histograms are normalised. Then, the global representation of ear image is obtained by concatenating all the local feature descriptors. After that, a discriminant representation of ear image is constructed using whitened linear discriminant analysis. Finally, the K-nearest neighbour classifier is used for identification. Experiments conducted on two ear databases (IIT Delhi-1 and IIT Delhi-2); show that the proposed approach provides a significant accuracy improvement compared to the state-of-the-art methods.