Large scale palmprint identification: Algorithms and evaluation

In palmprint identification, the identity of a query palmprint image is determined by searching its nearest neighbor from all the templates in the database. For large scale identification system, it is often necessary to speed up the nearest neighbor searching process. In this chapter, we introduce two novel large scale palmprint identification methods, multiscale competitive code and improved cover tree. Mutliscale competitive code adopts a hierarchical matching scheme for efficient palmprint representation and matching. Considering that palmprints are typically multiscale features, in filterbank design, we adopt the log-Gabor wavelets since of its less overlapping in the frequency domain for efficient multiscale palmprint feature extraction. In palmprint representation, competitive code is used to encoding the dominant orientation of the filter responses in each scale. To enforce fast and accurate palmprint identification, in palmprint matching, we propose a novel hierarchical matching scheme where a fusion rule is proposed to combine the distances obtained using different scales. Experimental results indicate that the proposed method would achieve higher accuracy and faster identification speed while compared with original competitive code and several other state-of-the-art methods. To facilitate the fast and accurate searching for the competitive code -based palmprint identification scheme, we also present an improved cover tree method. To utilize cover tree method, we first show that the angular distance used in competitive code can be treated as a metric, and then proposed an improved cover tree method by modifying the searching algorithm. Using the PolyU palmprint database (version 2) and a large scale palmprint database, our experimental results show that the proposed method could effectively speedup the nearest neighbor searching over brute force search.