Dual-tree Complex Wavelet Transform based Local Binary Pattern Weighted Histogram Method for Palmprint Recognition

In the paper, we improve the Local Binary Pattern Histogram (LBPH) approach and combine it with Dual-Tree Complex Wavelet Transform (DT-CWT) to propose a Dual-Tree Complex Wavelet Transform based Local Binary Pattern Weighted Histogram (DT-CWT based LBPWH) method for palmprint representation and recognition. The approximate shift invariant property of the DT-CWT and its good directional selectively in 2D make it a very appealing choice for palmprint representation. LBPH is a powerful texture description method, which considers both shape and texture information to represent an image. To enhance the representation capability of LBPH, a weight set is computed and assigned to the finial feature histogram. Here we needn't construct a palmprint model by a train sample set, which is not like some methods based on subspace discriminant analysis or statistical learning. In the approach, a palmprint image is first decomposed into multiple subbands by using DT-CWT. After that, each subband in complex wavelet domain is divided into non-overlapping sub-regions. Then LBPHs are extracted from each sub-region in each subband, and lastly, all of LBPHs are weighted and concatenated into a single feature histogram to effectively represent the palmprint image. A Chi square distance is used to measure the similarity of different feature histograms and the finial recognition is performed by the nearest neighborhood classifier. A group of optimal parameters is chosen by 20 verification tests on our palmprint database. In addition, the recognition results on our palmprint database and the database from the Hong Kong Polytechnic University show the proposed method outperforms other methods