Abstract

We develop an improved LPP method called scatter-difference discriminant locality preserving projections (SDLPP) for palmprint recognition. SDLPP has better classification capability which benefits from discriminant information obtained by maximizing the difference of between-class separability and within-class similarity. SDLPP avoids the singularity problem for the high-dimensional data matrix and can be directly applied to the small sample size problem while preserving more important discriminant information. Comparative recognition performance results obtained on public PolyU palmprint database also confirm the effectiveness of the proposed SDLPP approach.

Key words: Palmprint recognition, scatter-difference criteria, locality preserving projections, singularity problem.