Face Recognition Based on Non-Subsampled Contourlet Transform and Multi-order Fusion Binary Patterns

摘要

In this paper, we propose a novel face representation approach based on Non-subsampled Contourlet Transform (NSCT) and Multi-order Fusion Binary Patterns (MFBP). NSCT, which is a newly developed multi-resolution analysis tool in image denoising and enhancement, can be used to effectively capture image features of both geometrical structure and directional texture information. Due to the ability of extracting multi-order derivatives of texture patterns, the MFBP is applied on the NSCT coefficient images to achieve enhanced face representation. Furthermore, Block-based Fisher Linear Discriminant (BFLD) feature selection and weight scheme based on Fisher Separation Criteria (FSC) are chosen to further improve discriminative power of the proposed face representation. The experiments on public FERET database demonstrate that our approach outperforms many of the state-of-the-art methods.