De-Noising Model for Weberface-Based and Max-Filter-Based Illumination Invariant Face Recognition

摘要

In the topic of illumination invariant face recognition (IIFR), although the state-of-the-art Multi-scale Weberface (MSW) and Multi-scale Quotient Image (MQI) give best results against other illumination insensitive feature extraction methods, they are computationally heavy and easy affected by noises hiding in face shadow. In this paper, we propose a lightweight de-noising model to boost the IIFR system based on max-filter and Weberface called GMAX and GWEB respectively. In this model, we try to eliminate the influence of quantum noise and quantization noise on ill-illuminated images by average smoothing and Gaussian smoothing. After that, linear discriminant analysis (LDA) is adopted to improve verification rate. Never before, a comparative study on popular approaches in the literature fully implemented on the challenging data set Extended Yale B is also provided. The proposed method gives excellent results in term of both computational time and accuracy.