Design of a multi-stage hybrid model for face recognition in varied illumination conditions

Varied Illumination and unequal contrast are the critical challenges in the face recognition system. This paper presents a multi-stage hybrid model to handle extreme illumination, unequal contract, and varied pose challenges. These challenges are handled within pre-processing and feature generation stages. In the pre-processing stage, the content and structural feature-based feature exposing and region selection method is defined. In the second stage, the hybrid feature generation- adaptive weber and speeded up robust features(AWSURF) model is applied on rectified and normalized face images. In this stage, a weber filter is applied for generating the exposing the structural contents. The Speeded up Robust Features (SURF) method is applied to the weber face for extracting the illumination and pose robust features. The proposed model has experimented on Yale, Extended-Yale and CMU-PIE datasets. The proposed model is experimented and validated against EigneFace, Local Binary Pattern(LBP), Gabor,EigenFace+Gabor, Gabor+LBP,Weber, SURF,Weber+SURF feature processors. The experimental analysis is done against accuracy and error measures. The proposed model achieved the 92.88%, 91.77% and 99.09% accuracy against Yale, Extended-Yale and CMU-PIE datasets and outperformed the experimented handcraft approaches. The analysis results confirm that the proposed model achieved a higher average accuracy rate against state-of-art deep learning and handcraft approaches.

Automated summary

This paper presents a multi-stage hybrid model to handle challenges of extreme illumination, unequal contrast, and varied poses in face recognition, achieving high accuracy rates compared to traditional methods.