Title:Illumination Normalization via Merging Locally Enhanced Textures for Robust Face Recognition

Abstract:In order to improve the accuracy of face recognition under varying illumination conditions, a local texture enhanced illumination normalization method based on fusion of differential filtering images (FDFI-LTEIN) is proposed to weaken the influence caused by illumination changes. Firstly, the dynamic range of the face image in dark or shadowed regions is expanded by logarithmic transformation. Then, the global contrast enhanced face image is convolved with difference of Gaussian filters and difference of bilateral filters, and the filtered images are weighted and merged using a coefficient selection rule based on the standard deviation (SD) of image, which can enhance image texture information while filtering out most noise. Finally, the local contrast equalization (LCE) is performed on the fused face image to reduce the influence caused by over or under saturated pixel values in highlight or dark regions. Experimental results on the Extended Yale B face database and CMU PIE face database demonstrate that the proposed method is more robust to illumination changes and achieve higher recognition accuracy when compared with other illumination normalization methods and a deep CNNs based illumination invariant face recognition method