Robust Kernel Representation with Statistical Local Features for Face Recognition

Description

In this project , we proposed a statistical local feature based robust kernel representation (SLF-RKR) model for face recognition. A robust representation model to image outliers (e.g., occlusion and real disguise) was built in the kernel space, and a multi-partition max pooling technology was proposed to enhance the invariance of local pattern feature to image misalignment and pose variation. We evaluated the proposed method on different conditions, including variations of illumination, expression, misalignment and pose, as well as block occlusion and disguise occlusion. One big advantage of SLF-RKR is its high face recognition rates and robustness to various occlusions. The extensive experimental results demonstrated that SLF-RKR is superior to state-of-the-arts and has great potential to be applied in practical face recognition systems.

Relaxed Collaborative Representation for Patter Classification

Nonlocally centralized sparse representation for image interpolation