Abstract:
The human mind processes the information in a complex fashion including utilization of color, shape, texture and symmetry-related meta-information. but in conjunction with a strong (domain) knowledge, these can boost the overall performance. Inspired by this observation, we present a novel approach in building learning-based COST-S space. This space consists of meta-level features obtained from dictionary learning and combining it with task speci c class ers such as DenseNet for object recognition. Con dence based fusion mechanism is presented to supplement a task speci c class er using the proposed COST-S representation. The performance of the proposed framework is evaluated on four benchmark face recognition datasets: (i) Disguised Faces in the Wild (DFW), (ii) Labeled faces in the wild (LFW), (iii) IIITD Plastic Surgery dataset, and (iv) Point and Shoot Challenge (PaSC). Experimental results show the robustness of the proposed framework, in terms of improvement in face recognition accuracy.
