SOM-based binary coding for single sample face recognition

Due to the semantic gap between the insufficient facial features and facial identifying information, the single sample per person (SSPP) problem has always been a significant challenge in the field of facial recognition. To address this problem, this paper proposes a Self-Organizing Map (SOM)-based binary coding (SOM-BC) method, which extracts the middle-level semantic features by merging the SOM network with the Bag-of-Features (BoF) model. First, we extract the local features of the facial images using the SIFT descriptor. Next, inspired by human visual perception, we utilize a SOM neural network to obtain a visual words dictionary capable of reflecting the intrinsic structure of facial features in semantic space. Subsequently, a binary coding method is further proposed to map the local features into semantic space. Finally, we propose a simple but effective similarity measure method for classification. Experimental results on three public databases not only demonstrate the effectiveness of the proposed method, but also its high computational efficiency.