An Efficient Coding Model for Image Representation

Abstract

The role of early sensory neurons is to remove statistical redundancy in the sensory input. In this paper, we propose a novel efficient coding model combining sparse coding strategy and selective attention strategy for image representation. The model is divided into two modules. In the first module, we employ the sparse coding strategy for natural image feature extraction. Furthermore, inspired by the selective attention strategy in biological visual system, we propose a self-adaptive algorithm to further reduce the activated variables in the second module. Compared with standard sparse coding (SC), the experimental results show that the efficient coding model evidently decreases the number of coefficients which may be activated and preserves the main structural information at the same time. Moreover, our model employs fewer responses to preserve similar perceptual image quality than other models.