A Programmable CNN Architecture and Its Hardware-Software Co-design Approach for Image Processing and Stimulating Visual Illusions

Cellular neural network (CNN) shows a strong resemblance to biological visual system, and has found numerous applications in image and video signal processing, and in artificial intelligence. In this paper, a programmable architecture based on simplicial CNN is proposed for image processing operations. The proposed architecture provides a flexible and reconfigurable hardware platform that can be used to implement gray-level and binary image processing functions. Hardware-software co-design approach is applied to further improve its efficiency and completeness. Experimental results show that several common image processing tasks are fully completed by using a combination of the basic instructions. Additionally, studying on visual illusion is regarded as that it can provide fundamental insights for biological perception and cognition. It is also shown in the experimental results that the proposed model is responsible for stimulating visual illusions.