Modelling ecologically specialized biological visual systems

摘要

Computer vision is a broad area in artificial intelligence (AI) usually utilizing digital images. Digital image compression techniques are required to overcome significant storage and transmission problems in this area. A strong understanding of image characteristics enhances the effectiveness of these and many other image processing operations. Traditional methods typically maintain constant resolution throughout an image. However, a survey of the various visual systems present in the animal kingdom demonstrates the potential of variable resolution (VR) compression methods. The authors model several animate visual systems and outline novel image compression techniques based on foveated vision. Interesting variations on the simple fovea are presented, motivated by similar variations present in animate visual systems; specifically multiple, moving and weighted foveae as well as visual streaks. Techniques for the efficient modelling of fovea movement are also described.