A visual search system based on multi-scale population cell coding

Gaze movement plays an important role in human visual search system. How to simulate such a system to efficiently encode and decode gaze movement for target searching is a meaningful issue. There are two key points that should be addressed for this issue. First, eye movement is affected by the visual context that includes more than one object in images. It is important to study how to encode the spatial relationship between the target and the other objects, and decode the spatial relationship to drive the gaze moving to the target. Second, the human retina has a non-uniform distribution of light sensing neurons, which can be viewed as a composition of neuron arrays in different spatial resolutions. When a system searches a target, how many scales of visual fields should be involved to sense the image stimulus? In this paper, we propose a visual search system using the population cell coding mechanism and the multi-scale visual field as sensing input. As an example, the system is applied to human eye center searching. An experimental comparison of a Full-scale visual field coding system and a Gradual-scale coding system is carried out. The experiment results show the Gradual-scale coding system performs better than Full-scale coding system for target searching.