Based On The Classical Receptive Field Model And The Application Of The Visual Identity Of The Research

Human visual system, as the most major source of perception information forhuman brain, has perfect capability for information processing and outperforms thebest current computer vision system by almost any measure. As an importantcomponent of the brain and cognition science, the research of visual mechanism and itsmodeling may inspire efficient computational methods for many applications, such asimage analysis and understanding. In this paper, we mainly focus on understanding theroles of neurons for information processing with a computational evaluation.Based on the sufficient findings of physiology and psychophysics, this paperproposed several computational models for contour detection in natural scenes. Thecontributions of this paper mainly include two parts:(1) Based on the surround inhibition of V1neurons, this paper evaluated thefunctional role of co-existing V1neurons with orientation-selective and withnon-selective surround inhibition in the specific perceptual task of contour detection.Based on the computational finding that two types of neurons contribute to contourdetection with different performance, we combine two surround inhibition mechanismsand substantially improve the performance of contour detection in natural scenes.(2) Another content of this paper is on the role of various local visual cues inintegrating and improving contour detection. The magnitude of surround inhibition ofV1neurons can be modulated by various local visual features (like luminance,orientation etc.). Therefore, in this paper, different local features are extracted firstly,and a supervised learning is used to combine the information from different cues. Then,the combined information is used to modulate the magnitude of surround inhibition.The results demonstrate that combining multiple cues can remarkably improve theperformance of contour detection, and generally, luminance and contrast contributemuch more to the specific task of contour perception than other visual features.