Center-Surround Interaction with Adaptive Inhibition: A Computational Model for Contour Detection
Chi Zeng,Yongjie Li,Chaoyi Li
DOI: https://doi.org/10.1016/j.neuroimage.2010.11.067
IF: 5.7
2010-01-01
NeuroImage
Abstract:The broad region outside the classical receptive field (CRF) of a neuron in the primary visual cortex (V1), namely non-CRF (nCRF), exerts robust modulatory effects on the responses to visual stimuli presented within the CRF. This modulating effect is mostly suppressive, which plays important roles in visual information processing. One possible role is to extract object contours from disorderly background textures. In this study, a two-scale based contour extraction model, inspired by the inhibitory interactions between CRF and nCRF of V1 neurons, is presented. The kernel idea is that the side and end subregions of nCRF work in different manners, i.e., while the strength of side inhibition is consistently calculated just based on the local features in the side regions at a fine spatial scale, the strength of end inhibition adaptively varies in accordance with the local features in both end and side regions at both fine and coarse scales. Computationally, the end regions exert weaker inhibition on CRF at the locations where a meaningful contour more likely exists in the local texture and stronger inhibition at the locations where the texture elements are mainly stochastic. Our results demonstrate that by introducing such an adaptive mechanism into the model, the non-meaningful texture elements are removed dramatically, and at the same time, the object contours are extracted effectively. Besides the superior performance in contour detection over other inhibition-based models, our model provides a better understanding of the roles of nCRF and has potential applications in computer vision and pattern recognition.