外侧膝状体细胞对边缘的响应模型

摘要

Edges composing of stimuli of different intensities are common in both natural scenes and digital images.Edge detection is a basic step for machine visual systems,particularly machine biological visual systems.This paper establishes a model of lateral geniculate nucleus (LGN)cells in a primary visual pathway,describes the LGN cell response to edges,and provides feature representations for designing image processing approaches based on neural mechanism. According to physiological characteristics of retinal ganglion cells,the paper uses the classical model of difference of Gaussians to describe the LGN cell response to stimuli,and obtains a re-sponse function via reasonable simplifications.Through simple analyses,several mathematical properties of the response function are obtained,which agree with the physiological characteristics of neurons.By further simplifying the contrast of a stimulus,a normalized response function is obtained.Numeric experiments show that similarities exist between the function’s response curve and the physiological curve discovered in a previous neural science research,showing validity of the described model.%边缘检测是机器视觉系统与生物视觉系统处理视觉信息的基础阶段。为初级视觉通路中的外侧膝状体(LGN)细胞建立一个模型，描述其对边缘的响应，为构建基于神经机制的图像处理方法提供特征表征。根据神经节细胞感受野的生理特性，用经典的高斯差模型描述 LGN 细胞对刺激的响应，通过合理地简化得到相对简单的响应函数。通过简单数学分析，能够得到函数的几点数学性质，且这些性质都与神经元的生理特性相符。进一步简化刺激的对比度，得到归一化的响应函数。数值实验发现，函数的响应曲线和神经科学研究得到的生理曲线具有相似性，说明该数学模型的合理性。