Abstract: Contour detection is one of the key steps in the field of target recognition. Human visual system has excellent ability to extract target contour from cluttered scenes by using inhibition properties that can suppress the responses of central neurons in primary visual cortex（V1 area）. According to this physiological properties, traditional models employ ring-formed area to modulate distance weighting. Based on previous studies, a model in consideration of eye movement mechanism is proposed. Different from other models, eight sub-templates are established with uniformly-spaced orientations to modulate the eye movement mechanism, fixed numbers setting in peripheral region. The final inhibition response is computed in competition with different responses of sub-template. Compared with previous studies, the proposed model has a better performance evaluation. Note that this model can retain more object contour and inhibit more textures.

Key words: contour detection, non-classical receptive field, eye movement, primary visual cortex

摘要： 轮廓检测是目标识别中关键的步骤之一。人类视觉系统具有快速和有效地从复杂场景中提取轮廓特征的能力，初级视觉皮层（V1区）的非经典感受野对中心神经元刺激具有抑制特性。传统模型利用该感受野特性，采用圆环形的非经典感受野模板模拟纹理抑制的距离权重，在传统模型的基础上，提出一种引入人眼微动机制的轮廓检测新模型，该模型将圆环形模板按等间隔角度生成八个子模板，由子模板中的相应角度方位区域置换数值模拟眼动机制，通过竞争获得最终抑制权重。实验结果表明，该模型较传统模型具有较高的性能评测指标，在最大程度抑制背景纹理的同时，保留了更多的真实轮廓。

关键词: 轮廓检测, 非经典感受野, 人眼微动, 初级视觉皮层