基于双视通路交互感知的轮廓检测方法

基于生物视觉系统存在双视通路(VP)交互感知的机制，该文提出一种图像轮廓检测的新方法。首先针对皮层下视通路中视觉刺激流经多级不同尺度的感受野，提出一种多尺度轮廓融合的轮廓感知模型；接着基于皮层上视通路的对比度适应机制和方向敏感特性，获取显著性视觉特征；然后模拟双视通路的交互感知机制，分别在V1皮层中，构建一种信息流交互引导的脉冲编码模型，提取显著性轮廓；在上丘(SC)浅层提出一种特征调制的非经典感受野侧抑制模型，实现纹理抑制；最后对双视通路中的轮廓响应结果进行修正融合，得到最终轮廓响应。针对RUG40图像库的测试，整个数据集的最优平均P指标和每张图的最优平均P指标分别为0.51和 0.57；针对BSDS500图像库的测试，数据集尺度上最优(ODS)为0.68。结果表明该文方法能有效突显主体轮廓并且抑制纹理背景。通过该文提出的轮廓感知方法，为后续基于视觉机制的图像理解和分析提供了一种新的思路。

A Contour Detection Method Based on Interactive Perception Mechanism of Dual Visual Pathways

According to the mechanism of visual information interactive perception between dual Visual Pathways(VP) in the biological vision system, a new method of contour detection is proposed. Considering the visual stimulus in the hypocritical pathway flowing through multi-level and different-scale receptive fields, a multi-scale contour fusion contour perception model is proposed. Based on the contrast adaptation mechanism and directional sensitivity of the visual pathway on the cortex, salient visual features are extracted. The interactive perception mechanism of the dual vision pathway is simulated, a pulse coding model is constructed guided by the information flow interaction in the V1 cortex, to extract the saliency contour. An inhibition model of feature modulation non-classical receptive field is proposed in the Superior Colliculus(SC) shallow layer, to achieve texture inhibition. Finally, the contour response results in the dual-view path is modified and fused to obtain the final contour response. For the test of the RUG40 image library, the optimal average P index of the whole dataset and each graph is 0.51 and 0.57 respectively. For the test of the BSDS500 image library, the Optimal Scale (ODS) of the Dataset is 0.68. The results show that the method in this paper can effectively highlight the outline of the subject and suppress the textured background, which provides a new idea for the subsequent image understanding and analysis based on the visual mechanism.