基于视通路多感受野朝向性关联的轮廓检测方法

基于视通路中各层次的感受野特性以及层次间感受野的关联特性,提出一种图像轮廓检测新方法。根据神经元的放射状树突接收特性,构建检测空间差异性信息的视网膜神经元网络,对轮廓信息进行前级编码;建立LGN细胞的非经典感受野调节机制,结合感受野的层次变换特性,对前级编码结果进行全局调节;提出简单细胞的多感受野朝向性关联模型,模拟初级视皮层简单细胞的方向选择特性;融合多个方向上的轮廓响应,经过非极大值抑制和阈值处理,得到轮廓检测结果。以RuG图库40幅图片为实验对象,检测结果与基准轮廓图的平均P指标为0.43,结果显示该方法能够有效凸显主体轮廓,强化轮廓与纹理边缘的区分度。利用多感受野的层次关联特性,实现图像轮廓信息的编码与检测应用,为研究高级视皮层的图像理解和视觉认知提供新的思路。

A Contour Detection Method Based on Correlation of Orientation for Multiple Receptive Fields in Visual Pathway

Based on characteristics of receptive fields on different layers of human visual path and their correlation between every adjacent layer, a new method of image contour detection was proposed in this paper. Using the receptive feature of neurons with radial dendrite, a retina neuron network that has the capability to detect spatial difference was constructed for pre-coding of contour information. Taking the variety between adjacent layers into account, the adjustment mechanism for non-classical receptive field of LGN neurons was built to make a global adjustment on the pre-coding result. Next,a model of simple cell on primary visual cortex named orientation correlation of multiple receptive fields was proposed to realize the orientation selection of simple cells. At last,the fused contour response of all orientations was processed with non-maximum suppression and hysteresis thresholding to acquire the contour detecting result. In this paper, 40 images in the RuG library were taken for processing, and the mean value of measure P between detecting result and ground truth was 0.43, indicating that method in this paper could highlight principal contour and enhance the discrimination between contour and textured region. In this study, the correlate characteristics of multiple receptive fields were utilized to code and detect image contour, providing a new idea for the image processing and visual cognition of high-level visual cortex.