基于信息流多级结构响应的轮廓检测模型
Contour Detection Model Based on Multi-Level Structural Response of Information Flow
李健 1范影乐1
作者信息
- 1. 杭州电子科技大学模式识别与图像处理实验室,浙江 杭州 310018
- 折叠
摘要
考虑到视觉信息流在视通路多级结构中的处理方式,提出一种图像轮廓检测的新模型.首先,根据初级视皮层(V1区)4B层的简单细胞具有三重感受野结构并对朝向敏感的特性,感知图像方位信息,并经复杂细胞提取获得边缘轮廓响应;其次,根据V1 区 2/3 层细胞的抑制特性,引入稀疏性度量指标和神经元突触动态编码机制对边缘轮廓响应进行抑制,得到纹理抑制响应;最后,利用高级视皮层的融合修正机制,对边缘轮廓响应和纹理抑制响应进行优势互补,得到最终的轮廓检测结果.在RuG40 和BSDS500 图像数据集上进行实验,结果表明所提算法能够有效地区分图像的轮廓与纹理信息,凸显主体轮廓.所构建的基于信息流多级结构响应的轮廓检测模型对后续基于生物视觉机制的图像分析具有一定的参考价值.
Abstract
Considering the processing way of visual information flow in multi-level structure of visual pathway,a new image contour de-tection model is proposed.Firstly,according to the characteristics of simple cells in layer 4B of primary visual cortex(V1),which have triple receptive field structure and are sensitive to orientation,the image orientation information is perceived,and the edge contour re-sponse is obtained by complex cells extraction.Secondly,according to the inhibition characteristics of cells in layer 2/3 of V1,the sparseness measure and dynamic coding mechanism of neuron synapses are introduced to inhibit the edge contour response,and the tex-ture inhibition response is obtained.Finally,the fusion correction mechanism of higher visual cortex is used to complement the edge con-tour response and the texture inhibition response to obtain the final contour detection result.Experiments on RuG40 and BSDS500 image datasets show that the proposed algorithm can effectively distinguish the contour and texture information of images,and highlight the main contour.The constructed contour detection model based on multi-level structural response of information flow is valuable for the subsequent image analysis based on biological visual mechanism.
关键词
轮廓检测/视觉机制/信息流分层响应/三重感受野/突触动态编码Key words
contour detection/visual mechanism/information flow layered response/triple receptive field/synapse dynamic coding引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据