摘要
目的提出一种基于分频视觉机制的图像显著轮廓特征提取新方法,在多通道分离基础上融合视觉信息机制,以实现图像的有效性和完整性。方法利用具有频域响应特性的高斯导函数,模拟LGN感受野对视觉信息的分频特性,实现空间频域调谐作用;根据空间频率和朝向调谐之间的全局性抑制作用,构建一种基于对比度自适应的朝向敏感感受野,通过视觉感受野信息差异的检测,实现对外周纹理的选择性抑制,并利用分频视觉信息流的融合模型,表征初级视皮层中的上下文整合机制。结果以RuG40图库中的图像作为实验样本,经过非极大值抑制和阈值处理,检测结果与轮廓基准图的平均指标为0.48,优于其他对比方法。结论该方法能够实现图像轮廓的有效检测,为后续图像编码与融合提供新的思路。
Objective A new method of salient contour feature extraction based on frequency division visional was proposed,and the visual information mechanism was integrated on the basis of multi-channel separation to realize the effectiveness and integrity of images.Methods The Gaussian derivative function with frequency domain response characteristics was used to simulate the frequency division characteristics of the LGN receptive field to visual information,and the spatial frequency domain tuning was achieved.According to the global suppression between spatial frequency and orientation tuning,a contrast-based self-adapting sensitive receptive field was constructed.The selective suppression of peripheral texture was achieved through the detection of visual receptive field information differences.The fusion model of frequency-divided visual information flow was used to characterize the context integration mechanism in the primary visual cortex.Results With the images in the RuG40library as the experimental samples,the average index of the detection results and the contour standard map was 0.48after non-maximum suppression and threshold processing which was better than other comparison methods.Conclusion The method can effectively detect the contour of the image which provides a new idea for subsequent image coding and fusion.
作者
方琳灵
范影乐
房涛
武薇
Fang Linling;Fan Yingle;Fang Tao;Wu Wei(不详;Laboratory of Pattern Recognition and Image Processing,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China)
出处
《航天医学与医学工程》
CAS
CSCD
北大核心
2020年第6期522-532,共11页
Space Medicine & Medical Engineering
基金
国家自然科学基金资助项目(61501154)。
关键词
显著轮廓
分频视觉机制
选择性抑制
信息流融合
salient contour
frequency division visual mechanism
selective inhibition
information flow fusion
作者简介
第一作者:方琳灵,硕士研究生,研究方向为模式识别与图像处理。E-mail:182060252@hdu.edu.cn;通讯作者:范影乐,男,博士,教授,研究方向为神经信息学、机器视觉与机器认知。E-mail:fan@hdu.edu.cn。
