摘要
目的提出一种主视通路信息流层级传递和响应的新模型用于检测图像轮廓的新方法。方法以Ru G图库40幅图片为实验对象,利用非下采样轮廓波变换模拟外侧膝状体(lateral geniculate nucleu,LGN)对视觉信息的频域分离作用;构建LIF神经元网络模型来表达视觉神经系统中的电生理活动,通过CRF机制整合空间信息;同时,利用局部半波整流的高斯差函数来模拟n CRF的全局调节机制。有朝向性地将多个LGN细胞感受野进行关联。同时构建皮层下视通路来模拟它对于主视通路进行视觉信息处理的协同作用。然后,经过非极大值抑制和阈值处理,得到本文轮廓检测结果。最后将本文检测结果与3种经典方法(Noninh,SSC,ISO)的检测结果进行对比。结果本方法的检测结果与基准轮廓图的平均P指标为0.46,大于经典的3种检测方法(P指标分别为0.36、0.40、0.42)。结论本文算法不仅对纹理和背景具有抑制的作用,而且能有效区分纹理强边缘和主体轮廓,获得较佳的效果。
Objective To propose a new model of hierarchical transfer and response of the information flow in the cortical visual pathway to detect the image contour. Methods Forty images in the Ru G library were selected for processing. The non-subsampled contourlet transform was used to simulate the frequency-domain separation of visual information in lateral geniculate nucleu( LGN). The LIF neural network model was constructed to express the electrophysiological activities in the visual nervous system,and then the spatial information was integrated via CRF mechanism. At the same time,in order to simulate the global adjustment mechanism of the n CRF,the Gaussian difference function of the local half-wave rectification was utilized. After that,multiple LGN cells were associated into the field. Meanwhile,subcortical visual pathways were constructed to simulate the synergistic effect of visual information on the cortical pathways. The non-maximum suppression and threshold processing were used for processing and the contour detection results were obtained. In the end,the results were compared with that of other three classical methods( Noninh,SSC,ISO). Results The mean value of P between the detected result and the ground truth was 0. 46,which was better than the other three classical methods( P values were 0. 36,0. 40,0. 42 respectively). Conclusion The method not only has a deterrent effort on texture and background,but also can effectively distinguish the edge of the texture and the main contour,which can get better results.
引文
[1]Yi S,Labate D,Easley GR,et al.A shearlet approach to edge analysis and detection[J].IEEE Transactions on Image Processing,2009,18(5):929-941.
[2]Grigorescu C,Petkov N,Westenberg MA.Contour detection based on nonclassical receptive field inhibition[J].IEEE Transactions on Image Processing,2003,12(7):729-739.
[3]Ursino M,La Cara GE.A model of contextual interactions and contour detection in primary visual cortex[J].Neural Networks the Official Journal of the International Neural Network Society,2004,17(6):719-735.
[4]Tang Q,Sang N,Zhang T.Contour detection based on contextual influences[J].Image&Vision Computing,2007,25(8):1282-1290.
[5]Tang Q,Sang N,Liu H,et al.Detecting natural image contours by combining visual perception and machine learning[J].Scientia Sinica,2013,43(9):1124.
[6]李孟寒.仿生脑外侧膝状体模型及若干工作机理研究[D].长春:吉林大学,2011.Li MH.Study on the model and several working mechanism of LGN in the bionic brain[D].Changchun:Jilin University,2011.
[7]Da CA,Zhou J,Do MN.The nonsubsampled contourlet transform:theory,design,and applications[J].IEEE Transactions on Image Processing,2006,15(10):3089-3101.
[8]Yan CM,Guo BL,Meng YI.Fast algorithm for nonsubsampled contourlet transform[J].Acta Automatica Sinica,2014,40(4):757-762.
[9]张强,郭宝龙.基于非采样Contourlet变换多传感器图像融合算法[J].自动化学报,2008,34(2):135-141.Zhang Q,Guo BL.Fusion of multi-sensor images based on the nonsubsampled contourlet transform[J].Acta Automatica Sinica,2008,34(2):135-141.
[10]Hubel DH,Wiesel TN.Receptive fields,binocular interaction and functional architecture in the cat’s visual cortex[J].Journal of Physiology,1962,160(1):106.
[11]Chouhan AS.An analytical study of leaky integrate and-fire neuron model using,MATLAB simulation[C]//International Journal of Engineering Research and Technology.ESRSA Publications,2013,2(4):2242-2245.
[12]Alpert S,Galun M,Brandt A,et al.Image segmentation by probabilistic bottom-up aggregation and cue integration[J].IEEE Transactions on Pattern Analysis&Machine Intelligence,2012,34(2):315-326.
[13]Li C Y.Integration fields beyond the classical receptive field:Organization and functional properties[J].News in Physiological Sciences,1996,11(4):181-186.
[14]Wei H,Zuo Q,Lang B.Multi-scale image analysis based on non-classical receptive field mechanism[C]//International Conference on Neural Information Processing,Springer Berlin Heidelberg,2011:601-610.
[15]Yang K,Gao S,Li C,et al.Efficient color boundary detection with color-opponent mechanisms[C]//Computer Vision and Pattern Recognition,IEEE,2013:2810-2817.
[16]寿天德.视觉信息处理的脑机制[M].北京:中国科学技术大学出版社,1997.Shou TD.Brain Mechanism of Visual Information Processing[M].Beijing:Press of University of Science and Technology of China,1997
[17]Dicarlo J,Zoccolan D,Rust N.How does the brain solve visual object recognition?[J].Neuron,2012,73(3):415-434.
[18]Hung CP,Kreiman G,Poggio T,et al.Fast readout of object identity from macaque inferior temporal cortex[J].Science,2005,310(5749):863.
[19]Evans HM.The emotional brain:the mysterious underpinnings of emotional life[J].The Quarterly Review of Biology,1999,43(4):91-95.
[20]Panksepp J.Affective neuroscience:The foundations of human and animal emotions[J].American Journal of Psychiatry,2000,159(10):1805.
[21]Yang KF,Gao SB,Guo CF,et al.Boundary detection using double-opponency and spatial sparseness constraint[J].IEEE Transactions on Image Processing,2015,24(8):2565-2578.
[22]Chen M,Yan Y,Gong X,et al.Incremental integration of global contours through interplay between visual cortical areas[J].Neuron,2014,82(3):682.