基于突触连接视通路方位敏感的图像分级边缘检测方法

视觉通路上的多级方位敏感特性对于视觉轮廓感知起着关键作用,将为更高层次的视皮层图像理解提供重要的特征信息。从视觉方位敏感机制出发,提出一种图像边缘检测的新方法。利用神经节细胞以及外膝体神经元感受野向心分布的生理结构特性,构建具有突触连接和多方向敏感特性的视皮层下功能层,融合多方向上的神经元脉冲发放信息,将视觉激励映射为边缘敏感图像;构建具有去最优方位感受野特性的初级视皮层的功能层,对前级结构生成的脉冲序列按时间信息进行神经编码,经过感受野内侧向抑制和阈值处理,获得边缘检测结果。对层次模糊而细节丰富的菌落图像进行处理,并以边缘置信度和重构相似度以及两者的加权和作为边缘检测评价指标。结果表明,该方法在完整检测图像边缘的同时,并不引入纹理噪声,有着明显的优势,其对12幅图像的加权和指标均值为0.746 8,显著高于其他对比方法。所提出的方法可以模拟视通路中初级视皮层及视皮层下的方向敏感特性,提供一种基于视觉机制的图像处理和理解新思路。     

基于神经元颜色拮抗与动态编码的轮廓检测方法

基于神经元的颜色拮抗特性及神经元群体的动态编码机制,实现对图像的轮廓检测。模拟视皮层下神经元的颜色单拮抗特性,引入单拮抗感受野的动态调节机制,以充分响应颜色边界和亮度边界;利用单细胞的树突极性分布,构建初级视皮层的双拮抗神经元网络,实现对特定方位的视觉刺激响应,有效提取目标轮廓;在神经元的群体感受野内,考虑神经元的动态突触连接,融合单细胞的脉冲频率响应,实现对纹理信息的抑制作用。以BSDS500图库的图像为实验对象,结果显示该方法在提取主体轮廓的过程中能有效抑制纹理信息,其对100幅图像最佳检测结果的P值指标均值和标准差为0.58±0.04,相对CORF和CO等其他对比方法,可提高轮廓提取的准确率。所提出方法可有效实现图像的轮廓检测,为利用颜色信息以及神经元之间的动态编码、实现更高级皮层的图像理解或者视觉认知提供新的思路。     

基于抑制性突触多层神经元群放电编码的图像边缘检测

图像边缘检测所获得的目标轮廓细节质量,对于后续图像分析或理解过程具有重要作用。提出一种基于视觉机制的图像边缘检测新方法:构建抑制性突触连接的多层神经元群,在待检测图像的激励下,分析7像素×7像素视觉感受野窗口内互连接神经元的脉冲放电过程,记录发放时刻进行次序编码;同时考虑神经元之间的侧向抑制作用,在选择注意机制作用下获得增强图像;之后利用Log Gabor滤波器模拟视觉系统中的方向选择特性,获取8个方向的滤波结果,经过输出层神经元群的融合处理并经灰度映射到0～255的范围后获得边缘图像。对含有丰富边缘细节特性的24幅菌落图像进行处理,其处理结果的ROC评价指标均值为0.698 4,优于PCNN法的0.659 3;从评价指标的均方差来看,该结果具有更好的一致性。另外,从信息熵评价指标来看,该方法同样具有一定的优势,能够有效提取图像的边缘信息,而且也能反映更多层次的图像细节。所提出的方法为利用视觉生理特性进行图像处理提供了崭新而有效的思路。     

基于人工视网膜神经节细胞感受野的图像边缘检测

用光电池组装一个工感受野，在一维上模拟视网膜神经节细胞的零交叉滤波特性，进一步用该工工感受野构建一个图像边缘检测系统，并成功地用系统检测到图像边缘。一与般的基于视觉原理的图像系统相比，这种以人工感受野灵基础的系统，在图像输入的同时即可并行地提取图像边缘信息，而更接近于动物视觉系统原型，此结果将为人工视觉和图像技术的发展提供新的探索。     

融合感受野模块的卷积神经网络视杯视盘联合分割

青光眼是世界第一大不可逆致盲性眼病,早期诊断和及时治疗是预防青光眼致盲的有效措施。眼底图像中的杯盘比是青光眼早期筛查与临床诊断的重要指标。因此,精确的视杯视盘分割是准确计算杯盘比并提高青光眼计算机辅助诊断技术的关键。针对这一问题,在对眼底图像进行极坐标变换的基础上,提出一种融合感受野模块的卷积神经网络Seg-RFNet,以实现视杯视盘联合分割。Seg-RFNet以SegNet框架为基础,使用ResNet50作为编码层,增强图像的特征提取能力,并对编码层进行分支处理,进一步获得更多的深层语义信息;同时在编码层和解码层之间加入感受野模块,模拟人类视觉系统,在增大感受野的同时增强了有用特征的响应。使用MICCAI 2018公开数据集REFUGE中的800张眼底图像对所提出方法与其他方法进行性能验证和比较。结果表明,Seg-RFNet分割视杯和视盘的Jaccard相似度分别0.951 5和0.872 0,F分数达到了0.974 9和0.930 1,与常用的U-Net、SegNet 网络相比,Seg-RFNet具有更好的视杯视盘联合分割精度,为计算杯盘比提供了精确分割基础。     

基于视觉感光层功能的菌落图像多强度边缘检测研究

借鉴视觉神经系统在轮廓感知中的独特优势,提出一种基于视觉感光层功能的图像边缘检测新方法。构建以带漏感的积累发放（LIF）神经元电生理模型为基本单元的神经元网络;根据特定时间窗口内各个神经元的脉冲发放情况,对神经元的增强(ON)或抑制(OFF)类别进行判断;通过拮抗式感受野特性以及神经元激励的反馈增强模式,实现弱边缘的凸显;为克服视觉感光层所具有的适应性并凸显弱细节的对比度,对图像进行多方向、多距离尺度的移动,并融合感光层神经元网络脉冲发放率的差异信息,最后实现图像边缘的有效检测。以具有丰富边缘特性的20幅菌落图像为样本,以边缘置信度和重构相似度作为评价指标,对多强度边缘进行检测。结果表明,所提出方法可以有效完整地检测出图像多强度边缘,且其对弱边缘检测的重构相似度均值高于08,检测准确性有显著的提高(P<005)。所提出的利用生理视觉系统特性进行边缘检测,为包含多强度边缘信息的图像处理提供崭新的思路。     

视觉信息处理的表象结构

本世纪60年代以来,视觉研究进展很快,取得了一系列重要结果。Hubel和Wiesel关于视皮层细胞感受野的论述,Campbell和Robsen关于视觉系统空间频率通道的论述,则是其中两项最重要的研究成果。如果把视觉看成是一种信息处理过程,上述结果就形成两种不同的观点:一些人认为视皮层是特征检测器,另一些人则认为视皮层是某种空     

基于样本熵的神经元放电脉冲间隔序列分析

目的神经元动作电位的放电脉冲间隔(interspikeinterval,ISI)序列包含丰富的时空编码信息,但这种丰富的时空编码信息所蕴含的生理信息没有得到很好的揭示,本文采用样本熵工具量化并分析 ISI序列中可能包含的生理信息.方法采用胞外多通道微电极记录猫的初级视皮层神经元对正弦光栅的反应动作电位,在不同时间尺度下采用抖动方法生成与原始动作电位序列在编码层次上等价的ISI序列,然后利用样本熵工具分析这些 ISI序列包含的生理信息,将样本熵与神经元发放率进行关联分析.本文还研究了猫初级视觉皮层细胞的方位选择性与样本熵的关系,并分析不同时间尺度对样本熵的影响.结果神经元的样本熵与发放率之间呈反相关关系,并在所有时间尺度上都显著.在猫初级视觉皮层8个细胞的方位选择性与样本熵之间的关系中,6个细胞具有显著反相关关系.而时间尺度的选择对样本熵的影响较小,呈弱反相关关系.结论基于样本熵的神经元 ISI序列分析结果表明,样本熵一定程度上反映细胞的生理信息,为研究其他神经体系动作电位包含的生理信息提供参考.     

猫视皮层ERK mRNA表达的研究

丝裂原激活的蛋白激酶又称细胞外信号调节激酶,具有神经元信号传导功能,参与多种发育过程。它由细胞外特殊信号激活,将酪氨酸磷酸化过程转化为丝氨酸／苏氨酸磷酸化过程,后者进一步调节下游的蛋白激酶。因此,细胞外信号调节激酶在磷酸化的连锁反应过程中起着枢纽作用。本文应用地高辛标记寡核苷酸探针的原位杂交技术,研究了细胞外信号调节激酶的２ 种亚型 ＥＲＫ１ 和ＥＲＫ２ ｍ ＲＮＡ 在生后４ 周龄的幼猫视皮层各层的分布。结果表明,ＥＲＫ１ ｍ ＲＮＡ 在视皮层的表达层次明确,以第Ⅴ层和第Ⅲ层的表达为最强,Ⅱ层呈中等表达,Ⅳ层较弱,Ⅵ层最弱,即在视皮层的表达强度为Ⅴ、Ⅲ＞ Ⅱ＞ Ⅳ＞ Ⅵ。与ＥＲＫ１相比,ＥＲＫ２ 的表达普遍较强,但层次欠清晰,其表达强度为Ⅴ、Ⅱ＞ Ⅲ、Ⅳ＞ Ⅵ。ＥＲＫ 的２ 种亚型在视皮层的Ⅰ层即分子层均不表达。ＥＲＫ１ 和ＥＲＫ２ ｍ ＲＮＡ 在视皮层的这种特征性分布提示,在幼猫的视皮层发育过程中,ＥＲＫ 可能具有独特的信号传递作用。     

一种应用于视皮质假体图像压缩及编码系统的设计

背景:目前,视皮质假体已成为视觉修复实现的主要方法之一,图像压缩在视皮质假体的前期图像处理过程中占有重要地位。目的:探讨从空域及亮度信息两个方面分别对图像信息进行压缩的方法。方法:选取哈尔小波基对原始图像进行空间分辨率的压缩,并采用多尺度小波变换的方法对空间频率信号进行合理选取,实现信息的进一步压缩。同时,结合视觉皮质放大的特性,通过模拟视网膜对图像的前期处理,建立了非均匀压缩模型,使得图像压缩过程更接近视觉处理过程。还对亮度信息进行压缩及编码,实现了图像信息最终的压缩编码。结果与结论:构建了基于数字信号处理器的前期图像压缩及编码系统,最终实现了在10*10的点阵图阵列中进行图像信息的表达。     

Model Studies of the Mechanisms of Tuning of Visual Cortex Neurons to Incomplete Cross-Shaped Figures

Numerical simulation modeling of the receptive fields of visual cortex neurons able to detect cross-shaped figures with masked central or peripheral areas was performed. Receptive field models of two types were considered: those with antagonistic and cooperative interactions between the center and the periphery. Model neurons with receptive fields with reciprocal (antagonistic) interactions produced greater responses to peripheral or central crosses than to complete crosses. Studies using the model showed that the basis of this type of tuning could be provided by a disinhibition mechanism: blockade of the inhibitory zones in the center or periphery of the receptive field by activation of a lateral disinhibitory zone. A model with cooperative interactions between the center and periphery of the receptive field was also studied, in which responses to complete crosses were summed from the responses to the peripheral and central parts. Tuning of these model receptive fields was comparable to the sensitivity of real visual cortex neurons to the shape, size, and orientation of figures. The properties of model receptive fields (configuration, localization, and weightings of the various zones) allowing simulation of the properties of cat visual cortex field 17 neurons sensitive to the orientation and configuration of incomplete cross-shaped figures were identified.     

Unusually large receptive fields in cats with restricted visual experience

Summary The receptive fields of striate cortex neurons were analyzed in cats which had restricted or no visual experience. Two groups of animals were investigated: 1. cats which were deprived from contour vision over variable periods of time up to 1 year and 2. kittens whose visual experience was restricted to vertically oriented gratings of constant spatial frequency which moved unidirectionally at a fixed distance in front of the restrained animals. In both preparations exceedingly large receptive fields (up to 20° in diameter) were encountered, especially in cells located in supragranular layers. These large receptive fields never extended over more than 2° into the ipsilateral hemifield. Their sensitivity profile was frequently asymmetric and contained discontinuities. Many of these large receptive fields consisted of several excitatory subregions which were separated from each other by as much as 15°. Often but not always the most sensitive area was located where the retinotopic map predicted the receptive field center. The orientation and direction selectivity and also the angular separation of such multiple excitatory bands often matched precisely the orientation, direction and spatial frequency of the experienced moving grating. In other fields with multiple excitatory subregions such a correspondence could not be established; the various subregions could even have different orientation and direction selectivities. From these unconventional receptive fields it is concluded that the function of cat striate cortex is not confined to a point by point analysis of the visual field in retinotopically organized and functionally isolated columns.     

Receptive-field properties of neurons in different laminae of visual cortex of the cat

1. Receptive-field properties of neurons in the different layers of the visual cortex of normal adult cats were analyzed quantitatively. Neurons were classified into one of two groups: 1) S-cells, which have discrete on- and/or off-regions in their receptive fields and possess inhibitory side bands; 2) C-cells, which do not have discrete on- and off-regions in their receptive fields but display an on-off response to flashing stimuli. Neurons of this type rarely display side-band inhibition. 2. As a group, S-cells display lower relative degrees of binocularity and are more selective for stimulus orientation than C-cells. In addition, within a given lamina the S-cells have smaller receptive fields, lower cutoff velocities, lower peak responses to visual stimulation, and lower spontaneous activity than do the C-cells. 3. S-cells in all layers of the cortex display similar orientation sensitivities, mean spontaneous discharge rates, peak response to visual stimulation, and degrees of binocularity. 4. Many of the receptive-field properties of cortical cells vary with laminar location. Receptive-field sizes and cutoff velocities of S-cells and of C-cells are greater in layers V and VI than in layers II-IV. For S-cells, preferred velocities are also greater in layers V and VI than in layers II-IV. Furthermore, C-cells in layers V and VI display high mean spontaneous discharge rates, weak orientation preferences, high relative degrees of binocularity, and higher peak responses to visual stimulation when compared to C-cells in layers II and III. 5. The receptive-field properties of cells in layers V-VI of the striate cortex suggest that most neurons that have their somata in these laminae receive afferents from LGNd Y-cells. Hence, our results suggest that afferents from LGNd Y-cells may play a major part in the cortical control of subcortical visual functions.     

Spatial structure and symmetry of simple-cell receptive fields in macaque primary visual cortex

I present measurements of the spatial structure of simple-cell receptive fields in macaque primary visual cortex (area V1). Similar to previous findings in cat area 17, the spatial profile of simple-cell receptive fields in the macaque is well described by two-dimensional Gabor functions. A population analysis reveals that the distribution of spatial profiles in primary visual cortex lies approximately on a one-parameter family of filter shapes. Surprisingly, the receptive fields cluster into even- and odd-symmetry classes with a tendency for neurons that are well tuned in orientation and spatial frequency to have odd-symmetric receptive fields. The filter shapes predicted by two recent theories of simple-cell receptive field function, independent component analysis and sparse coding, are compared with the data. Both theories predict receptive fields with a larger number of subfields than observed in the experimental data. In addition, these theories do not generate receptive fields that are broadly tuned in orientation and low-pass in spatial frequency, which are commonly seen in monkey V1. The implications of these results for our understanding of image coding and representation in primary visual cortex are discussed.     

Population receptive fields of ON and OFF thalamic inputs to an orientation column in visual cortex

The primary visual cortex of primates and carnivores is organized into columns of neurons with similar preferences for stimulus orientation, but the developmental origin and function of this organization are still matters of debate. We found that the orientation preference of a cortical column is closely related to the population receptive field of its ON and OFF thalamic inputs. The receptive field scatter from the thalamic inputs was more limited than previously thought and matched the average receptive field size of neurons at the input layers of cortex. Moreover, the thalamic population receptive field (calculated as ON - OFF average) had separate ON and OFF subregions, similar to cortical neurons in layer 4, and provided an accurate prediction of the preferred orientation of the column. These results support developmental models of orientation maps that are based on the receptive field arrangement of ON and OFF visual inputs to cortex.     

Primary visual cortex neurons that contribute to resolve the aperture problem

It is traditional to believe that neurons in primary visual cortex are sensitive only or principally to stimulation within a spatially restricted receptive field (classical receptive field). It follows from this that they should only be capable of encoding the direction of stimulus movement orthogonal to the local contour, since this is the only information available in their classical receptive field "aperture." This direction is not necessarily the same as the motion of the entire object, as the direction cue within an aperture is ambiguous to the global direction of motion, which can only be derived by integrating with unambiguous components of the object. Recent results, however, show that primary visual cortex neurons can integrate spatially and temporally distributed cues outside the classical receptive field, and so we reexamined whether primary visual cortex neurons suffer the "aperture problem." With the stimulation of an optimally oriented bar drifting across the classical receptive field in different global directions, here we show that a subpopulation of primary visual cortex neurons (25/81) recorded from anesthetized and paralyzed marmosets is capable of integrating informative unambiguous direction cues presented by the bar ends, well outside their classical receptive fields, to encode global motion direction. Although the stimuli within the classical receptive field were identical, their directional responses were significantly modulated according to the global direction of stimulus movement. Hence, some primary visual cortex neurons are not local motion energy filters, but may encode signals that contribute directly to global motion processing.     

Development of cortical orientation selectivity in the absence of visual experience with contour

Visual cortical neurons are selective for the orientation of lines, and the full development of this selectivity requires natural visual experience after eye opening. Here we examined whether this selectivity develops without seeing lines and contours. Juvenile ferrets were reared in a dark room and visually trained by being shown a movie of flickering, sparse spots. We found that despite the lack of contour visual experience, the cortical neurons of these ferrets developed strong orientation selectivity and exhibited simple-cell receptive fields. This finding suggests that overt contour visual experience is unnecessary for the maturation of orientation selectivity and is inconsistent with the computational models that crucially require the visual inputs of lines and contours for the development of orientation selectivity. We propose that a correlation-based model supplemented with a constraint on synaptic strength dynamics is able to account for our experimental result.