Masking unveils pre-amodal completion representation in visual search

When one object is partly occluded by another, its occluded parts are perceptually 'filled in', that is, the occluded object appears to continue behind its occluder. This process is known as amodal completion. The completion of a partially occluded object takes about 200 ms, and pre-completion information (that is, information from before amodal completion has occurred) exists in the visual system for that duration. It has been suggested, however, that observers cannot make use of this information, even when it is beneficial to do so: visual search for a target that appears to be partly occluded, for example, is slower than for a target that does not undergo occlusion, despite both targets being physically identical. Here we show that visual search does have access to pre-completion representations, but only for a limited time that depends on the size of the occluded region.     

Anticipation of moving stimuli by the retina

A flash of light evokes neural activity in the brain with a delay of 30-100 milliseconds, much of which is due to the slow process of visual transduction in photoreceptors. A moving object can cover a considerable distance in this time, and should therefore be seen noticeably behind its actual location. As this conflicts with everyday experience, it has been suggested that the visual cortex uses the delayed visual data from the eye to extrapolate the trajectory of a moving object, so that it is perceived at its actual location. Here we report that such anticipation of moving stimuli begins in the retina. A moving bar elicits a moving wave of spiking activity in the population of retinal ganglion cells. Rather than lagging behind the visual image, the population activity travels near the leading edge of the moving bar. This response is observed over a wide range of speeds and apparently compensates for the visual response latency. We show how this anticipation follows from known mechanisms of retinal processing.     

Motion direction, speed and orientation in binocular matching

The spatial differences between the images seen by the two eyes, called binocular disparities, can be used to recover the volumetric (three-dimensional) aspects of a scene. The computation of disparity depends upon the correct identification of corresponding features in the two images. Understanding what image features are used by the brain to solve this matching problem is one of the main issues in stereoscopic vision. Many cortical neurons in visual areas V1 (ref. 2), MT (refs 3, 4) and MST (refs 5, 6) that are tuned to binocular disparity are also tuned to orientation, motion direction and speed. Although psychophysical work has shown that motion direction can facilitate binocular matching, the psychophysical literature on the role of orientation is mixed, and it has been argued that speed differences are ineffective in aiding correspondence. Here we use a different psychophysical paradigm to show that the visual system uses similarities in orientation, motion direction and speed to achieve binocular correspondence. These results indicate that cells that multiplex orientation, motion direction, speed and binocular disparity may help to solve the binocular matching problem.     

Gaze direction controls response gain in primary visual-cortex neurons

To localize objects in space, the brain needs to combine information about the position of the stimulus on the retinae with information about the location of the eyes in their orbits. Interaction between these two types of information occurs in several cortical areas, but the role of the primary visual cortex (area V1) in this process has remained unclear. Here we show that, for half the cells recorded in area V1 of behaving monkeys, the classically described visual responses are strongly modulated by gaze direction. Specifically, we find that selectivity for horizontal retinal disparity-the difference in the position of a stimulus on each retina which relates to relative object distance-and for stimulus orientation may be present at a given gaze direction, but be absent or poorly expressed at another direction. Shifts in preferred disparity also occurred in several neurons. These neural changes were most often present at the beginning of the visual response, suggesting a feedforward gain control by eye position signals. Cortical neural processes for encoding information about the three-dimensional position of a stimulus in space therefore start as early as area V1.     

基于仿射不变块相似度量的BM3D图像去噪算法

针对使用规则的正方形块去噪可能产生伪影, 或图像受到仿射形变时使用正方形块寻找到的相似块数量将会减少的问题, 提出一种基于仿射不变块相似度量的BM3D(block matching and 3D filtering)图像去噪算法. 首先, 在块匹配过程中, 使用椭圆块代替规则的正方形块进行块匹配分组, 形成三维椭圆块组; 其次, 使用仿射不变块相似度量计算两个椭圆块之间的距离判断其相似度, 该度量通过自动调整块大小与形状进行块比较, 能寻找到更多的相似块, 将其用于图像去噪. 实验结果表明, 该算法可有效提高去噪效果, 获得良好的视觉效果, 与原始BM3D算法相比, 具有较好的峰值信噪比.     

Eye-specific effects of binocular rivalry in the human lateral geniculate nucleus

When dissimilar images are presented to the two eyes, they compete for perceptual dominance so that each image is visible in turn for a few seconds while the other is suppressed. Such binocular rivalry is associated with relative suppression of local, eye-based representations that can also be modulated by high-level influences such as perceptual grouping. However, it is currently unclear how early in visual processing the suppression of eye-based signals can occur. Here we use high-resolution functional magnetic resonance imaging (fMRI) in conjunction with a new binocular rivalry stimulus to show that signals recorded from the human lateral geniculate nucleus (LGN) exhibit eye-specific suppression during rivalry. Regions of the LGN that show strong eye-preference independently show strongly reduced activity during binocular rivalry when the stimulus presented in their preferred eye is perceptually suppressed. The human LGN is thus the earliest stage of visual processing that reflects eye-specific dominance and suppression.     

Three-dimensional illusory contours and surfaces.

Under general viewing conditions, objects are often partially camouflaged, obscured or occluded, thereby limiting information about their three-dimensional position, orientation and shape to incomplete and variable image cues. When presented with such partial cues, observers report perceiving 'illusory' contours and surfaces (forms) in regions having no physical image contrast. Here we report that three-dimensional illusory forms share three fundamental properties with 'real' forms: (1) the same forms are perceived using either stereo or motion parallax cues (cue invariance); (2) they retain their shape over changes in position and orientation relative to an observer (view stability); and (3) they can take the shape of general contours and surfaces in three dimensions (morphic generality). We hypothesize that illusory contours and surfaces are manifestations of a previously unnoticed visual process which constructs a representation of three-dimensional position, orientation and shape of objects from available image cues.     

An unexpected specialization for horizontal disparity in primate primary visual cortex

The horizontal separation of the eyes means that objects nearer or farther than the fixation point project to different locations on the two retinae, differing principally in their horizontal coordinates (horizontal binocular disparity). Disparity-selective neurons have generally been studied with disparities applied in only one direction (often horizontal), which cannot determine whether the encoding is specialized for processing disparities along the horizontal axis. It is therefore unclear if disparity selectivity represents a specialization for naturally occurring disparities. I used random dot stereograms to study disparity-selective neurons from the primary visual cortex (V1) of awake fixating monkeys. Many combinations of vertical and horizontal disparity were used, characterizing the surface of responses as a function of two-dimensional disparity. Here I report that the response surface usually showed elongation along the horizontal disparity axis, despite the isotropic stimulus. Thus these neurons modulated their firing rate over a wider range of horizontal disparity than vertical disparity. This demonstrates that disparity-selective cells are specialized for processing horizontal disparity, and that existing models of disparity selectivity require substantial revision.     

基于样本的图像修补

提出了一种基于样本的图像修补方法。该方法首先计算填充前缘上每个修补块的优先权,确定填充前缘上最高优先权的修补块;然后在样本区域中寻找最高优先权修补块的最佳样本块,拷贝最佳样本块的有关图像数据到最高优先权修补块的未填充好的像素点位置;再更新这些刚填充的像素点的自信度。重复这些步骤直至整个图像受损区域填充修补完毕。实验结果表明,该方法能同时有效地修补图像受损区域的纹理和结构,还能较好地去除图像上不需要的物体。     

Probing the human stereoscopic system with reverse correlation.

Our two eyes obtain slightly different views of the world. The resulting differences in the two retinal images, called binocular disparities, provide us with a stereoscopic sense of depth. The primary visual cortex (V1) contains neurons that are selective for the disparity of individual elements in an image, but this information must be further analysed to complete the stereoscopic process. Here we apply the psychophysical technique of reverse correlation to investigate disparity processing in human vision. Observers viewed binocular random-dot patterns, with 'signal' dots in a specific depth plane plus 'noise' dots with randomly assigned disparities. By examining the correlation between the observers' ability to detect the plane and the particular sample of 'noise' disparities presented on each trial, we revealed detection 'filters', whose disparity selectivity was remarkably similar to that of individual neurons in monkey V1. Moreover, if the noise dots were of opposite contrast in the two eyes, the tuning inverted, just like the response patterns of V1 neurons. Reverse correlation appears to probe disparity processing at the earliest stages of binocular combination, prior to the generation of a full stereoscopic depth percept.     

Ketamine-xylazine anaesthesia blocks consolidation of ocular dominance changes in kitten visual cortex

In the visual cortex of mammals, response properties of single neurons can be changed by restricted visual experience during early postnatal development. Covering one eye for four to eight hours when kittens are at the peak of the sensitive period is sufficient to weaken the influence of the occluded eye on cortical neurons resulting in a noticeable shift of ocular dominance towards the open eye. The underlying changes in synaptic connections do not occur so readily when a kitten is anaesthetized and paralysed. We report here that an ocular dominance shift is prevented in alert kittens that receive repeated brief monocular exposures when these are followed by ketamine-xylazine anaesthesia. This retrograde effect on cortical plasticity suggests that the process by which synaptic activity is converted into structural changes has been disturbed.     

Recognition of partially occluded objects based on segmentation by median filtering of extended direction code

The image contour is segmented into lines, arcs and smooth curves by median filtering of extended direction code. Based on this segmentation, a set of new local invariant features are proposed to recognize partially occluded objects, which is more reasonable compared with conventional corner features. The matching results of some typical examples shows that these features are robust, effective in recognition. Supported by the National Defence Foundation of China Wang Yanping: born in Aug. 1937, Professor     

Topography of contextual modulations mediated by short-range interactions in primary visual cortex.

Neurons in primary visual cortex (V1) respond differently to a simple visual element presented in isolation from when it is embedded within a complex image. This difference, a specific modulation by surrounding elements in the image, is mediated by short- and long-range connections within V1 and by feedback from other areas. Here we study the role of short-range connections in this process, and relate it to the layout of local inhomogeneities in the cortical maps of orientation and space. By measuring correlation between neuron pairs located in optically imaged maps of V1 orientation columns we show that the strength of local connections between cells is a graded function of lateral separation across cortex, largely radially symmetrical and relatively independent of orientation preferences. We then show the contextual influence of flanking visual elements on neuronal responses varies systematically with a neuron's position within the cortical orientation map. The strength of this contextual influence on a neuron can be predicted from a model of local connections based on simple overlap with particular features of the orientation map. This indicates that local intracortical circuitry could endow neurons with a graded specialization for processing angular visual features such as corners and T junctions, and this specialization could have its own functional cortical map, linked with the orientation map.     

基于感知的大范围人造对象图像结构特征抽取

基于内容的图像检索(CBIR)系统传统上仅仅使用颜色特征和纹理特征进行图像的底层分析,不足以满足检索性能的要求.要提高基于内容图像检索的鲁棒性,需要检索系统能分析更多的图像底层特征.对于包含大范围人造对象(比如:建筑物、高塔、桥梁和其他结构对象)的图像,提出应用感知分组(perceptual grouping)规则抽取图像的结构特征,并结合使用该结构特征,对特定的图像库进行分类和客观评价.     

梯度直方图约束的多尺度块先验模型

块先验模型在图像复原领域取得了较大的成功,但其整体模型强制局部性的缺点,易出现局部伪影、视觉观感较差的问题,提出一种新的集成多尺度块先验和梯度直方图先验的图像复原方法.对原始图像实施滤波和下采样以保持尺度不变性,在多尺度上施加同一局部块模型,即保持块低维模型的简单性,又在图像较大区域实施非局部性;将梯度直方图全局统计先验加入正则约束中,利用Wasserstein距离对复原图像与参考直方图的相似性进行度量.借助半二次分裂和最优传递理论,求解所提出的模型.通过在图像去噪和去模糊实验,相比传统方法无论在客观质量评价还是视觉观感上都更有优势,验证了方法的有效性.     

Image Tagging by Semantic Neighbor Learning Using User-Contributed Social Image Datasets

The explosive increase in the number of images on the Internet has brought with it the great challenge of how to effectively index, retrieve, and organize these resources. Assigning proper tags to the visual content is key to the success of many applications such as image retrieval and content mining. Although recent years have witnessed many advances in image tagging, these methods have limitations when applied to high-quality and large-scale training data that are expensive to obtain. In this paper, we propose a novel semantic neighbor learning method based on user-contributed social image datasets that can be acquired from the Web's inexhaustible social image content. In contrast to existing image tagging approaches that rely on high-quality image-tag supervision, we acquire weak supervision of our neighbor learning method by progressive neighborhood retrieval from noisy and diverse user-contributed image collections. The retrieved neighbor images are not only visually alike and partially correlated but also semantically related. We offer a step-by-step and easy-to-use implementation for the proposed method. Extensive experimentation on several datasets demonstrates that the performance of the proposed method significantly outperforms others.     

视频序列中面向行人的多目标跟踪算法

为实现视频序列中多行人目标跟踪,基于多信息融合方法,考虑多目标间严重遮挡,建立面向行人的多目标跟踪算法.提出多信息融合算法融合目标颜色和运动信息,结合均值漂移算法思想,实现常态下目标跟踪.针对多行人目标参与的遮挡,通过理论分析遮挡过程中目标面积变化,提出遮挡因子判别遮挡发生、辨识遮挡者和被遮挡对象、确认被遮挡对象重新出现等.实验结果表明,该方法能够正确跟踪行人目标,判断并处理多目标间的严重遮挡.      

稳健水印的验证算法

研究了稳健水印验证模型,提出一种以人类视觉掩蔽值作为量化阶的DWT域稳健水印认证算法.该算法能够区分普通的图像处理(如图像增强、压缩等)和非法纂改(如剪切攻击),而且对于一般的图像处理操作还能根据水印误差图样来判断图像经过了何种处理.最后详细论述提出的算法并给出仿真结果.     

Neural mechanisms of perceptual grouping in human visual cortex

The current work examined neural substrates of perceptual grouping in human visual cortex using event-related potential (ERP) recording. Stimulus arrays consisted of local elements that were either evenly spaced (uniform stimuli) or grouped into columns or rows by proximity or color similarity (grouping stimuli). High-density ERPs were recorded while subjects identified orientations of perceptual groups in stimulus arrays that were presented randomly in one of the four quadrants of the visual field.Both uniform and grouping stimulus arrays elicited an early ERP component (C1), which peaked at about 70ms after stimulus onset and changed its polarity as a function of stimulated elevations. Dipole modeling based on realistichead boundary-element models revealed generators of the C1 component in the calcarine cortex. The C1 was modulated by perceptual grouping of local elements based on proximity, and this grouping effect was stronger in the upper than in the lower visual field. The findings provide ERP evidence for the engagement of human primary visual cortex in the early stage of perceptual grouping.     

Dynamic programming based ray-space interpolation for free-viewpoint video system

This paper presents a free viewpoint video (FVV) system based on ray-space interpolation method. The new algorithm matches individual pixels in corresponding scanline pairs by using a dynamic programming technique. A sparse intermediate view disparity map is projected from matched pixels firstly, and the holes (occluded pixels) are filled in by propagating the disparity of neighboring background pixels. After interpolating dense view images, an arbitrary virtual view image can be easily rendered from the dense ray-space converted from these view images. The proposed method is evaluated on the Middlebury data set and compared with other methods, experimental results show that the better quality of the intermediate view is obtained and the corresponding computational complexity is reduced significantly.