Dynamic properties of neurons in cortical area MT in alert and anaesthetized macaque monkeys.

In order to see the world with high spatial acuity, an animal must sample the visual image with many detectors that restrict their analyses to extremely small regions of space. The visual cortex must then integrate the information from these localized receptive fields to obtain a more global picture of the surrounding environment. We studied this process in single neurons within the middle temporal visual area (MT) of macaques using stimuli that produced conflicting local and global information about stimulus motion. Neuronal responses in alert animals initially reflected predominantly the ambiguous local motion features, but gradually converged to an unambiguous global representation. When the same animals were anaesthetized, the integration of local motion signals was markedly impaired even though neuronal responses remained vigorous and directional tuning characteristics were intact. Our results suggest that anaesthesia preferentially affects the visual processing responsible for integrating local signals into a global visual representation.     

基于小波明显区域及空间分布的图像检索

提出了一种基于小波变换的明显区域检测方法,并改进了环型分割算法,使对视觉有意义的区域和区域特征提取更加快捷、方便。该算法不仅考虑到区域内的图像特征,而且还考虑到明显区域的空间分布信息,并把环型区域的颜色矩和在明显区域附近的Gabor特点,作为索引图像的特征向量。使用Corel图像库测试了提出的方法。实验表明,该方法切实可行。     

Perceiving visual expansion without optic flow

When an observer moves forward in the environment, the image on his or her retina expands. The rate of this expansion conveys information about the observer's speed and the time to collision. Psychophysical and physiological studies have provided abundant evidence that these expansionary motions are processed by specialized mechanisms in mammalian visual systems. It is commonly assumed that the rate of expansion is estimated from the divergence of the optic-flow field (the two-dimensional field of local translational velocities). But this rate might also be estimated from changes in the size (or scale) of image features. To determine whether human vision uses such scale-change information, we have synthesized stochastic texture stimuli in which the scale of image elements increases gradually over time, while the optic-flow pattern is random. Here we show, using these stimuli, that observers can estimate expansion rates from scale-change information alone, and that pure scale changes can produce motion after-effects. These two findings suggest that the visual system contains mechanisms that are explicitly sensitive to changes in scale.     

边缘检测的分层松弛方法

本文基于图象的分层表示,提出了一种边缘检测的分层松弛方法。实验结果表明,本文方法计算简单,运算速度快,而且不受噪声影响,能有效地应用于细胞图象的边缘检测。     

基于多分支CNN和多尺度特征融合的非受控人脸年龄估计

现有的人脸年龄估计不能很好地兼顾全局-局部细节的特征表达,因而非受控人脸年龄估计的精度存在一定的提升空间。为解决此问题,提出了一种基于多分支卷积神经网络(convolutional neural networks,CNN)和多尺度特征融合的非受控人脸年龄估计方法。该方法根据人脸关键点对人脸图片剪裁得到包含人脸的全局图像和分别包含眼睛、鼻子、嘴巴的局部图像;使用多分支CNN网络提取对应的深层全局特征和局部特征,使用多尺度特征融合网络探索局部特征间的相关性信息从而进行局部特征选择;将融合的局部特征与全局特征拼接得到兼顾全局-局部细节的年龄特征;使用softmax损失函数优化模型进行人脸年龄估计。根据MORPH Album2、FG-NET、LAP2016人脸年龄数据集上的实验结果表明,提出的方法是有效的。     

基于深度CNN的改进弱监督学习方法设计与验证

针对包含目标、尺度和平移变化较强的空间信息难以获取大量训练样本的问题,提出一种基于深度卷积神经网络(deep convolutional neural network,DCCN)的弱监督学习方法,从3个层面对当前卷积神经网络进行扩展。为了提取分辨率更高的局部特征,同时考虑到全卷积网络(full convolution network,FCN)在全监督式学习下的高效性能,使用FCN作为后端模块;为了获取更多的通用特征,增加一个多映射弱监督学习的传输层,对与补充性类模态相关的多个局部特征进行显式学习;为了优化训练过程,改进了池化层,使用全局图像标签进行训练,将空间得分聚合为全局预测。使用图像分类、弱监督逐点目标定位和图像分割3种常用的机器视觉任务进行评估。多个公开数据库的实验结果表明,所提方法能够有效地学习强局部特征,具有良好的分类和定位效果。     

结合卷积与转置卷积特征的模糊车牌复原方法

深度学习算法在图像去噪领域已经得到了很好的效果;但目前对于深度学习算法在模糊图像复原领域的研究没有更深入的研究。直接应用图像去噪的方法对模糊车牌进行复原实际上可行的,但会产生复原图像细节缺失,时间代价高的缺点。针对这些问题,吸取去噪方法的优点,提出将原始图像信息与转置卷积复原后的图像信息相结合的方法,重新构建了图像复原网络结构;并根据图像特点自定了损失函数。实验通过与已有的方法进行对比说明,提出的复原方法在复原车牌图像质量上和复原效率上都有很好的表现;同时对模糊运动角度与不同噪声具有健壮性;而模糊运动像素越大的图片,复原图像的质量也会下降。     

利用多颜色空间特征融合方法检测近似目标

以棉花中羊毛、白头发、塑料膜等杂质的检测为应用背景,提出一种利用多颜色空间特征融合方法。该方法构建了颜色特征评价函数,对近似目标在不同颜色空间的特征表现进行评估,从中抽取近似目标的若干最优特征;再利用区域信息相关度权值小波分析算法进行多特征融合,获取近似目标的图像。实验结果表明,融合图像比原始图像及单色空间图像具有较高的图像信息量值,近似目标视觉特征明显增强。此方法为提取与背景特征相近的近似目标提供一条新思路。     

Correction of Breathing Motion in the Thorax for Helical CT

This paper investigates a reconstruction method for helical computed tomography which compensates for the motion artifacts in the thorax caused by patient breathing.The method takes into account a motion vector field determined from a four-dimensional(4-D) uncompensated image data set.Surface models of the lung and the ribs are tracked through the 4-D data set to create motion information within the entire thorax.Finally,an image is reconstructed using motion compensated back-projection.The results show tha...     

基于人工免疫算法的光学影像与SAR影像配准方法

提出了一种基于人工免疫算法的光学影像和SAR影像配准方法,该方法从影像上的面状地物入手,仅从识别性较好的光学影像上提取面状地物,先随机给定一组配准参数,将光学影像上面状地物的坐标经仿射变换获得新的坐标,以转换后新坐标在SAR影像上对应区域的均质性为评价标准,并利用人工免疫算法对配准参数进行优化,从而得到影像配准结果.最后,利用WorldView-2和RadarSat-2影像的配准实验验证该方法的有效性,结果表明该方法配准精度可优于2像素.     

Honeybee dances communicate distances measured by optic flow

In honeybees, employed foragers recruit unemployed hive mates to food sources by dances from which a human observer can read the distance and direction of the food source. When foragers collect food in a short, narrow tunnel, they dance as if the food source were much farther away. Dancers gauge distance by retinal image flow on the way to their destination. Their visually driven odometer misreads distance because the close tunnel walls increase optic flow. We examined how hive mates interpret these dances. Here we show that recruited bees search outside in the direction of the tunnel at exaggerated distances and not inside the tunnel where the foragers come from. Thus, dances must convey information about the direction of the food source and the total amount of image motion en route to the food source, but they do not convey information about absolute distances. We also found that perceived distances on various outdoor routes from the same hive could be considerably different. Navigational errors are avoided as recruits and dancers tend to fly in the same direction. Reported racial differences in honeybee dances could have arisen merely from differences in the environments in which these bees flew.     

融合全局与局部特征的UWB雷达人体动作识别算法

为了消除雷达信号中杂波和噪声对人体动作识别的干扰,提高小样本数据下动作识别的精度,在去除杂波及噪声干扰的基础上,提出一种融合全局与局部特征的超宽带（ultra-wideband,UWB）雷达人体动作识别算法。用动目标指示（moving target indication,MTI）结合自适应中值滤波对雷达原始回波信号进行预处理,再对人体动作的雷达二维特征图像利用主成分分析（principal component analysis,PCA）提取主要分量作为全局特征表征,并用二维离散小波变换（2D discrete wavelet transform,2D-DWT）结合奇异值分解（singular value decomposition,SVD）获取特征图像在不同方向与尺度划分下动作的局部特征表征,并将全局与局部特征进行串联融合;根据融合特征,在网格搜索算法（grid search,GS）优化的支持向量机（support vector machines,SVM）模型中实现人体动作的识别分类。实验结果表明,该算法能有效获取雷达信号中的人体动作信息,平均识别准确率为95.63%,具有良好的识别性能。     

局部特征与全局信息联合的自适应图像增强算法

针对传统基于修正直方图的图像增强算法不能兼顾局部特征和全局信息的问题,提出一种局部特征与全局信息联合的自适应图像增强算法. 该算法将增强分为局部增强和全局增强两部分,局部增强利用像素的邻域信息和局部与全局对比度的比例信息作为幂次变换的伽马值,对图像进行伽马校正,提高图像的亮度和局部对比度;全局增强利用区域相似直方图统计抑制噪声,避免过度增强. 实验结果表明,本文算法在客观性能上优于其它传统图像增强算法,并且可以有效提高复杂光照下人脸图像的检测率.      

一种基于视觉感知的复合医学图像分割算法

为提高医学图像分割的视觉效果,依据人类视觉感知的分层特性,提出了一种新的复合医学图像分割方法.该方法通过提取医学图像的底层特征,利用Fuzzy-ART神经网络作为像素的分类器,对医学图像进行连续两次分割.实验结果表明,该医学图像分割方法能有效地解决局部信息与整体分布边缘淡化等相关问题,达到良好的分割视觉效果.     

Interaction between colour and motion in human vision

There is a wealth of anatomical and psychological evidence which suggests that when people look at an object in the visual world, its various attributes such as colour, 'form', motion and depth are analysed by separate channels in the visual system. If so, how are these attributes put back together again to create a unified picture of the object? And if the object moves rapidly, how is perfect perceptual synchrony maintained between different features on its surface, if it is indeed true that they are being processed separately? Our evidence suggests that the visual system extracts certain conspicuous image features based on luminance contrast, and that the signals derived from these are then attributed to other features on the object, a process that we call 'capture'. Specifically, we find that when either illusory contours or random-dot patterns are moved in the vicinity of a colour-border, the colour border will also seem to move in the same direction even though it is physically stationary.     

基于级联卷积神经网络的人脸特征点识别算法实现

针对在有冗余图像信息干扰下进行人脸有效特征点提取时精度不高的问题,提出了基于级联卷积神经网络的人脸特征点检测算法.在该算法中:输入层读入规则化的原始图像,神经元提取图像的局部特征;池化层进行局部平均和降采样操作,对卷积结果降低维度;卷积层和池化层分布连接,迭代训练,输出特征点检测结果.该算法采用Python语言编程实现,在人脸数据集进行仿真实验,结果表明该算法对人脸特征点有较高的识别率.     

Semi-reference image quality metric for virtual view images in free-viewpoint television

A semi-reference image quality assessment metric based on similarity measurement for synthesized virtual viewpoint image (VVI) in free-viewpoint television system (FTV) is proposed in this paper. The key point of the proposed metric is taking resemblant information between VVI and its neighbor view images for quality assessment to make our metric to be extended to multi-semi-reference image quality assessment easily. The proposed metric first extracts impact factors from image features, then combines an image synthesis technique and similarity functions, in which, disparity information are taken into account for registering the resemblant regions. Experiments are divided into three phases. Phase I is to verify the validation of the proposed metric by taking impaired images and original reference into account. The experimental results show the agreement between evaluation scores and bio-characteristic of human visual system. Phase II shows the accordance with Phase I by taking neighbor view as reference. The proposed metric can be taken as a full reference one to evaluate the image quality even though the original reference is absent. Phase III is then performed to evaluate the quality of VVI. Evaluation scores in the experimental results are able to evaluate the quality of VVI.     

序列图像运动自适应V1-MT光流估计算法

针对传统算法在抗光照变化影响、大位移光流和异质点滤除等方面的不足,从人类视觉认知机理出发,提出了一种基于机器学习和生物模型的运动自适应V1-MT(motion-adaptive V1-MT,MAV1MT)序列图像光流估计算法.首先,引入基于ROF模型的结构纹理分解(structure-texture decomposition,STD)技术,有效解决了光照和色彩变化的影响.其次,利用多V1细胞加权组合及非线性正则化模拟MT细胞模型,并结合岭回归训练学习得到运动自适应的权重,解决对目标的运动速度感知问题.最后,引入由粗到精的增强方法和图像金字塔局部运动估计采样,将V1-MT运动估计模型应用于实际大位移视频序列.理论分析和实验结果表明,新方法能更加拟合人眼视觉信息处理特性,对视频序列具有普适、有效、鲁棒的运动感知性能.     

Weight-Edge Convolution Neural Network for Point Clouds Learning

As a kind of flexible three-dimensional geometric data, point clouds can accomplish many challenging tasks so long as the rich information in the geometric topology architecture can be deeply analyzed. On account of that point cloud data is sparse, disordered and rotation-invariant, the success of convolutional neural network in 2 D image cannot be directly reproduced on point cloud. In this paper, we propose WECNN, namely, Weight-Edge Convolution Neural Network, which has an excellent ability to utilize local structural features. As the core of WECNN, a novel convolution operator called WEConv tries to capture structural features by constructing a fixed number of directed graphs and extracting the edge information of the graph to further analyze the local regions of point cloud. Moreover, a weight function is designed for different tasks to assign weights to the edges, so that feature extractions on the edges can be more fine-grained and robust. WECNN gets overall accuracy of 93.8% and mean class accuracy of 91.6% on Model Net40 dataset. At the same time, it gets a mean Io U of 85.5% on Shape Net Part dataset. Results of extensive experiments show that our WECNN outperforms other classification and segmentation approaches on challenging benchmarks.     

用于焊缝检测的X射线实时图像序列时空域滤波

X射线实时图像处理技术是在线自动检测焊缝缺陷的关键环节。由于原始图像噪声大,严重影响了焊缝缺陷X射线实时自动检测技术的实际应用。该文先对焊缝X射线实时图像序列进行匹配,在此基础上采用扩展的局部线性最小均方差法进行时空域滤波,从而综合利用X射线实时图像序列的时域信息和空域信息。实验结果表明:焊缝X射线实时图像时空域滤波效果明显优于现有的空域滤波和时域滤波,为后续焊缝缺陷检测的图像处理提供了良好条件。