Lens binarity vs limb darkening in close-impact galactic microlensing events

合集下载

Learning to detect natural image boundaries using local brightness, color, and texture cues

David R.Martin Charless C.Fowlkes Jitendra MalikComputer Science Division,EECS,U.C.Berkeley,Berkeley,CA94720dmartin,fowlkes,malik@AbstractThe goal of this work is to accurately detect and localize boundaries innatural scenes using local image measurements.We formulate featuresthat respond to characteristic changes in brightness and texture associatedwith natural boundaries.In order to combine the information from thesefeatures in an optimal way,a classiﬁer is trained using human labeledimages as ground truth.We present precision-recall curves showing thatthe resulting detector outperforms existing approaches.1IntroductionConsider the image patches in Figure1.Though they lack global context,it is clear which contain boundaries and which do not.The goal of this paper is to use features extracted from the image patch to estimate the posterior probability of a boundary passing through the center point.Such a local boundary model is integral to higher-level segmentation algo-rithms,whether based on grouping pixels into regions[21,8]or grouping edge fragments into contours[22,16].The traditional approach to this problem is to look for discontinuities in image brightness. For example,the widely employed Canny detector[2]models boundaries as brightness step edges.The image patches show that this is an inadequate model for boundaries in natural images,due to the ubiquitous phenomenon of texture.The Canny detector willﬁre wildly inside textured regions where high-contrast contours are present but no boundary exists.In addition,it is unable to detect the boundary between textured regions when there is only a subtle change in average image brightness.These signiﬁcant problems have lead researchers to develop boundary detectors that ex-plicitly model texture.While these work well on synthetic Brodatz mosaics,they have problems in the vicinity of brightness edges.Texture descriptors over local windows that straddle a boundary have different statistics from windows contained in either of the neigh-boring regions.This results in thin halo-like regions being detected around contours. Clearly,boundaries in natural images are marked by changes in both texture and brightness. Evidence from psychophysics[18]suggests that humans make combined use of these two cues to improve detection and localization of boundaries.There has been limited work in computational vision on addressing the difﬁcult problem of cue combination.For example, the authors of[8]associate a measure of texturedness with each point in an image in order to suppress contour processing in textured regions and vice versa.However,their solution is full of ad-hoc design decisions and hand chosen parameters.The main contribution of this paper is to provide a more principled approach to cue com-bination by framing the task as a supervised learning problem.A large dataset of natural images that have been manually segmented by multiple human subjects[10]provides theground truth label for each pixel as being on-or off-boundary.The task is then to model the probability of a pixel being on-boundary conditioned on some set of locally measured image features.This sort of quantitative approach to learning and evaluating boundary detectors is similar to the work of Konishi et al.[7]using the Sowerby dataset of En-glish countryside scenes.Our work is distinguished by an explicit treatment of texture and brightness,enabling superior performance on a more diverse collection of natural images. The outline of the paper is as follows.In Section2we describe the oriented energy and texture gradient features used as input to our algorithm.Section3discusses the classiﬁers we use to combine the local features.Section4presents our evaluation methodology along with a quantitative comparison of our method to existing boundary detection methods.We conclude in Section5.2Image Features2.1Oriented EnergyIn natural images,brightness edges are more than simple steps.Phenomena such as spec-ularities,mutual illumination,and shading result in composite intensity proﬁles consisting of steps,peaks,and roofs.The oriented energy(OE)approach[12]can be used to detect and localize these composite edges[14].OE is deﬁned as:where and are a quadrature pair of even-and odd-symmetricﬁlters at orientation and scale.Our even-symmetricﬁlter is a Gaussian second-derivative,and the corre-sponding odd-symmetricﬁlter is its Hilbert transform.has maximum response for contours at orientation.We compute OE at3half-octave scales starting atthe image diagonal.Theﬁlters are elongated by a ratio of3:1along the putative boundary direction.2.2Texture GradientWe would like a directional operator that measures the degree to which texture varies at a location in direction.A natural way to operationalize this is to consider a disk of radius centered on,and divided in two along a diameter at orientation.We can then compare the texture in the two half discs with some texture dissimilarity measure. Oriented texture processing along these lines has been pursued by[19].What texture dissimilarity measure should one use?There is an emerging consensus that for texture analysis,an image shouldﬁrst be convolved with a bank ofﬁlters tuned to various orientations and spatial frequencies[4,9].Afterﬁltering,a texture descriptor is then constructed using the empirical distribution ofﬁlter responses in the neighborhood of a pixel.This approach has been shown to be very powerful both for texture synthesis[5]as well as texture discrimination[15].Puzicha et al.[15]evaluate a wide range of texture descriptors in this framework.We choose the approach developed in[8].Convolution with aﬁlter bank containing both even and oddﬁlters at multiple orientations as well as a radially symmetric center-surround ﬁlter associates a vector ofﬁlter responses to every pixel.These vectors are clustered using k-means and each pixel is assigned to one of the cluster centers,or textons.Texture dissimilarities can then be computed by comparing the histograms of textons in the two disc halves.Let and count how many pixels of texton type occur in each half disk.Image IntensityN o n -B o u n d a r i e sB o u n d a r i e sFigure 1:Local image features.In each row,the ﬁrst panel shows the image patch.The followingpanels show feature proﬁles along the line marked in each patch.The features are raw image intensity,raw oriented energy ,localized oriented energy ,raw texture gradient ,and localized texture gradient .The vertical line in each proﬁle marks the patch center.The challenge is to combine these features in order to detect and localize boundaries.We deﬁne the texture gradient (TG)to be the distance between these two histograms:The texture gradient is computed at each pixelover 12orientations and 3half-octavescales starting at of the image diagonal.2.3LocalizationThe underlying function we are trying to learn is tightly peaked around the location of image boundaries marked by humans.In contrast,Figure 1shows that the features we have discussed so far don’t have this structure.By nature of the fact that they pool information over some support,they produce smooth,spatially extended outputs.The texture gradient is particularly prone to this effect,since the texture in a window straddling the boundary is distinctly different than the textures on either side of the boundary.This often results in a wide plateau or even double peaks in the texture gradient.Since each pixel is classiﬁed independently,these spatially extended features are partic-ularly problematic as both on-boundary pixels and nearby off-boundary pixels will have large OE and TG.In order to make this spatial structure available to the classiﬁer we trans-form the raw OE and TG signals in order to emphasize local maxima.Given a featuredeﬁned over spatial coordinate orthogonal to the edge orientation,consider the derived feature,where is theﬁrst-order approximation of the distance to the nearest maximum of.We use the stabilized version1Windowed parabolicﬁtting is known as2nd-order Savitsky-Golayﬁltering.We also considered Gaussian derivativeﬁlters to estimate with nearly identical results.2Theﬁtted values are=0.1,0.075,0.013and=2.1,2.5,3.1for OE,and=.057,.016,.005 and=6.66,9.31,11.72for TG.is measured in pixels.00.250.50.75100.250.50.751P r e c i s i o nRecall Raw Featuresall F=.65oe0 F=.59oe1 F=.60oe2 F=.61tg0 F=.64tg1 F=.64tg2 F=.610.250.50.75100.250.50.751P r e c i s i o nRecall Localized Featuresall F=.67oe0 F=.60oe1 F=.62oe2 F=.63tg0 F=.65tg1 F=.65tg2 F=.63Figure 2:Performance of raw (left)and localized features (right).The precision and recall axes aredescribed in Section 4.Curves towards the top (lower noise)and right (higher accuracy)are more desirable.Each curve is scored by the F-measure,the value of which is shown in the legend.In all the precision-recall graphs in this paper,the maximum F-measure occurs at a recall of approximately 75%.The left plot shows the performance of the raw OE and TG features using the logistic regression classiﬁer.The right plot shows the performance of the features after applying the localization process of Equation 1.It is clear that the localization function greatly improves the quality of the individual features,especially the texture gradient.The top curve in each graph shows the performance of the features in combination.While tuning each feature’sparameters individually is suboptimal,overall performance still improves.consider small binary trees up to a depth of 3(8experts).The model is initialized in a greedy,top-down manner and ﬁt with EM.Support Vector Machines We use the SVM package libsvm [3]to do soft-margin clas-siﬁcation using Gaussian kernels.The optimal parameters were =0.2and =0.2.The ground truth boundary data is based on the dataset of [10]which provides 5-6human segmentations for each of 1000natural images from the Corel image database.We used 200images for training and algorithm development.The 100test images were used only to generate the ﬁnal results for this paper.The authors of [10]show that the segmentations of a single image by the different subjects are highly consistent,so we consider all human-marked boundaries valid.We declare an image location to be on-boundary if it is within =2pixels and =30degrees of any human-marked boundary.The remainder are labeled off-boundary.This classiﬁcation task is characterized by relatively low dimension,a large amount of data (100M samples for our 240x160-pixel images),and poor separability.The maximum fea-sible amount of data,uniformly sampled,is given to each classiﬁer.This varies from 50M samples for density estimation to 20K samples for the SVM.Note that a high degree of class overlap in any local feature space is inevitable because the human subjects make use of both global constraints and high-level information to resolve locally ambiguous bound-aries.4ResultsThe output of each classiﬁer is a set of oriented images ,which provide the probability of a boundary at each image location based on local information.For several of the00.250.50.75100.250.50.751P r e c i s i o nRecall (a) Feature Combinationsall F=.67oe2+tg1 F=.67tg* F=.66oe* F=.6300.250.50.75100.250.50.751P r e c i s i o n Recall (b) ClassifiersDensity Estimation F=.68Classification Tree F=.68Logistic Regression F=.67Quadratic LR F=.68Boosted LR F=.68Hier. Mix. of Experts F=.68Support Vector Machine F=.66Figure 3:Precision-recall curves for (a)different feature combinations,and (b)different classiﬁers.The left panel shows the performance of different combinations of the localized features using the logistic regression classiﬁer:the 3OE features (oe*),the 3TG features (tg*),the best performing single OE and TG features (oe2+tg1),and all 6features together.There is clearly independent infor-mation in each feature,but most of the information is captured by the combination of one OE and one TG feature.The right panel shows the performance of different classiﬁers using all 6features.All the classiﬁers achieve similar performance,except for the SVM which suffers from the poor sep-aration of the data.Classiﬁcation trees performs the best by a slim margin.Based on performance,simplicity,and low computation cost,we favor the logistic regression and its variants.classiﬁers we consider,the image provides actual posterior probabilities,which is par-ticularly appropriate for the local measurement model in higher-level vision applications.For the purpose of evaluation,we take the maximum over orientations.In order to evaluate the boundary model against the human ground truth,we use the precision-recall framework,a standard evaluation technique in the information retrieval community [17].It is closely related to the ROC curves used for by [1]to evaluate bound-ary models.The precision-recall curve captures the trade-off between accuracy and noise as the detector threshold is varied.Precision is the fraction of detections which are true pos-itives,while recall is the fraction of positives that are detected.These are computed using a distance tolerance of 2pixels to allow for small localization errors in both the machine and human boundary maps.The precision-recall curve is particularly meaningful in the context of boundary detection when we consider applications that make use of boundary maps,such as stereo or object recognition.It is reasonable to characterize higher level processing in terms of how much true signal is required to succeed,and how much noise can be tolerated.Recall provides the former and precision the latter.A particular application will deﬁne a relative cost between these quantities,which focuses attention at a speciﬁc point on the precision-recall curve.The F-measure ,deﬁned as ,captures this trade-off.The location of the maximum F-measure along the curve provides the optimal threshold given ,which we set to 0.5in our experiments.Figure 2shows the performance of the raw and localized features.This provides a clear quantitative justiﬁcation for the localization process described in Section 2.3.Figure 3a shows the performance of various linear combinations of the localized features.The com-bination of multiple scales improves performance,but the largest gain comes from using OE and TG together.00.250.50.75100.250.50.751P r e c i s i o n Recall (a) Detector ComparisonHuman F=.75Us F=.67Nitzberg F=.65Canny F=.570.40.50.60.70.8F -M e a s u r e Tolerance (in pixels)(b) F-Measure vs. Tolerance Figure 4:The left panel shows precision-recall curves for a variety of boundary detection schemes,along with the precision and recall of the human segmentations when compared with each other.The right panel shows the F-measure of each detector as the distance tolerance for measuring precision and recall varies.We take the Canny detector as the baseline due to its widespread use.Our detector outperforms the learning-based Nitzberg detector proposed by Konishi et al.[7],but there is still a signiﬁcant gap with respect to human performance.The results presented so far use the logistic regression classiﬁer.Figure 3b shows the per-formance of the 7different classiﬁers on the complete feature set.The most obvious trend is that they all perform similarly.The simple non-parametric models –the classiﬁcation tree and density estimation –perform the best,as they are most able to make use of the large quantity of training data to provide unbiased estimates of the posterior.The plain logistic regression model performs extremely well,with the variants of logistic regression –quadratic,boosted,and HME –performing only slightly better.The SVM is a disap-pointment because of its lower performance,high computational cost,and fragility.These problems result from the non-separability of the data,which requires 20%of the training examples to be used as support vectors.Balancing considerations of performance,model complexity,and computational cost,we favor the logistic model and its variants.3Figure 4shows the performance of our detector compared to two other approaches.Be-cause of its widespread use,MATLAB’s implementation of the classic Canny [2]detector forms the baseline.We also consider the Nitzberg detector [13,7],since it is based on a similar supervised learning approach,and Konishi et al.[7]show that it outperforms pre-vious methods.To make the comparisons fair,the parameters of both Canny and Nitzberg were optimized using the training data.For Canny,this amounts to choosing the optimal scale.The Nitzberg detector generates a feature vector containing eigenvalues of the 2nd moment matrix;we train a classiﬁer on these 2features using logistic regression.Figure 4also shows the performance of the human data as an upper-bound for the algo-rithms.The human precision-recall points are computed for each segmentation by com-paring it to the other segmentations of the same image.The approach of this paper is a clear improvement over the state of the art in boundary detection,but it will take the addi-tion of high-level and global information to close the gap between the machine and human performance.5ConclusionWe have deﬁned a novel set of brightness and texture cues appropriate for constructing a local boundary model.By using a very large dataset of human-labeled boundaries in natural images,we have formulated the task of cue combination for local boundary detection as a supervised learning problem.This approach models the true posterior probability of a boundary at every image location and orientation,which is particularly useful for higher-level algorithms.Based on a quantitative evaluation on100natural images,our detector outperforms existing methods.References[1]K.Bowyer,C.Kranenburg,and S.Dougherty.Edge detector evaluation using empirical ROCcurves.Proc.IEEE put.Vision and Pattern Recognition,1999.[2]J.Canny.A computational approach to edge detection.IEEE Trans.Pattern Analysis andMachine Intelligence,8:679–698,1986.[3] C.Chang and C.Lin.LIBSVM:a library for support vector machines,2001.Software availableat .tw/˜cjlin/libsvm.[4]I.Fogel and D.Sagi.Gaborﬁlters as texture discriminator.Bio.Cybernetics,61:103–13,1989.[5] D.J.Heeger and J.R.Bergen.Pyramid-based texture analysis/synthesis.In Proceedings ofSIGGRAPH’95,pages229–238,1995.[6]M.I.Jordan and R.A.Jacobs.Hierarchical mixtures of experts and the EM algorithm.NeuralComputation,6:181–214,1994.[7]S.Konishi,A.L.Yuille,J.Coughlan,and S.C.Zhu.Fundamental bounds on edge detection:an information theoretic evaluation of different edge cues.Proc.IEEE put.Vision and Pattern Recognition,pages573–579,1999.[8]J.Malik,S.Belongie,T.Leung,and J.Shi.Contour and texture analysis for image segmenta-tion.Int’l.Journal of Computer Vision,43(1):7–27,June2001.[9]J.Malik and P.Perona.Preattentive texture discrimination with early vision mechanisms.J.Optical Society of America,7(2):923–32,May1990.[10] D.Martin,C.Fowlkes,D.Tal,and J.Malik.A database of human segmented natural imagesand its application to evaluating segmentation algorithms and measuring ecological statistics.In Proc.8th Int’puter Vision,volume2,pages416–423,July2001.[11]M.Meil˘a and J.Shi.Learning segmentation by random walks.In NIPS,2001.[12]M.C.Morrone and D.C.Burr.Feature detection in human vision:a phase dependent energymodel.Proc.R.Soc.Lond.B,235:221–45,1988.[13]M.Nitzberg,D.Mumford,and T.Shiota.Filtering,Segmentation and Depth.Springer-Verlag,1993.[14]P.Perona and J.Malik.Detecting and localizing edges composed of steps,peaks and roofs.Inputer Vision,pages52–7,Osaka,Japan,Dec1990.[15]J.Puzicha,T.Hofmann,and J.Buhmann.Non-parametric similarity measures for unsupervisedtexture segmentation and image retrieval.In Computer Vision and Pattern Recognition,1997.[16]X.Ren and J.Malik.A probabilistic multi-scale model for contour completion based on imagestatistics.Proc.7th put.Vision,2002.[17] C.Van rmation Retrieval,2nd ed.Dept.of Comp.Sci.,Univ.of Glasgow,1979.[18]J.Rivest and P.Cavanagh.Localizing contours deﬁned by more than one attribute.VisionResearch,36(1):53–66,1996.[19]Y.Rubner and C.Tomasi.Coalescing texture descriptors.ARPA Image Understanding Work-shop,1996.[20]R.E.Schapire and Y.Singer.Improved boosting algorithms using conﬁdence-rated predictions.Machine Learning,37(3):297–336,1999.[21]Z.Tu,S.Zhu,and H.Shum.Image segmentation by data driven markov chain monte carlo.InProc.8th Int’puter Vision,volume2,pages131–138,July2001.[22]L.R.Williams and D.W.Jacobs.Stochastic completionﬁelds:a neural model of illusory contourshape and salience.In Proc.5th puter Vision,pages408–15,June1995.。

基于轮廓波维纳滤波的图像压缩传感重构.

第30卷第10期仪器仪表学报V ol.30 No. 10 2009年10月Chinese Journal of Scientific Instrument Oct. 2009 基于轮廓波维纳滤波的图像压缩传感重构*李林，孔令富，练秋生（燕山大学信息科学与工程学院秦皇岛066004）摘要：图像压缩传感重构利用自然图像可稀疏表示的先验知识，从比奈奎斯特采样率低得多的随机投影观测值中重构原始图像。

为了克服传统的压缩传感重构中正交小波方向选择性差和未利用变换系数的邻域统计特性的缺点，利用了轮廓波维纳滤波去噪算子替代迭代阈值法中的阈值算子，进而提出了基于轮廓波维纳滤波的图像压缩传感的重构算法。

实验结果表明，该算法提高了重构图像的峰值信噪比和视觉效果，保护了图像的细节，加快了重构算法的收敛速度。

关键词：压缩传感；轮廓波；迭代维纳滤波；邻域系数中图分类号：TP391文献标识码：A国家标准学科分类代码：510.4050Image compressed sensing reconstruction based on contourlet Wiener filteringLi Lin, Kong Lingfu, Lian Qiusheng（Institute of Information Science and Technology,Yanshan University, Qinhuangdao 066004, China）Abstract：Image compressed sensing reconstruction uses the sparse prior of the natural image and can reconstruct the original image from far fewer measurements of random projection than Nyquist sampling rate. To overcome the poor directional selectivity of orthogonal wavelet bases and not exploiting the statistic characteristics of trans-form coefficients in neighborhood, the contourlet Wiener filtering denoising operator is used to replace the thresh-old operator in iterative thresholding algorithm. And then a reconstruction algorithm is proposed, which combines contourlet with Wiener filtering. Experiment results show that the proposed algorithm improves the peak sig-nal-to-noise ratio and visual quality, protects image details, and accelerates the convergence of reconstruction al-gorithm.Key words：compressed sensing; contourlet; iterative Wiener filtering; neighborhood coefficient1引言随着科技进步，信号带宽越来越宽，并在某些场合是不可知的，传统的香农抽样定理在应用中受到了挑战。

基于邻域相似性的暗原色先验图像去雾方法

第３１卷第５期
２１年５月０１
计算机应用
ＪｕｎｌｏｏｕｅｐｉａｉｎｏｒａｆＣｍｐｔｒＡｐｌｔｓｃｏ
Ｖ１３．ｏ．１Ｎｏ５Ｍａ０１ｙ２１
文章编号：０１— ０１２１）５— １２０１０９８（０１００２４— ３
ｍｄ１ｈｘｅｉｅｔｓｌｈｗｔａｔｅｍｔｏａａｐｔｅｅｇｄｉｐｏｅｈａｔｏｅｄｇｄｄｉａｅｏｅ．Ｔｅｅｐｒｎｒｕｔｓｏｔｈｅｄｃｎｓｒｄｅａｒｖｅｑｌｙｆｅａｅｇ．ｍａｅｓｌｈｈｈｈｎｍｔｕｉｔｈｒｍ
ＧｉＵＯＪａ，ＷＡＮＧＸｉｏｔｎＨＵＣｅｇｐｎａ．ｇ一，ｏｈｎ。ｅｇ，ＸＵＸｉｏｇｎａ．ａｇ
（．ＩｓｔｔｏｈｔｌｔｃＴｃｎｌｙａｉｎＮｖｌｃｄｍ，ＤｌｎＬａｎｇ１６１，Ｃｉ；１ｎｔｕｉｅｆＰｏｅｃｉｅｏｇ，Ｄｌａａａｅｙａｉｉｏｉ１０８ｈｎｏｅｒｈｏａＡａｎａ
－
２ｅａｔｅｔａｉｔｎａｉａａＡａｍ，ＤｌｎＬａｎｎ１０８ｈｎ；．ＤｐｒｎｏＮｖａｉ，ＤｌｎＮｖｌｃｄｙａｉｉｏｉ１６１，Ｃｉｍｆｇｏａｅａｇａ３ｅａｍｅｔｆＥ＠￣ｎｕｏａｉｔｎａｉａａＡａｅｙ．ＤｐｒｎｑｍｔＡｔｍｔａｉ，ＤｌｎＮｖｌｃｄｍ，Ｄａｉｉｏｉ１０８ｈｎ）ｔｏｅｚｏａｌｎＬａｎｎ１６１，Ｃｉａｇａ

(光电信息工程专业英语)专业英语第七讲Geometrical Optics

牛顿1672年使用的6英寸反射式望远镜复制品，为皇家学会所拥有。
1994年，不列颠哥伦比亚大学开始建造一台口径为6米的旋转水银面望远镜——大型天顶望远镜（LZT），并于2003年建成
1.9 Lens Aberrations
9. By choosing materials with indices of refraction that depend in different ways on the color of the light, lens designers can ensure that positive and negative lens elements compensate for each other by having equal but opposite chromatic effects.
1.9 Lens Aberrations
4. Up to now, our discussion of lenses has not taken into account some of the optical imperfections that are inherent in single lenses made of uniform material with spherical surfaces. These failures of a lens to give a perfect image are known as aberrations. 句子结构： our discussion … has not taken into account … imperfections…
vt.提供；给予（provide的过去式）
参考翻译：
在上述的推导中，我们检验了物在会聚透镜焦点以外的特殊情况。然而，如果遵守下述的符号规则，薄透镜公式对于会聚和发散薄透镜都是有效的，且不论物体远近。

超像素分割算法研究综述

超像素分割算法研究综述超像素分割是计算机视觉领域的一个重要研究方向，它的目标是将图像分割成一组紧密连接的区域，每个区域都具有相似的颜色和纹理特征。

超像素分割可以在许多计算机视觉任务中发挥重要作用，如图像分割、目标检测和图像语义分割等。

本综述将介绍一些常见的超像素分割算法及其应用。

1. SLIC (Simple Linear Iterative Clustering)：SLIC是一种基于k-means聚类的超像素分割算法。

它首先在图像中均匀采样一组初始超像素中心，并通过迭代的方式将每个像素分配给最近的超像素中心。

SLIC算法结合了颜色和空间信息，具有简单高效的特点，适用于实时应用。

2.QuickShift：QuickShift是一种基于密度峰值的超像素分割算法。

它通过利用图片的颜色相似性和空间相似性来计算每个像素的相似度，并通过移动像素之间的边界来形成超像素。

QuickShift算法不依赖于预定义的超像素数量，适用于不同大小和形状的图像。

3. CPMC (Constrained Parametric Min-Cuts)：CPMC是一种基于图割的超像素分割算法。

该算法通过求解最小割问题来获得具有边界连通性的超像素分割结果。

CPMC算法能够生成形状规则的超像素，适用于对形状准确性要求较高的应用。

4. LSC (Linear Spectral Clustering)：LSC是一种基于线性谱聚类的超像素分割算法。

它通过构建图像的颜色和空间邻接图，并对其进行谱分解来获取超像素分割结果。

LSC算法具有良好的分割结果和计算效率，适用于大规模图像数据的处理。

5. SEEDS (Superpixels Extracted via Energy-Driven Sampling)：SEEDS是一种基于随机采样的超像素分割算法。

它通过迭代的方式将像素相似度转化为能量函数，并通过最小化能量函数来生成超像素。

SEEDS算法能够快速生成具有边界连通性的超像素，并适用于实时应用。

基于L-BFGS-B局部极小化的自适应尺度CLEAN算法

第38卷第1期2021年1月Vol, 33 No. 1Ju. 0021贵州大学学报(自然科学版)Journai of Guizhou UniversiW ( Naturai Sciecces)文章编号 10002269(2021)012038 27 DOI ：10.15759/(. chdi. adxPzrb. 0021.01.06基于L-BFGS-B 局部极小化的自适应尺度CLEAN 算法张利**6,肖一凡6米立功6卢梅6赵庆超6王 b 1,刘祥4,张明4,谢泉1 (6贵州大学大数据与信息工程学院，贵州贵阳550028 ；2,中国科学院新疆天文台，新疆乌鲁木齐830011)收稿日期:202021-23基金项目：平方公里阵列射电望远镜专项资助项目(2020SKA0n0300);国家重点研发计划资助项目(2418YFA0404242)；国家自然科学基金资助项目(11663403);贵州省教育厅青年科技人才成长资助项目(黔教合KY 字［2018 ］ 119)作者简介：张利(1987—)，男，特聘教授，博士，研究方向：天文大数据、医疗大数据、图像处理,E-mi ：/zhang. scmoce@OmaW com.* 通讯作者：张利,E-maii : /zhang. science@ gmaii. com.摘要：干涉阵列存在的i 扩展函数旁瓣使得观测到的射电源出现不同程度的失真，对重建宇宙真实结构图景和理解宇宙起源造成影响。

为解决观测中出现的伪影，本文在现有的CLEAN 算法的基础上，提出了基于L-2FGS-2局部极小化的自适应尺度CLEAN 算法。

首先，基于L 西FGS- B 局部极小化算法通过最小化目标函数，寻找最优分量，构建自适应尺度模型；其次，通过CASA实现对测试图像的重建,对比目前广泛使用的HBgdom CLEAN 算法的重建图像，评估本文算法性能；最后，对反卷积算法在射电天文图像处理领域的发展做出展望。

细胞超分辨成像技术的最新进展

细胞超分辨成像技术的最新进展细胞是构成生物体的基本单位，其内部结构鲜有人能够观测到。

传统的光学显微镜像素级别的分辨率限制了观察细胞内部微观结构的深度。

探测到分子级别的微观细节对理解细胞的生理和疾病过程至关重要，而超分辨成像技术为解决这个问题提供了切实可行的方法。

传统的液体显微镜的行程始于 1889 年，当时德国物理学家夫琅和费发展了同步位移的干涉显微术。

20世纪60年代，Minsky提出了激光点扫描显微镜（LSM）的方法。

这种方法采用荧光标记的生物样品，LED或激光扫描围绕着一个样品的轨迹，由发射到检测器的光产生的图像，从而得到一个三维点扫描图像的序列。

然而，这种方法的分辨率仍远远不能突破经典阿贝分辨率极限。

在过去的二十年中，超分辨成像技术已经得到了相当大的进步，并广泛应用于各个学科领域。

总体上，超分辨成像技术的原理有两种：超分辨现场（SIM）和刺激发射衰减（STED）显微镜。

超分辨现场显微镜利用多种景深渐变技术，超分辨现场显微镜将景深分为称为“焦点”的多个体积，继而用多个重叠图像来重建最终超分辨率图像。

而 STED 利用激光点阵、相位板和偏振调节器，通过对较小的局部区域进行刺激发射衰减，根据表面下的荧光谐振能量传递强点测量荧光，得到超分辨率成像的效果。

值得一提的是，STED 显微镜技术实现的分辨率甚至可以小于10纳米。

近年来，一种被称为单分子定位瞬态成像（PALM) 的超分辨率显微镜也引起了广泛关注。

此技术基于光刻制造微型结构，通过控制荧光标识剂的激发，进而控制陆续点亮像素的显示。

然后，通过这些对局部成像进行重复图像叠加的方式，显微镜利用那些位置信息不断瞬移的单个分子的图像信息获得更高的空间分辨率。

据报道，使用 PALM 测量的细胞器内蛋白质的可视化最高分辨率可以达到10％。

超分辨成像技术促进了生命科学研究领域的多个方面，包括细胞吞噬，肿瘤学，神经科学等。

它使研究者们能够观察和分析某些非常微小的细胞结构以及计量分子分布。

达到光学分辨率极限的“最清晰”图像问世

达到光学分辨率极限的“最清晰”图像问世
佚名
【期刊名称】《创新时代》
【年(卷),期】2012(000)009
【摘要】英国《自然·纳米技术》杂志8月12日在线刊登报告说，新加坡研究人员完成了一幅常用作图像测试的彩色女子头像“莱娜”，整幅头像大小只有50微米见方，它的清晰程度可达到光学分辨率的理论极限，即化每250纳米距离上安放一个像素点。

【总页数】1页(P15-15)
【正文语种】中文
【中图分类】TP334.22
【相关文献】
1.雾天车辆超分辨率视频图像清晰度识别仿真 [J], 汤嘉立;杜卓明
2.基于块旋转和清晰度的图像超分辨率重建算法 [J], 尧潞阳;解凯;李桐
3.达光学分辨率极限“最清晰”图像问世 [J], W．KJ
4.每250nm距离安放一个像素点达光学分辨力极限的“最清晰”图像问世 [J],
5.达到理论极限的带光学读出的非致冷焦平面列阵 [J], 高国龙
因版权原因，仅展示原文概要，查看原文内容请购买。

复杂光照下的人脸肤色检测方法

复杂光照下的人脸肤色检测方法
李全彬;王小明;刘锦高;李明
【期刊名称】《计算机应用》
【年(卷),期】2010(0)6
【摘要】复杂光照对人脸肤色检测具有重要影响.在YCbCr颜色空间建立复杂光照条件下的人脸肤色模型,然后利用该模型检测人脸图像的肤色区域,并对检测结果利用4-连通区域的几何特征消除非人脸区域,最后利用连通元复原误检的人脸肤色区域.实验结果表明,该方法可以实现复杂光照下人脸肤色区域的准确检测.
【总页数】3页(P1594-1596)
【作者】李全彬;王小明;刘锦高;李明
【作者单位】华东师范大学,信息科学技术学院,上海,200062;徐州师范大学,物理与电子工程学院,江苏,徐州,221116;华东师范大学,信息科学技术学院,上海,200062;华东师范大学,信息科学技术学院,上海,200062;华东师范大学,信息科学技术学院,上海,200062
【正文语种】中文
【中图分类】TP391
【相关文献】
1.一种基于肤色的复杂背景人脸检测方法 [J], 何为;李见为;蒋邦持
2.较强光照下肤色结合发色检测人脸的方法研究 [J], 黄亦佳;潘巍
3.一种融合改进型AdaBoost和肤色模型的人脸检测方法 [J], 姚子怡;张清勇;李雪
琪
4.基于双肤色模型和改进的SNoW算法的人脸检测方法 [J], 丁勇;吴彦儒
5.基于肤色特征的人脸检测方法研究 [J], 唐云龙;何麒;陈平
因版权原因，仅展示原文概要，查看原文内容请购买。

致我们终将朴素的近邻

致我们终将朴素的近邻
1980
【期刊名称】《中国眼镜科技杂志》
【年(卷),期】2014(0)18
【摘要】以前,我们就说过,如果作为日本眼镜的狂热分子,当然会对那些个性十足的眼镜产品歇斯底里,然后努力寻找那些憧憬已久的限量新品。

但是,等到它们在国内上柜时,你也许还是会有些失望：我喜欢的那款＂OKI-NI未完成款＂在哪里？SOLID BLUE又在哪里？还有我中意的LESS THAN HUMAN,它难道只在秀场或者是海报中才活着？
【总页数】10页(P26-35)
【作者】1980
【作者单位】
【正文语种】中文
【相关文献】
1.基于互近邻一致性的近邻传播算法
2.近邻点一致性的随机抽样一致性算法
3.基于K-近邻法的局部加权朴素贝叶斯分类算法
4.基于K-近邻法的局部加权朴素贝叶斯分类算法
5.基于深K近邻和朴素贝叶斯分类算法的肿瘤诊断
因版权原因，仅展示原文概要，查看原文内容请购买。

美国警察采用帕拉蓬纳米实验室的DNA人脸技术破案

美国警察采用帕拉蓬纳米实验室的DNA人脸技术破案
佚名
【期刊名称】《警察技术》
【年(卷),期】2016(0)3
【摘要】美国弗吉尼亚州的帕拉蓬纳米实验室（Parabon Nanolabs）研发了一种能够基于基因序列来预测人脸外观的新技术，该技术称为“人脸法医系统”。

帕拉蓬生物信息部主任艾伦·麦克雷·格雷塔克告诉《大众科学》，
【总页数】1页(P96-96)
【关键词】新技术;实验室;人脸;纳米;美国;DNA;破案;警察
【正文语种】中文
【中图分类】Q523
【相关文献】
1.细胞DNA拓扑异构酶Ⅱ抑制剂吡喃萘醌诱导抗肿瘤药α-拉帕醌和β-拉帕醌的新机制 [J],
2.从美国FDA批准新药看生物制品和化学药在临床审评方面考量的异同——以HER2阳性转移性乳腺癌药品帕妥珠单抗及拉帕替尼为例 [J], 董可;李大蔚;姚立新;郑强
3.横滨橡胶推出采用纳米技术的“DNA S．drive” [J],
4.美国女素描专家绘肖像帮警察破案 [J],
5.美国梦的追寻——对《塞拉斯·拉帕姆的发迹》中的塞拉斯·拉帕姆与《了不起的盖茨比》中的盖茨比的比较分析 [J], 刘琪
因版权原因，仅展示原文概要，查看原文内容请购买。

利用颜色编码辅助计算绝对相位的傅里叶轮廓术

利用颜色编码辅助计算绝对相位的傅里叶轮廓术
鲁超;李永新
【期刊名称】《应用光学》
【年(卷),期】2013(34)5
【摘要】针对传统去卷积算法时间需求的弊端,提出一种新的使用颜色编码辅助的绝对相位并行计算方法.该算法采用对光栅数目需求最少的傅里叶变换轮廓术(FTP)做为卷积相位求取的方法;颜色编码光栅被用来标识轮廓的序数.直接使用FTP计算出的卷积相位以及从彩色光栅中获得的轮廓序数,即可方便求出当前像素的绝对相位值;同时只用一副图像标识轮廓序数也比其他轮廓序数标识方法简单.本方法由于使用绝对相位计算方法,局部相位误差不会扩展.实验结果也证明了此算法对于多个分离物体以及复杂物体的有效性.
【总页数】6页(P831-836)
【作者】鲁超;李永新
【作者单位】中国科学技术大学精密机械与精密仪器系,安徽合肥230027;中国科学技术大学精密机械与精密仪器系,安徽合肥230027
【正文语种】中文
【中图分类】TN911.73;TN247
【相关文献】
1.基于傅里叶相位差的抗噪声位移估计算法 [J], 吴元昊;于前洋
2.利用辅助函数法进行傅里叶展开 [J], 李洪才
3.利用傅里叶积分性质计算信号频谱函数的补充说明 [J], 张国强;杜清珍
4.基于微分法的新傅里叶测量轮廓术(英文) [J], 赵焕东;李志能;毕岗
5.利用静止卫星云图进行二维傅里叶相位导风试验 [J], 张红;王振会;许建明
因版权原因，仅展示原文概要，查看原文内容请购买。

粒子群优化的深海图像暗边缘检测优化算法

粒子群优化的深海图像暗边缘检测优化算法
邹倩颖;陈晖阳;李永生;胡力雯;王小芳
【期刊名称】《应用科学学报》
【年(卷),期】2023(41)1
【摘要】为解决深海资源探测图像识别难题,提出一种基于粒子群优化的图像暗边缘检测优化算法。

该算法通过指数型线性单元和高斯误差线性单元改进激活函数,根据Marr-Hildreth算子检测结果并结合改进激活函数构建暗边缘检测算法,利用粒子群对改进暗边缘检测算法进行训练和优化。

最后,采用不同算法对水下11个数据集进行比较的结果表明:改进算法的峰值信噪比、结构相似度和边缘保持指数最高,分别达到18.7696 dB、0.6607和0.8345;图像均方误差最低,为3750.2253;平均检测时间为0.6674 s,比其他对比实验中性能最好的算法缩短了14%。

【总页数】17页(P153-169)
【作者】邹倩颖;陈晖阳;李永生;胡力雯;王小芳
【作者单位】吉利学院智能科技学院;电子科技大学成都学院行知学院
【正文语种】中文
【中图分类】TP391.41
【相关文献】
1.粒子群优化算法在图像边缘检测中的应用
2.分层差分粒子群优化算法对图像边缘检测的研究与应用
3.粒子群优化算法在图像边缘检测中的研究应用
4.基于Canny 边缘检测的图像预处理优化算法
5.基于粒子群优化算法的深海运载器模块化设计
因版权原因，仅展示原文概要，查看原文内容请购买。

基于有界高斯混合模型的高光谱图像去噪方法

基于有界高斯混合模型的高光谱图像去噪方法
刘昊;贾小宁;成丽波;李喆
【期刊名称】《应用数学进展》
【年(卷),期】2022(11)12
【摘要】针对高光谱图像的噪声去除问题,本文提出了一种基于有界高斯混合模型的高光谱图像去噪方法。

在该方法中,我们使用张量表示高光谱图像,并对其进行Tucker分解,最后采用有界高斯混合模型对噪声进行拟合,从而将图像的固有特征和噪声建模相结合。

我们将图像和噪声的先验信息表述为一个完整的贝叶斯模型,并设计了变分贝叶斯算法来封闭更新模型中所涉及的变量。

最后,我们将本文所提出的去噪算法与其他算法进行对比,验证了该方法的先进性。

【总页数】12页(P8473-8484)
【作者】刘昊;贾小宁;成丽波;李喆
【作者单位】长春理工大学数学与统计学院长春
【正文语种】中文
【中图分类】TP3
【相关文献】
1.基于小波域高斯混合模型与中值滤波的图像去噪研究
2.基于主成分分析和字典学习的高光谱遥感图像去噪方法
3.基于多尺度小波变换的高斯混合模型SAR图像去噪
4.基于RTF的高光谱图像去噪方法
因版权原因，仅展示原文概要，查看原文内容请购买。

快自旋小天体跨尺度光照不变匹配算法

快自旋小天体跨尺度光照不变匹配算法
李帅;李晋屹;刘延杰;邵巍;黄翔宇
【期刊名称】《深空探测学报（中英文）》
【年(卷),期】2024(11)1
【摘要】小天体探测器附着过程中,图像存在尺度、视角和光照变化,传统特征匹配算法难以获得精准匹配。

提出一种小天体跨尺度光照不变匹配算法。

针对附着过程中图像存在尺度变化的问题,将全局注意力机制与空洞卷积相结合,构建尺度自适应
调整模块;设计视角不变特征提取模块解决特征匹配算法大视角变化下匹配正确率
低的问题;结合自注意力机制与互注意力机制建立特征依赖关系,提取光照不变特征。

利用谷神星及贝努的真实图像分别进行了实验验证,结果表明,本文所提出算法在大
尺度、视角及光照变化下,特征匹配准确率达89%以上。

【总页数】7页(P56-62)
【作者】李帅;李晋屹;刘延杰;邵巍;黄翔宇
【作者单位】青岛科技大学自动化与电子工程学院;山东省深空自主着陆技术重点
实验室;北京控制工程研究所;空间智能控制技术全国重点实验室
【正文语种】中文
【中图分类】V448.2
【相关文献】
1.一种基于颜色尺度不变和FANN搜索的图像匹配算法
2.高效的光照、旋转、尺
度不变纹理分类算法3.融合彩色和光照信息的尺度不变特征变换图像匹配算法4.
基于尺度不变Harris特征的准稠密匹配算法5.基于支持向量机与尺度不变特征转换算法相结合的线状矢量数据匹配方法
因版权原因，仅展示原文概要，查看原文内容请购买。

基于小波的小尺寸物体的图像边缘提取方法

基于小波的小尺寸物体的图像边缘提取方法
李艳敏; 张恒; 李孟超
【期刊名称】《《微计算机信息》》
【年(卷),期】2004(20)12
【摘要】基于小波变换的“多分辨”特性,采用一种利用小波模极大值,找到图像中的边界点,从而实现对灰度图像的边界检测的方法.通过分析图像中是否有灰度突变特征,来判断边界的存在。

在进行边界检测之前,首先要滤除原图像的寄生噪声的干扰。

实验结果表明:基于小波变换的边界检测方法可对图像边界进行敏锐提取,从而得到小尺寸物体的图像的细微特征。

【总页数】2页(P126-127)
【作者】李艳敏; 张恒; 李孟超
【作者单位】上海理工大学
【正文语种】中文
【中图分类】TP391
【相关文献】
1.基于小波变换的手骨图像边缘提取方法 [J], 付秀丽;宁刚;廖小丽
2.一种基于小波变换的雷达图像边缘提取方法 [J], 刘佳敏;周荫清
3.基于B样条小波融合的平面足迹图像边缘特征提取方法研究 [J], 李元金;杨斌
4.基于不可分加性小波与形态学梯度的图像边缘提取方法 [J], 刘斌;孙斌;关淼苗;邢倩
5.一种改进的基于小波变换的图像边缘提取方法 [J], 李红
因版权原因，仅展示原文概要，查看原文内容请购买。

小波变换和硬C均值聚类的数字水印算法

小波变换和硬C均值聚类的数字水印算法
孟娜;王冰
【期刊名称】《计算机工程与应用》
【年(卷),期】2010(046)006
【摘要】提出一种基于硬C均值聚类的小波域数字图像水印算法.该算法利用人类视觉特性对小波域的低频子图进行硬C均值聚类,确定可嵌信息区域,在可嵌信息区域内通过交换相邻小波域系数值嵌入二值水印图像;双混沌序列的使用增强了水印系统的安全性.实验结果表明:该算法不仅具有良好的透明性,而且对叠加噪声、JPEG压缩、几何剪切,改变图像核心部分等恶意攻击具有高鲁棒性.
【总页数】5页(P197-200,220)
【作者】孟娜;王冰
【作者单位】西北大学,信息科学与技术学院,西安,710127;西北大学,信息科学与技术学院,西安,710127
【正文语种】中文
【中图分类】TP391
【相关文献】
1.利用小波变换和K均值聚类实现字幕区域分割 [J], 王建宇;张峰;周献中;史迎春;骆文
2.模糊C均值聚类与硬聚类的性能比较及改进 [J], 杨建
3.硬C-means聚类和DT-CWT变换的数字图像水印算法 [J], 林克正;姚欢
4.小波变换域均值调制的可读图像盲水印算法 [J], 郑成勇
5.结合小波变换和K均值聚类的SAR舰船目标分割 [J], 杨晓波
因版权原因，仅展示原文概要，查看原文内容请购买。

基于改进Canny和灰度矩的水下图像边缘检测方法研究

基于改进Canny和灰度矩的水下图像边缘检测方法研究陈振娅;刘增力
【期刊名称】《农业装备与车辆工程》
【年(卷),期】2024(62)3
【摘要】针对水下图像对比度差、模糊,致使检测出的边缘存在不连续和伪边缘的问题,提出一种基于改进的Canny和亚像素的水下图像边缘检测方法。

采用改进的Canny算法检测水下图像的像素级边缘,在此基础上采用灰度矩提取图像的亚像素边缘特征,提高边缘的定位精度和检测率。

实验结果表明,提出的算法相较于传统的像素级边缘检测算法在边缘轮廓的提取、减少伪边缘、提升边缘精度方面有较大优势,尤其是对深海环境中光源单一以及图像对比度差的物体边缘的检测具有更显著的效果和定位精度。

【总页数】5页(P107-110)
【作者】陈振娅;刘增力
【作者单位】昆明理工大学信息工程与自动化学院
【正文语种】中文
【中图分类】TP391.41
【相关文献】
1.基于改进的Canny算子和Zernike矩的亚像素边缘检测方法
2.基于 Canny 算子的图像边缘检测方法改进研究
3.基于改进Canny算子的磁共振T2加权图像边
缘检测方法研究4.基于改进Canny算法的图像边缘检测方法研究5.基于改进RGHS和Canny算子的水下图像边缘检测方法
因版权原因，仅展示原文概要，查看原文内容请购买。

基于贝叶斯算法的弱监督细粒度图像分类方法

基于贝叶斯算法的弱监督细粒度图像分类方法
崔西宁;孙红雨;李克龙
【期刊名称】《计算机仿真》
【年(卷),期】2022(39)9
【摘要】针对细粒度图像具有类内差距大、类间差距小的特点,而传统的图像分类方法存在分类效率低的问题,基于贝叶斯算法设计了新的弱监督细粒度图像分类方法。

先建立弱监督模型,并利用该模型定位图像的显著性细粒度区域,提取该区域中的细粒度图像特征并进行量化处理,并利用贝叶斯算法设计分类器,将提取的特征作为分类器的输入项,经过分类器的处理得出弱监督细粒度图像的分类结果。

对比实验结果表明,相比于传统方法,上述细粒度图像分类方法的分类精度有所提升,且有效的降低了图像分类所消耗的时间,从而提升了图像分类的整体效率。

【总页数】5页(P467-470)
【作者】崔西宁;孙红雨;李克龙
【作者单位】山东科技大学;西北师范大学
【正文语种】中文
【中图分类】TP391
【相关文献】
1.基于多分支神经网络模型的弱监督细粒度图像分类方法
2.基于注意力自身线性融合的弱监督细粒度图像分类算法
3.基于DDT算法和EfficientNet-B4的弱监督细
粒度农作物病害图像分类4.基于注意力机制的弱监督细粒度图像分类5.基于Xception网络的弱监督细粒度图像分类
因版权原因，仅展示原文概要，查看原文内容请购买。

基于肤色模型和梯度算子的正面人脸轮廓提取

基于肤色模型和梯度算子的正面人脸轮廓提取
吴绿芳;雷蕴奇;魏昇;吴众山
【期刊名称】《电脑与信息技术》
【年(卷),期】2008(016)001
【摘要】文章提出了一种新的正面人脸封闭轮廓提取方法,首先采用肤色模型对肤色区域进行初定位,然后经过筛选去除非人脸区域,最后提取人脸初步轮廓,并结合梯度算子提取下巴轮廓,进而得到一个连续、封闭的人脸轮廓.该算法基本克服了下巴轮廓难以从颈部提取的问题,满足了人脸轮廓特征提取的要求,同时具有较高的时效性.
【总页数】4页(P1-4)
【作者】吴绿芳;雷蕴奇;魏昇;吴众山
【作者单位】厦门大学信息科学与技术学院,福建,厦门,361005;厦门大学信息科学与技术学院,福建,厦门,361005;厦门大学信息科学与技术学院,福建,厦门,361005;厦门大学信息科学与技术学院,福建,厦门,361005
【正文语种】中文
【中图分类】TP391.41
【相关文献】
1.基于多尺度梯度角和SVM的正面人脸识别方法 [J], 赵武锋;严晓浪
2.基于肤色和人脸形状约束的正面人脸轮廓提取算法 [J], 黄福珍;佘星星
3.基于保边滤波和肤色模型的人脸美颜技术研究与实现 [J], 王志强;苗翔宇
4.基于肤色模型、神经网络和人脸结构模型的平面旋转人脸检测 [J], 张洪明;赵德斌;高文
5.基于肤色模型与人脸结构特征的人脸检测 [J], 肖红;南威治
因版权原因，仅展示原文概要，查看原文内容请购买。