渐进式前景更新和区域关联的图像协同分割

目的 图像协同分割技术是通过多幅参考图像以实现前景目标与背景区域的分离，并已被广泛应用于图像分类和目标识别等领域中。不过，现有多数的图像协同分割算法只适用于背景变化较大且前景几乎不变的环境。为此，提出一种新的无监督协同分割算法。方法 本文方法是无监督式的，在分级图像分割的基础上通过渐进式优化框架分别实现前景和背景模型的更新估计，同时结合图像内部和不同图像之间的分级区域相似度关联进一步增强上述模型估计的鲁棒性。该无监督的方法不需要进行预先样本学习，能够同时处理两幅或多幅图像且适用于同时存在多个前景目标的情况，并且能够较好地适应前景物体类的变化。结果 通过基于iCoseg和MSRC图像集的实验证明，该算法无需图像间具有显著的前景和背景差异这一约束，与现有的经典方法相比更适用于前景变化剧烈以及同时存在多个前景目标等更为一般化的图像场景中。结论 该方法通过对分级图像分割得到的超像素外观分布分别进行递归式估计来实现前景和背景的有效区分，并同时融合了图像内部以及不同图像区域之间的区域关联性来增加图像前景和背景分布估计的一致性。实验表明当前景变化显著时本文方法相比于现有方法具有更为鲁棒的表现。     

基于YOLOv5和FCN-DenseNet水下图像多目标语义分割算法

带视觉系统的水下机器人作业离不开对水下目标准确的分割,但水下环境复杂,场景感知精度和识别精度不高等问题会严重影响目标分割算法的性能.针对此问题本文提出了一种综合YOLOv5和FCN-DenseNet的多目标分割算法.本算法以FCN-DenseNet算法为主要分割框架, YOLOv5算法为目标检测框架.采用YOLOv5算法检测出每个种类目标所在位置;然后输入针对不同类别的FCN-DenseNet语义分割网络,实现多分支单目标语义分割,最后融合分割结果实现多目标语义分割.此外,本文在Kaggle竞赛平台上的海底图片数据集上将所提算法与PSPNet算法和FCN-DenseNet算法两种经典的语义分割算法进行了实验对比.结果表明本文所提的多目标图像语义分割算法与PSPNet算法相比,在MIoU和IoU指标上分别提高了14.9%和11.6%;与FCN-DenseNet算法在MIoU和IoU指标上分别提高了8%和7.7%,更适合于水下图像分割.     

Multiphase Soft Segmentation with Total Variation and <Emphasis Type="Italic">H</Emphasis><Superscript>1</Superscript> Regularization

In this paper, we propose a variational soft segmentation framework inspired by the level set formulation of multiphase Chan-Vese model. We use soft membership functions valued in [0,1] to replace the Heaviside functions of level sets (or characteristic functions) such that we get a representation of regions by soft membership functions which automatically satisfies the sum to one constraint. We give general formulas for arbitrary N-phase segmentation, in contrast to Chan-Vese’s level set method only 2 ^m -phase are studied. To ensure smoothness on membership functions, both total variation (TV) regularization and H ¹ regularization used as two choices for the definition of regularization term. TV regularization has geometric meaning which requires that the segmentation curve length as short as possible, while H ¹ regularization has no explicit geometric meaning but is easier to implement with less parameters and has higher tolerance to noise. Fast numerical schemes are designed for both of the regularization methods. By changing the distance function, the proposed segmentation framework can be easily extended to the segmentation of other types of images. Numerical results on cartoon images, piecewise smooth images and texture images demonstrate that our methods are effective in multiphase image segmentation.     

Efficient Exact Algorithms on Planar Graphs: Exploiting Sphere Cut Decompositions

We present a general framework for designing fast subexponential exact and parameterized algorithms on planar graphs. Our approach is based on geometric properties of planar branch decompositions obtained by Seymour and Thomas, combined with refined techniques of dynamic programming on planar graphs based on properties of non-crossing partitions. To exemplify our approach we show how to obtain an  $O(2^{6.903\sqrt{n}})We present a general framework for designing fast subexponential exact and parameterized algorithms on planar graphs. Our approach is based on geometric properties of planar branch decompositions obtained by Seymour and Thomas, combined with refined techniques of dynamic programming on planar graphs based on properties of non-crossing partitions. To exemplify our approach we show how to obtain an  O(2^6.903?n)O(2^{6.903\sqrt{n}}) time algorithm solving weighted Hamiltonian Cycle on an n-vertex planar graph. Similar technique solves Planar Graph Travelling Salesman Problem with n cities in time O(2^9.8594?n)O(2^{9.8594\sqrt{n}}) . Our approach can be used to design parameterized algorithms as well. For example, we give an algorithm that for a given k decides if a planar graph on n vertices has a cycle of length at least k in time O(2^13.6?kn+n³)O(2^{13.6\sqrt{k}}n+n^{3}) .     

Discrete Surfaces and Frontier Orders

Many applications require the extraction of an object boundary from a discrete image. In most cases, the result of such a process is expected to be, topologically, a surface, and this property might be required in subsequent operations. However, only through careful design can such a guarantee be provided. In the present article we will focus on partially ordered sets and the notion of n-surfaces introduced by Evako et al. to deal with this issue. Partially ordered sets are topological spaces that can represent the topology of a wide range of discrete spaces, including abstract simplicial complexes and regular grids. It will be proved in this article that (in the framework of simplicial complexes) any n-surface is an n-pseudomanifold, and that any n-dimensional combinatorial manifold is an n-surface. Moreover, given a subset of an n-surface (an object), we show how to build a partially ordered set called frontier order, which represents the boundary of this object. Similarly to the continuous case, where the boundary of an n-manifold, if not empty, is an (n−1)-manifold, we prove that the frontier order associated to an object is a union of disjoint (n−1)-surfaces. Thanks to this property, we show how topologically consistent Marching Cubes-like algorithms can be designed using the framework of partially ordered sets.X. Daragon is a Ph.D. student at ESIEE, A2SI laboratory. He received a DEA in computer science from Marne-La-Vallee University in 2000. His research focuses on order theory and its applications to image analysis and computer graphics, mainly in the field of 3D medical imaging (segmentation of the cerebral neo-cortex).Michel Couprie received his Ingénieur’s degree from the École Supérieure d’Ingénieurs en Électrotechnique et Électronique (Paris, France) in 1985 and the Ph.D. degree from the Pierre et Marie Curie University (Paris, France) in 1988. Since 1988 he has been working in ESIEE where he is an Associate Professor. He is a member of the Laboratoire Algorithmique et Architecture des Systèmes Informatiques, ESIEE, Paris, and of the Institut Gaspard Monge, Université de Marne-la-Vallée. His current research interests include image analysis and discrete mathematics.Gilles Bertrand received his Ingénieur’s degree from the École Centrale des Arts et Manufactures in 1976. Until 1983 he was with the Thomson-CSF company, where he designed image processing systems for aeronautical applications. He received his Ph.D. from the École Centrale in 1986. He is currently teaching and doing research with the Laboratoire Algorithmique et Architecture des Systèmes Informatiques, ESIEE, Paris, and with the Institut Gaspard Monge, Université de Marne-la-Vallée. His research interests are image analysis, pattern recognition, mathematical morphology and digital topology.     

结合超像元和子空间投影支持向量机的高光谱图像分类

目的 高光谱图像包含了丰富的空间、光谱和辐射信息,能够用于精细的地物分类,但是要达到较高的分类精度,需要解决高维数据与有限样本之间存在矛盾的问题,并且降低因噪声和混合像元引起的同物异谱的影响。为有效解决上述问题,提出结合超像元和子空间投影支持向量机的高光谱图像分类方法。方法 首先采用简单线性迭代聚类算法将高光谱图像分割成许多无重叠的同质性区域,将每一个区域作为一个超像元,以超像元作为图像分类的最小单元,利用子空间投影算法对超像元构成的图像进行降维处理,在低维特征空间中执行支持向量机分类。本文高光谱图像空谱综合分类模型,对几何特征空间下的超像元分割与光谱特征空间下的子空间投影支持向量机（SVMsub）,采用分割后进行特征融合的处理方式,将像元级别转换为面向对象的超像元级别,实现高光谱图像空谱综合分类。结果 在AVIRIS（airbone visible/infrared imaging spectrometer）获取的Indian Pines数据和Reflective ROSIS（optics system spectrographic imaging system）传感器获取的University of Pavia数据实验中,子空间投影算法比对应的非子空间投影算法的分类精度高,特别是在样本数较少的情况下,分类效果提升明显;利用马尔可夫随机场或超像元融合空间信息的算法比对应的没有融合空间信息的算法的分类精度高;在两组数据均使用少于1%的训练样本情况下,同时融合了超像元和子空间投影的支持向量机算法在两组实验中分类精度均为最高,整体分类精度高出其他相关算法4%左右。结论 利用超像元处理可以有效融合空间信息,降低同物异谱对分类结果的不利影响;采用子空间投影能够将高光谱数据变换到低维空间中,实现有限训练样本条件下的高精度分类;结合超像元和子空间投影支持向量机的算法能够得到较高的高光谱图像分类精度。     

Enumerate and Expand: Improved Algorithms for Connected Vertex Cover and Tree Cover

We present a new method of solving graph problems related to Vertex Cover by enumerating and expanding appropriate sets of nodes. As an application, we obtain dramatically improved runtime bounds for two variants of the Vertex Cover problem. In the case of Connected Vertex Cover, we take the upper bound from O ^*(6 ^k ) to O ^*(2.7606 ^k ) without large hidden factors. For Tree Cover, we show a complexity of O ^*(3.2361 ^k ), improving over the previous bound of O ^*((2k) ^k ). In the process, faster algorithms for solving subclasses of the Steiner tree problem on graphs are investigated. Supported by the DFG under grant RO 927/6-1 (TAPI).     

Globally Optimal Geodesic Active Contours

An approach to optimal object segmentation in the geodesic active contour framework is presented with application to automated image segmentation. The new segmentation scheme seeks the geodesic active contour of globally minimal energy under the sole restriction that it contains a specified internal point p_int. This internal point selects the object of interest and may be used as the only input parameter to yield a highly automated segmentation scheme. The image to be segmented is represented as a Riemannian space S with an associated metric induced by the image. The metric is an isotropic and decreasing function of the local image gradient at each point in the image, encoding the local homogeneity of image features. Optimal segmentations are then the closed geodesics which partition the object from the background with minimal similarity across the partitioning. An efficient algorithm is presented for the computation of globally optimal segmentations and applied to cell microscopy, x-ray, magnetic resonance and cDNA microarray images.Ben Appleton received degrees in engineering and in science from the University of Queensland in 2001 and was awarded a university medal. In 2002 he began a Ph.D at the University of Queensland in the field of image analysis. He is supported by an Australian Postgraduate Award and the Commonwealth Scientific and Industrial Research Organisation (CSIRO), Mathematical and Information Sciences. He has been a teaching assistant in image analysis at the University of Queensland since 2001. He has also contributed 10 research papers to international journals and conferences and was recently awarded the prize for Best Student Paper at Digital Image Computing: Techniques and Applications. His research interests include image segmentation, stereo vision and algorithms.Hugues Talbot received the engineering degree from École Centrale de Paris in 1989, the D.E.A. (Masters) from University Paris VI in 1990 and the Ph.D from École des Mines de Paris in 1993, under the guidance of Dominique Jeulin and Jean Serra. He has been affiliated with the Commonwealth Scientific and Industrial Research Organisation (CSIRO), Mathematical and Information Sciences since 1994. He has worked on numerous applied projects in relation with industry, he has contributed more than 30 research papers in international journals and conferences and he has co-edited two sets of international conference proceedings on image analysis. He now also teaches image processing at the University of Sydney, and his research interest include image segmentation, linear structure analysis, texture analysis and algorithms.     

Shape-Based Mutual Segmentation

We present a novel variational approach for simultaneous segmentation of two images of the same object taken from different viewpoints. Due to noise, clutter and occlusions, neither of the images contains sufficient information for correct object-background partitioning. The evolving object contour in each image provides a dynamic prior for the segmentation of the other object view. We call this process mutual segmentation. The foundation of the proposed method is a unified level-set framework for region and edge based segmentation, associated with a shape similarity term. The suggested shape term incorporates the semantic knowledge gained in the segmentation process of the image pair, accounting for excess or deficient parts in the estimated object shape. Transformations, including planar projectivities, between the object views are accommodated by a registration process held concurrently with the segmentation. The proposed segmentation algorithm is demonstrated on a variety of image pairs. The homography between each of the image pairs is estimated and its accuracy is evaluated.     

Image Parsing: Unifying Segmentation,Detection, and Recognition

In this paper we present a Bayesian framework for parsing images into their constituent visual patterns. The parsing algorithm optimizes the posterior probability and outputs a scene representation as a parsing graph, in a spirit similar to parsing sentences in speech and natural language. The algorithm constructs the parsing graph and re-configures it dynamically using a set of moves, which are mostly reversible Markov chain jumps. This computational framework integrates two popular inference approaches—generative (top-down) methods and discriminative (bottom-up) methods. The former formulates the posterior probability in terms of generative models for images defined by likelihood functions and priors. The latter computes discriminative probabilities based on a sequence (cascade) of bottom-up tests/filters. In our Markov chain algorithm design, the posterior probability, defined by the generative models, is the invariant (target) probability for the Markov chain, and the discriminative probabilities are used to construct proposal probabilities to drive the Markov chain. Intuitively, the bottom-up discriminative probabilities activate top-down generative models. In this paper, we focus on two types of visual patterns—generic visual patterns, such as texture and shading, and object patterns including human faces and text. These types of patterns compete and cooperate to explain the image and so image parsing unifies image segmentation, object detection, and recognition (if we use generic visual patterns only then image parsing will correspond to image segmentation (Tu and Zhu, 2002. IEEE Trans. PAMI, 24(5):657–673). We illustrate our algorithm on natural images of complex city scenes and show examples where image segmentation can be improved by allowing object specific knowledge to disambiguate low-level segmentation cues, and conversely where object detection can be improved by using generic visual patterns to explain away shadows and occlusions.     

Simultaneous Segmentation and Pose Estimation of Humans Using Dynamic Graph Cuts

This paper presents a novel algorithm for performing integrated segmentation and 3D pose estimation of a human body from multiple views. Unlike other state of the art methods which focus on either segmentation or pose estimation individually, our approach tackles these two tasks together. Our method works by optimizing a cost function based on a Conditional Random Field (CRF). This has the advantage that all information in the image (edges, background and foreground appearances), as well as the prior information on the shape and pose of the subject can be combined and used in a Bayesian framework. Optimizing such a cost function would have been computationally infeasible. However, our recent research in dynamic graph cuts allows this to be done much more efficiently than before. We demonstrate the efficacy of our approach on challenging motion sequences. Although we target the human pose inference problem in the paper, our method is completely generic and can be used to segment and infer the pose of any rigid, deformable or articulated object.     

Generating rooted triangulations without repetitions

We use the reverse search technique to give algorithms for generating all graphs onn points that are 2- and 3-connected planar triangulations withr points on the outer face. The triangulations are rooted, which means the outer face has a fixed labelling. The triangulations are produced without duplications inO(n ²) time per triangulation. The algorithms useO(n) space. A program for generating all 3-connected rooted triangulations based on this algorithm is available by ftp.This research was supported by N.S.E.R.C. Grant Number A3013, F.C.A.R. Grant Number EQ1678, and a bilateral exchange from J.S.P.S./N.S.E.R.C.     

Efficient Pose Clustering Using a Randomized Algorithm

Pose clustering is a method to perform object recognition by determining hypothetical object poses and finding clusters of the poses in the space of legal object positions. An object that appears in an image will yield a large cluster of such poses close to the correct position of the object. If there are m model features and n image features, then there are O(m ³ n ³ ) hypothetical poses that can be determined from minimal information for the case of recognition of three-dimensional objects from feature points in two-dimensional images. Rather than clustering all of these poses, we show that pose clustering can have equivalent performance for this case when examining only O(mn) poses, due to correlation between the poses, if we are given two correct matches between model features and image features. Since we do not usually know two correct matches in advance, this property is used with randomization to decompose the pose clustering problem into O(n ² ) problems, each of which clusters O(mn) poses, for a total complexity of O(mn ³ ) . Further speedup can be achieved through the use of grouping techniques. This method also requires little memory and makes the use of accurate clustering algorithms less costly. We use recursive histograming techniques to perform clustering in time and space that is guaranteed to be linear in the number of poses. Finally, we present results demonstrating the recognition of objects in the presence of noise, clutter, and occlusion.     

Image segmentation method by merging and correction of sets of pixels

The Otsu method (1979) and the Mumford-Shah model (1985) for image segmentation described by image approximations by a step (piecewise constant) function are described and developed to formalize automatic object detection. Segmentation here means image preprocessing performed without tuning parameters, assumptions about image content, or training data from the user. A sequence of partitions of image pixels into sets that consist of either pixels of specific intensity ranges or pixels of connected image segments is considered a result of segmentation. The resulting partitions of image pixels need to generate a sequence of optimal approximations of an image averaged over the sets according to intensity with the lowest standard deviation of the approximation from the image. In general, the sequence of optimal approximations of an image is nonhierarchical, and the conventionally used merging of sets is insufficient to calculate it. Therefore, the merging of sets is complemented by a correction operation. The paper presents an analytical validation for correction of sets by reclassification of pixels and discusses the stability of optimal approximations during the reclassification operation of image pixels. Experimental results are demonstrated, and a comparative analysis of image approximations obtained in different algorithms is given. The prospects for improving segmentation with respect to standard deviation are analyzed.     

A New Algorithm for Finding Trees with Many Leaves

We present an algorithm that finds out-trees and out-branchings with at least k leaves in directed graphs. These problems are known as Directed Maximum Leaf Out-Tree and Directed Maximum Leaf Out-Branching, respectively, and—in the case of undirected graphs—as Maximum Leaf Spanning Tree. The run time of our algorithm is O(4 ^k nm) on directed graphs and O(poly(n)+4 ^k k ²) on undirected graphs. This improves over the previously fastest algorithms for these problems with run times of 2 ^O(klog k) poly(n) and O(poly(n)+6.75 ^k poly(k)) respectively.     

OBJCUT: Efficient Segmentation Using Top-Down and Bottom-Up Cues

We present a probabilistic method for segmenting instances of a particular object category within an image. Our approach overcomes the deficiencies of previous segmentation techniques based on traditional grid conditional random fields (CRF), namely that 1) they require the user to provide seed pixels for the foreground and the background and 2) they provide a poor prior for specific shapes due to the small neighborhood size of grid CRF. Specifically, we automatically obtain the pose of the object in a given image instead of relying on manual interaction. Furthermore, we employ a probabilistic model which includes shape potentials for the object to incorporate top-down information that is global across the image, in addition to the grid clique potentials which provide the bottom-up information used in previous approaches. The shape potentials are provided by the pose of the object obtained using an object category model. We represent articulated object categories using a novel layered pictorial structures model. Nonarticulated object categories are modeled using a set of exemplars. These object category models have the advantage that they can handle large intraclass shape, appearance, and spatial variation. We develop an efficient method, OBJCUT, to obtain segmentations using our probabilistic framework. Novel aspects of this method include: 1) efficient algorithms for sampling the object category models of our choice and 2) the observation that a sampling-based approximation of the expected log-likelihood of the model can be increased by a single graph cut. Results are presented on several articulated (e.g., animals) and nonarticulated (e.g., fruits) object categories. We provide a favorable comparison of our method with the state of the art in object category specific image segmentation, specifically the methods of Leibe and Schiele and Schoenemann and Cremers.     

An efficient parallel algorithm for geometrically characterising drawings of a class of 3-D objects

Labelling the lines of a planar line drawing of a 3-D object in a way that reflects the geometric properties of the object is a much studied problem in computer vision, considered to be an important step towards understanding the object from its 2-D drawing. Combinatorially, the labellability problem is a Constraint Satisfaction Problem and has been shown to be NP-complete even for images of polyhedral scenes. In this paper, we examine scenes that consist of a set of objects each obtained by rotating a polygon around an arbitrary axis. The objects are allowed to arbitrarily intersect or overlay. We show that for these scenes, there is a sequential lineartime labelling algorithm. Moreover, we show that the algorithm has a fast parallel version that executes inO(log³ n) time on an Exclusive-Read-Exclusive-Write Parallel Random Access Machine withO(n ³/log³ n) processors. The algorithm not only answers the decision problem of labellability, but also produces a legal labelling, if there is one. This parallel algorithm should be contrasted with the techniques of dealing with special cases of the constraint satisfaction problem. These techniques employ an effective, but inherently sequential, relaxation procedure in order to restrict the domains of the variables.This research was partially supported by the European Community ESPRIT Basic Research Program under contracts 7141 (project ALCOM II) and 6019 (project Insight II).     

融合分割先验的多图像目标语义分割

目的 在序列图像或多视角图像的目标分割中，传统的协同分割算法对复杂的多图像分割鲁棒性不强，而现有的深度学习算法在前景和背景存在较大歧义时容易导致目标分割错误和分割不一致。为此，提出一种基于深度特征的融合分割先验的多图像分割算法。方法 首先，为了使模型更好地学习复杂场景下多视角图像的细节特征，通过融合浅层网络高分辨率的细节特征来改进PSPNet-50网络模型，减小随着网络的加深导致空间信息的丢失对分割边缘细节的影响。然后通过交互分割算法获取一至两幅图像的分割先验，将少量分割先验融合到新的模型中，通过网络的再学习来解决前景/背景的分割歧义以及多图像的分割一致性。最后通过构建全连接条件随机场模型，将深度卷积神经网络的识别能力和全连接条件随机场优化的定位精度耦合在一起，更好地处理边界定位问题。结果 本文采用公共数据集的多图像集进行了分割测试。实验结果表明本文算法不但可以更好地分割出经过大量数据预训练过的目标类，而且对于没有预训练过的目标类，也能有效避免歧义的区域分割。本文算法不论是对前景与背景区别明显的较简单图像集，还是对前景与背景颜色相似的较复杂图像集，平均像素准确度（PA）和交并比（IOU）均大于95%。结论 本文算法对各种场景的多图像分割都具有较强的鲁棒性，同时通过融入少量先验，使模型更有效地区分目标与背景，获得了分割目标的一致性。