有效重采样的颜色编辑扩散

目的 颜色编辑扩散是图像处理领域中的一个重要研究内容。以往方法中,为了取得期望的编辑效果,要求交互选定的样本颜色尽可能丰富。为了降低输入条件,提出一种新的有效重采样的颜色编辑扩散技术。方法 首先对图像做超像素分割并提取每个超像素中心作为重采样样本。遵循超像素中心非边缘像素,只服从一种编辑模式的原则,计算其与每种编辑模式的相似度,如果与某种编辑模式相似度绝对占优,则该样本的编辑模式直接确定。对于待定样本,依照相邻相似样本服从相同编辑模式的思想来判断其归属。然后将重采样样本的颜色编辑遵循相似颜色编辑后仍相似的原则采用增量的方式扩散到整幅图像。扩散时每个像素自适应地选择样本的数量,以避免相似度低样本的干扰。结果 与以往的方法进行比较,该方法对输入条件的敏感度较低,且在相同的相对简单的输入条件下能取得更好的视觉效果,可视化的梯度显示细节保持也更理想。结论 该方法能有效简化用户交互,将稀疏的画笔像素的扩散转化为颜色信息相对丰富的超像素中心样本的扩散,且生成的图像能忠于编辑颜色及较好地保持图像细节。     

Learning bidirectional asymmetric similarity for collaborative filtering via matrix factorization

Memory-based collaborative filtering (CF) aims at predicting the rating of a certain item for a particular user based on the previous ratings from similar users and/or similar items. Previous studies in finding similar users and items have several drawbacks. First, they are based on user-defined similarity measurements, such as Pearson Correlation Coefficient (PCC) or Vector Space Similarity (VSS), which are, for the most part, not adaptive and optimized for specific applications and data. Second, these similarity measures are restricted to symmetric ones such that the similarity between A and B is the same as that for B and A, although symmetry may not always hold in many real world applications. Third, they typically treat the similarity functions between users and functions between items separately. However, in reality, the similarities between users and between items are inter-related. In this paper, we propose a novel unified model for users and items, known as Similarity Learning based Collaborative Filtering (SLCF) , based on a novel adaptive bidirectional asymmetric similarity measurement. Our proposed model automatically learns asymmetric similarities between users and items at the same time through matrix factorization. Theoretical analysis shows that our model is a novel generalization of singular value decomposition (SVD). We show that, once the similarity relation is learned, it can be used flexibly in many ways for rating prediction. To take full advantage of the model, we propose several strategies to make the best use of the proposed similarity function for rating prediction. The similarity can be used either to improve the memory-based approaches or directly in a model based CF approaches. In addition, we also propose an online version of the rating prediction method to incorporate new users and new items. We evaluate SLCF using three benchmark datasets, including MovieLens, EachMovie and Netflix, through which we show that our methods can outperform many state-of-the-art baselines.     

The IKBALS project: Multi-modal reasoning in legal knowledge based systems

In attempting to build intelligent litigation support tools, we have moved beyond first generation, production rule legal expert systems. Our work integrates rule based and case based reasoning with intelligent information retrieval.When using the case based reasoning methodology, or in our case the specialisation of case based retrieval, we need to be aware of how to retrieve relevant experience. Our research, in the legal domain, specifies an approach to the retrieval problem which relies heavily on an extended object oriented/rule based system architecture that is supplemented with causal background information. We use a distributed agent architecture to help support the reasoning process of lawyers.Our approach to integrating rule based reasoning, case based reasoning and case based retrieval is contrasted to the CABARET and PROLEXS architectures which rely on a centralised blackboard architecture. We discuss in detail how our various cooperating agents interact, and provide examples of the system at work. The IKBALS system uses a specialised induction algorithm to induce rules from cases. These rules are then used as indices during the case based retrieval process.Because we aim to build legal support tools which can be modified to suit various domains rather than single purpose legal expert systems, we focus on principles behind developing legal knowledge based systems. The original domain chosen was theAccident Compensation Act 1989 (Victoria, Australia), which relates to the provision of benefits for employees injured at work. For various reasons, which are indicated in the paper, we changed our domain to that ofCredit Act 1984 (Victoria, Australia). This Act regulates the provision of loans by financial institutions.The rule based part of our system which provides advice on the Credit Act has been commercially developed in conjunction with a legal firm. We indicate how this work has lead to the development of a methodology for constructing rule based legal knowledge based systems. We explain the process of integrating this existing commercial rule based system with the case base reasoning and retrieval architecture.     

非下采样剪切波变换的医学图像融合

目的 由于单模态医学图像所提供的信息有限,不能反映相关组织所有细节信息,可能会造成临床医学误诊。针对这一问题,提出一种基于非下采样剪切波变换（NSST）的医学图像融合算法,对多模态医学图像进行融合,丰富融合图像信息,提高图像质量,为临床诊断提供依据。方法 首先,将源图像进行NSST变换得到低频子带和若干高频方向子带;其次,根据低频子带图像的特点,提出低频系数与脉冲耦合神经网络PCNN （pulse coupled neural network）相结合的方法;根据高频子带间结构相似度SSIM （structure similarity）不同,分为低相似和高相似子带图像;对低相似子带系数采用视觉敏感度系数VSC （visual sensitivity coefficient）与改进梯度能量相结合的策略;高相似子带系数采用VSC与区域能量相结合的方法;进而,选取结构相似度与边缘信息评价因子Q^ABF（edge based similarity measure）之和作为目标函数,自适应地优化可调参数;最后,经逆NSST变换重构图像。结果 对灰度图像和彩色图像进行实验仿真,并与其他4种融合方法进行比较,在主观视觉效果和客观评价标准,本文方法取得良好的融合效果,其中边缘评价因子和标准差都是最好的,其他指标相对较好;与靳珍怡提出的基于非下采样轮廓波变换的多模态医学图像融合相比,5组实验空间频率分别提高了11.8%、24.7%、83.4%、11.9%、30.3%;边缘评价因子分别提高了6.7%、15%、40%、50%、12%;结构相似度分别提高了0.7%、7.3%、2.4%、-3.6%、2.1%;交叉熵分别降低了16.9%、1.6%、-27.4%、6.1%、0.4%。结论 本文算法有效提高多模态医学图像融合质量,增加不同模态间的互补信息;与现有医学图像融合算法相比,本文算法更加优越。融合图像细节信息更为突出,整体信息更丰富,更符合人眼视觉特性。     

Discovery of a perceptual distance function for measuring image similarity

For more than a decade, researchers have actively explored the area of image/video analysis and retrieval. Yet one fundamental problem remains largely unsolved: how to measure perceptual similarity between two objects. For this purpose, most researchers employ a Minkowski-type metric. Unfortunately, the Minkowski metric does not reliably find similarities in objects that are obviously alike. Through mining a large set of visual data, our team has discovered a perceptual distance function. We call the discovered function the dynamic partial function (DPF). When we empirically compare DPF to Minkowski-type distance functions in image retrieval and in video shot-transition detection using our image features, DPF performs significantly better. The effectiveness of DPF can be explained by similarity theories in cognitive psychology.     

Approximate entity extraction in temporal databases

We study the problem of efficiently extracting K entities, in a temporal database, which are most similar to a given search query. This problem is well studied in relational databases, where each entity is represented as a single record and there exist a variety of methods to define the similarity between a record and the search query. However, in temporal databases, each entity is represented as a sequence of historical records. How to properly define the similarity of each entity in the temporal database still remains an open problem. The main challenging is that, when a user issues a search query for an entity, he or she is prone to mix up information of the same entity at different time points. As a result, methods, which are used in relational databases based on record granularity, cannot work any further. Instead, we regard each entity as a set of “virtual records”, where attribute values of a “virtual record” can be from different records of the same entity. In this paper, we propose a novel evaluation model, based on which the similarity between each “virtual record” and the query can be effectively quantified, and the maximum similarity of its “virtual records” is taken as the similarity of an entity. For each entity, as the number of its “virtual records” is exponentially large, calculating the similarity of the entity is challenging. As a result, we further propose a Dominating Tree Algorithm (DTA), which is based on the bounding-pruning-refining strategy, to efficiently extract K entities with greatest similarities. We conduct extensive experiments on both real and synthetic datasets. The encouraging results show that our model for defining the similarity between each entity and the search query is effective, and the proposed DTA can perform at least two orders of magnitude improvement on the performance comparing with the naive approach.     

A rigorous methodology for specification and verification of business processes

Both specification and verification of business processes are gaining more and more attention in the field. Most of the existing works in the last years are dealing with important, yet very specialized, issues. Among these, we can enumerate compensation constructs to cope with exceptions generated by long running business transactions, fully programmable fault and compensation handling mechanism, web service area, scope-based compensation and shared-labels for synchronization, and so on. The main purpose of this paper is to present a semi-automatized framework to describe and analyse business processes. Business analysts can now use a simple specification language (e.g., BPMN [Obj06]) to describe any type of activity in a company, in a concurrent and modular fashion. The associated programs (e.g., BPDs [Obj06]) have to be executed in an appropriate language (e.g., BPEL4WS [ACD⁺03]). Much more, they have to be confirmed to be sound, via some prescribed (a priori) conditions. We suggest how all the issues can be embedded in a unifying computer tool. We link our work with similar approaches and we justify our particular choices (besides BPMN and BPD): the TLA+ language for expressing the imposed behavioural conditions and Petri Nets ([EB87], [EB88]) to describe an intermediate semantics. In fact, we want to manage in an appropriate way the general relationship diagram (Fig. 1). Examples and case studies are provided. General relationship diagram     

Overlapping correlation clustering

We introduce a new approach for finding overlapping clusters given pairwise similarities of objects. In particular, we relax the problem of correlation clustering by allowing an object to be assigned to more than one cluster. At the core of our approach is an optimization problem in which each data point is mapped to a small set of labels, representing membership in different clusters. The objective is to find a mapping so that the given similarities between objects agree as much as possible with similarities taken over their label sets. The number of labels can vary across objects. To define a similarity between label sets, we consider two measures: (i) a 0–1 function indicating whether the two label sets have non-zero intersection and (ii) the Jaccard coefficient between the two label sets. The algorithm we propose is an iterative local-search method. The definitions of label set similarity give rise to two non-trivial optimization problems, which, for the measures of set-intersection and Jaccard, we solve using a greedy strategy and non-negative least squares, respectively. We also develop a distributed version of our algorithm based on the BSP model and implement it using a Pregel framework. Our algorithm uses as input pairwise similarities of objects and can thus be applied when clustering structured objects for which feature vectors are not available. As a proof of concept, we apply our algorithms on three different and complex application domains: trajectories, amino-acid sequences, and textual documents.     

Centralized and distributed algorithms for on-line synthesis of maximal control policies under partial observation

This paper deals with the on-line control of partially observed discrete event systems (DES). The goal is to restrict the behavior of the system within a prefix-closed legal language while accounting for the presence ofuncontrollable andunobservable events. In the spirit of recent work on the on-line control of partially observed DES (Heymann and Lin 1994) and on variable lookahead control of fully observed DES (Ben Hadj-Alouane et al. 1994c), we propose an approach where, following each observable event, a control action is computed on-line using an algorithm oflinear worst-case complexity. This algorithm, calledVLP-PO, has the following additional properties: (i) the resulting behavior is guaranteed to be amaximal controllable and observable sublanguage of the legal language; (ii) different maximals may be generated by varying the priorities assigned to the controllable events, a parameter ofVLP-PO; (iii) a maximal containing the supremal controllable and normal sublanguage of the legal language can be generated by a proper selection of controllable event priorities; and (iv) no off-line calculations are necessary. We also present a parallel/distributed version of theVLP-PO algorithm calledDI-VLP-PO. This version uses several communicating agents that simultaneously run (on-line) identical versions of the algorithm but on possibly different parts of the system model and the legal language, according to the structural properties of the system and the specifications. While achieving the same behavior asVLO-PO, DI-VLP-PO runs at a total complexity (for computation and communication) that is significantly lower than its sequential counterpart.     

Anytime similarity measures for faster alignment

Image alignment refers to finding the best transformation from a fixed reference image to a new image of a scene. This process is often optimizing a similarity measure between images, computed based on the image data. However, in time-critical applications state-of-the-art methods for computing similarity are too slow. Instead of using all the image data to compute similarity, one could use only a subset of pixels to improve the speed, but often this comes at the cost of reduced accuracy. These kinds of tradeoffs between the amount of computation and the accuracy of the result have been addressed in the field of real-time artificial intelligence as deliberation control problems. We propose that the optimization of a similarity measure is a natural application domain for deliberation control using the anytime algorithm framework. In this paper, we present anytime versions for the computation of two common image similarity measures: mean squared difference and mutual information. Off-line, we learn a performance profile specific to each measure, which is then used on-line to select the appropriate amount of pixels to process at each optimization step. When tested against existing techniques, our method achieves comparable quality and robustness with significantly less computation.     

A Flexible Algorithm for Generating All the Spanning Trees in Undirected Graphs

In this paper we propose an algorithm for generating all the spanning trees in undirected graphs. The algorithm requires O (n+m+ τ n) time where the given graph has n vertices, m edges, and τ spanning trees. For outputting all the spanning trees explicitly, this time complexity is optimal. Our algorithm follows a special rooted tree structure on the skeleton graph of the spanning tree polytope. The rule by which the rooted tree structure is traversed is irrelevant to the time complexity. In this sense, our algorithm is flexible. If we employ the depth-first search rule, we can save the memory requirement to O (n+m). A breadth-first implementation requires as much as O (m+ τ n) space, but when a parallel computer is available, this might have an advantage. When a given graph is weighted, the best-first search rule provides a ranking algorithm for the minimum spanning tree problem. The ranking algorithm requires O (n+ m + τ n) time and O (m+ τ n) space when we have a minimum spanning tree. Received January 21, 1995; revised February 19, 1996.     

Online variance minimization

We consider the following type of online variance minimization problem: In every trial t our algorithms get a covariance matrix C ^t and try to select a parameter vector w ^t−1 such that the total variance over a sequence of trials ?_t=1^T (w^t-1)^T C^tw^t-1\sum_{t=1}^{T} (\boldsymbol {w}^{t-1})^{\top} \boldsymbol {C}^{t}\boldsymbol {w}^{t-1} is not much larger than the total variance of the best parameter vector u chosen in hindsight. Two parameter spaces in ℝ ⁿ are considered—the probability simplex and the unit sphere. The first space is associated with the problem of minimizing risk in stock portfolios and the second space leads to an online calculation of the eigenvector with minimum eigenvalue of the total covariance matrix ?_t=1^T C^t\sum_{t=1}^{T} \boldsymbol {C}^{t}. For the first parameter space we apply the Exponentiated Gradient algorithm which is motivated with a relative entropy regularization. In the second case, the algorithm has to maintain uncertainty information over all unit directions u. For this purpose, directions are represented as dyads uu ^⊤ and the uncertainty over all directions as a mixture of dyads which is a density matrix. The motivating divergence for density matrices is the quantum version of the relative entropy and the resulting algorithm is a special case of the Matrix Exponentiated Gradient algorithm. In each of the two cases we prove bounds on the additional total variance incurred by the online algorithm over the best offline parameter.     

Assessing single-pair similarity over graphs by aggregating first-meeting probabilities

Link-based similarity plays an important role in measuring similarities between nodes in a graph. As a widely used link-based similarity, SimRank scores similarity between two nodes as the first-meeting probability of two random surfers. However, due to the large scale of graphs in real-world applications and dynamic change characteristic, it is not viable to frequently update the whole similarity matrix. Also, people often only concern about the similarities of a small subset of nodes in a graph. In such a case, the existing approaches need to compute the similarities of all node-pairs simultaneously, suffering from high computation cost.In this paper, we propose a new algorithm, Iterative Single-Pair SimRank (ISP), based on the random surfer-pair model to compute the SimRank similarity score for a single pair of nodes in a graph. To avoid computing similarities of all other nodes, we introduce a new data structure, position matrix, to facilitate computation of the first-meeting probabilities of two random surfers, and give two optimization techniques to further enhance their performance. In addition, we theoretically prove that the time cost of ISP is always less than the original algorithm SimRank. Comprehensive experiments conducted on both synthetic and real datasets demonstrate the effectiveness and efficiency of our approach.     

SimClus: an effective algorithm for clustering with a lower bound on similarity

Clustering algorithms generally accept a parameter k from the user, which determines the number of clusters sought. However, in many application domains, like document categorization, social network clustering, and frequent pattern summarization, the proper value of k is difficult to guess. An alternative clustering formulation that does not require k is to impose a lower bound on the similarity between an object and its corresponding cluster representative. Such a formulation chooses exactly one representative for every cluster and minimizes the representative count. It has many additional benefits. For instance, it supports overlapping clusters in a natural way. Moreover, for every cluster, it selects a representative object, which can be effectively used in summarization or semi-supervised classification task. In this work, we propose an algorithm, SimClus, for clustering with lower bound on similarity. It achieves a O(log n) approximation bound on the number of clusters, whereas for the best previous algorithm the bound can be as poor as O(n). Experiments on real and synthetic data sets show that our algorithm produces more than 40% fewer representative objects, yet offers the same or better clustering quality. We also propose a dynamic variant of the algorithm, which can be effectively used in an on-line setting.     

A cross-modal method of labeling music tags

In this paper, we discuss various features of music objects in two kinds of domain. Among these features, Mel-frequency cepstral coefficients (MFCCs) are further discussed and described by Gaussian mixture model (GMM). Also, the similarity between GMMs are investigated accordingly. Then, we employ the multimedia graph as a cross-modal method to associate MFCCs and genre tags of music objects. By applying link analysis algorithm in the graph, we label appropriate genre tags for target music objects. Also, we perform experiments to show performance, effectiveness, and parameter setting of our approach.     

Supporting Efficient Record Linkage for Large Data Sets Using Mapping Techniques

This paper describes an efficient approach to record linkage. Given two lists of records, the record-linkage problem consists of determining all pairs that are similar to each other, where the overall similarity between two records is defined based on domain-specific similarities over individual attributes. The record-linkage problem arises naturally in the context of data cleansing that usually precedes data analysis and mining. Since the scalability issue of record linkage was addressed in [21], the repertoire of database techniques dealing with multidimensional data sets has significantly increased. Specifically, many effective and efficient approaches for distance-preserving transforms and similarity joins have been developed. Based on these advances, we explore a novel approach to record linkage. For each attribute of records, we first map values to a multidimensional Euclidean space that preserves domain-specific similarity. Many mapping algorithms can be applied, and we use the Fastmap approach [16] as an example. Given the merging rule that defines when two records are similar based on their attribute-level similarities, a set of attributes are chosen along which the merge will proceed. A multidimensional similarity join over the chosen attributes is used to find similar pairs of records. Our extensive experiments using real data sets show that our solution has very good efficiency and recall. Part of this article was published in [28]. In addition to the prior materials, this article contains more analysis, a complete proof, and more experimental results that were not included in the original paper.     

Rough set-based approach to rule generation and rule induction

During the last decade, databases have been growing rapidly in size and number as a result of rapid advances in database capacity and management techniques. This expansive growth in data and databases has caused a pressing need for the development of more powerful techniques to convert the vast pool of data into valuable information. For the purpose of strategic and decision-making, many companies and researchers have recognized mining useful information and knowledge from large databases as a key research topic and as an opportunity for major revenues and improving competitiveness. In this paper, we will explore a new rule generation algorithm (based on rough sets theory) that can generate a minimal set of rule reducts, and a rule generation and rule induction program (RGRIP) which can efficiently induce decision rules from conflicting information systems. All the methods will also be illustrated with numerical examples.     

An adaptive version of <Emphasis Type="Italic">k</Emphasis>-medoids to deal with the uncertainty in clustering heterogeneous data using an intermediary fusion approach

This paper introduces Hk-medoids, a modified version of the standard k-medoids algorithm. The modification extends the algorithm for the problem of clustering complex heterogeneous objects that are described by a diversity of data types, e.g. text, images, structured data and time series. We first proposed an intermediary fusion approach to calculate fused similarities between objects, SMF, taking into account the similarities between the component elements of the objects using appropriate similarity measures. The fused approach entails uncertainty for incomplete objects or for objects which have diverging distances according to the different component. Our implementation of Hk-medoids proposed here works with the fused distances and deals with the uncertainty in the fusion process. We experimentally evaluate the potential of our proposed algorithm using five datasets with different combinations of data types that define the objects. Our results show the feasibility of the our algorithm, and also they show a performance enhancement when comparing to the application of the original SMF approach in combination with a standard k-medoids that does not take uncertainty into account. In addition, from a theoretical point of view, our proposed algorithm has lower computation complexity than the popular PAM implementation.     

A regression model based on the nearest centroid neighborhood

The renowned k-nearest neighbor decision rule is widely used for classification tasks, where the label of any new sample is estimated based on a similarity criterion defined by an appropriate distance function. It has also been used successfully for regression problems where the purpose is to predict a continuous numeric label. However, some alternative neighborhood definitions, such as the surrounding neighborhood, have considered that the neighbors should fulfill not only the proximity property, but also a spatial location criterion. In this paper, we explore the use of the k-nearest centroid neighbor rule, which is based on the concept of surrounding neighborhood, for regression problems. Two support vector regression models were executed as reference. Experimentation over a wide collection of real-world data sets and using fifteen odd different values of k demonstrates that the regression algorithm based on the surrounding neighborhood significantly outperforms the traditional k-nearest neighborhood method and also a support vector regression model with a RBF kernel.     

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。