Search

Virtual Local Area Networks (VLANs) are extensively used in enterprise networks. However, their configuration remains an ad-hoc, complex and error-prone process today. We believe that to eliminate these difficulties, there is need for automation tools, and also need for visualization tools. In this paper, we report on our experience building a VLAN management toolkit, which automates and visualizes common VLAN configuration tasks. We begin by describing common misconfigurations, and their impact on network performance and security. We next present a set of algorithms that automate the VLAN configuration tasks. These algorithms form the back end of the toolkit. The front end of the toolkit consists of an interactive graphical user interface which provides visualization of VLAN operations at multiple granularities, and can be accessed remotely from a web browser. We are in the process of deploying the toolkit at a large campus network which has thousands of switches, and around 800 VLANs. Our initial operational experience shows that the toolkit is effective in both automating configuration tasks, and identifying common misconfigurations. In particular, we have found that (i) more than 40% of the VLANs in the network have redundant links that may lead to security and performance issues. (ii) more than 30% of the VLANs in the network have missing links which may result in connectivity issues and (iii) the root-bridge placements of more than 30% of the VLANs are not optimum, which again may result in performance issues. We believe these insights highlight the benefit and importance of such a toolkit.

IP network operators face the challenge of making and managing router configuration changes to serve rapidly evolving user and organizational needs. Changes are expressed in low-level languages, and often impact multiple parts of a configuration file and multiple routers. These dependencies make configuration changes difficult for operators to reason about, detect problems in, and troubleshoot. In this paper, we present a methodology to extract network-wide correlations of changes. From longitudinal snapshots of low-level router configuration data, our methodology identifies syntactic configuration blocks that changed, applies data mining techniques to extract correlated changes, and highlights changes of interest via operator feedback. Employing our methodology, we analyze an 11-month archive of router configuration data from 5 different large-scale enterprise Virtual Private Networks (VPNs). Our study shows that our techniques effectively extract correlated configuration changes, within and across individual routers, and shed light on the prevalence and causes of system-wide and intertwined change operations. A deeper understanding of correlated changes has potential applications in the design of an auditing system that can help operators proactively detect errors during change management. To demonstrate this, we conduct an initial study analyzing the prevalence and causes of anomalies in system-wide changes.

Enterprise networks are important, with size and complexity even surpassing carrier networks. Yet, the design of enterprise networks is ad-hoc and poorly understood. In this paper, we show how a systematic design approach can handle two key areas of enterprise design: virtual local area networks (VLANs) and reachability control. We focus on these tasks given their complexity, prevalence, and time-consuming nature. Our contributions are three-fold. First, we show how these design tasks may be formulated in terms of network-wide performance, security, and resilience requirements. Our formulations capture the correctness and feasibility constraints on the design, and they model each task as one of optimizing desired criteria subject to the constraints. The optimization criteria may further be customized to meet operator-preferred design strategies. Second, we develop a set of algorithms to solve the problems that we formulate. Third, we demonstrate the feasibility and value of our systematic design approach through validation on a large-scale campus network with hundreds of routers and VLANs.

In this paper we present a study characterizing VLAN us- age in a large-sized campus network. Despite their extensive prevalence in enterprise and campus networks, the usage of VLANs has received little systematic treatment in the re- search community. Our study is conducted using a white- box approach, involving data such as router configuration files obtained from network operators, and through itera- tive interactions with them. Our study shows that the use of virtualization is prevalent to enable users belonging to physically disparate locations to be treated as a group. We demonstrate and characterize the performance inefficiencies resulting from virtualization. We show the inefficiencies are exacerbated by sub-optimal placement policies. We also dis- cuss potential sources of errors that may arise with configu- ration of VLANs, and demonstrate their prevalence in real configurations. We believe these results are a key step to- wards gaining deeper insights into operational practices in enterprise and campus networks, and the design of abstrac- tions to simplify management.

In order to deal with the small sample size (SSS) problem in linear discriminant analysis (LDA), a two-dimensional linear discriminant analysis (2DLDA) has been proposed [16, 25]. However, some discriminant information could be lost because the 2DLDA can be viewed as operating on the image rows or columns independently [24]. We know that the spatial correlation between neighbouring pixels in both vertical and horizontal directions tend to be high. This motivates us to consider other decompositions of the image. In this paper, a multi-block 2DLDA (MB2DLDA) is developed to consider the spatial correlation in which an image is decomposed into blocks of pixels where some of this local correlations could be captured when the 2DLDA is applied to the subimages instead of the whole image. Moreover, different facial regions have different degrees of importance in face recognition, so we combine the 2DLDA features of the different facial regions based on their degrees of importance. This can be done by weighting the 2DLDA feature of a block. In doing so, the verification performance is improved. We evaluate the approach on the face verification on the XM2VTS database based on a standard protocol defined in the database. The results show that the weighted MB2DLDA outperforms the 2DLDA.

Support vector machine has obtained much success in machine learning. But it requires to solve a quadratic optimization (QP) problem so that its training time increases dramatically with the increase of training set. Hence, standard SVM with batch learning has difficulty in handling large scale problems. In this paper, we introduce a SVM accelerator, called sequential bootstrapped SVM (SeqSVM), to speed up the training of SVM. At the beginning, the SeqSVM trains a SVM classifier on a small part of all training samples and then keeps on selecting the so-called convex hull samples from the given large training set to retrain this SVM until all convex hull samples are selected. The key principle in our method is to help the SVM pick the convex hull sample that is wrongly classified by the current SVM and furthest from the current SVM solution. The convex hull sample, which disagrees most with the SVM solution, will lie on the convex hull of each class distribution and all support vectors lie on the convex hull in the case of linearly separable classes. Two difficulties have to be overcome. The first is that the SeqSVM's iterations will take too many if there are too many support vectors. The second is that when the class distributions are not separable, it is not easy to pick convex hull samples. In this paper, we show how these two difficulties are overcome. Experimental results on both artificial database and benchmark databases demonstrated the effectiveness of proposed algorithm to reduce the learning time of SVM on the whole training set.

In this article, we focus on designing an algorithm, named AdaBoostSVM, using SVM as weak learners for AdaBoost. To obtain a set of effective SVM weak learners, this algorithm adaptively adjusts the kernel parameter in SVM instead of using a fixed one. Compared with the existing AdaBoost methods, the AdaBoostSVM has advantages of easier model selection and better generalization performance. It also provides a possible way to handle the over-fitting problem in AdaBoost. An improved version called Diverse AdaBoostSVM is further developed to deal with the accuracy/diversity dilemma in Boosting methods. By implementing some parameter adjusting strategies, the distributions of accuracy and diversity over these SVM weak learners are tuned to achieve a good balance. To the best of our knowledge, such a mechanism that can conveniently and explicitly balances this dilemma has not been seen in the literature. Experimental results demonstrated that both proposed algorithms achieve better generalization performance than AdaBoost using other kinds of weak learners. Benefiting from the balance between accuracy and diversity, the Diverse AdaBoostSVM achieves the best performance. In addition, the experiments on unbalanced data sets showed that the AdaBoostSVM performed much better than SVM.

Optic disk is one of the main components on retina. It is an indicator of various ophthalmic pathologies. This paper presents a novel algorithm to segment the optic disk, based on snake model. The proposed method improves and extends original snake technique in two aspects: clustering and smoothing update. The contour first deforms to the location with the minimal energy, and then self-separates into edge-point group or uncertain-point group by means of weighted k-means algorithm. The contour points are finally updated by variable updating sample numbers. These modifications directly solve the blood vessel problem, which has not been effectively solved in the earlier work on disk boundary detection up to now. The comparative results on the 100 testing images shown that the proposed method achieves better success rate (94%) when compared to those obtained by GVF-snake (12%) and modified ASM method (82%).

This paper presents a method that matches interest point features detected on two images taken from different view points. A new multi-scale Plessey corner detector (MPCD) is used to detect the interest points. The geometric constraint between two images is exploited as in [R. Deriche, et al., 1994]: the fundamental matrix is derived using least median of squares (LMedS) from an initial set of matches, and then it is used to guide new matches. However, we propose a new energy function that can approximate affine transformation in a more effective way. Then, the initial set of matches is derived by minimizing the energy function. As a result, our method can perform well even when the pose variation is large between the two images. We compare our method using the proposed MPCD against two standard corner detectors on image matching. Also we evaluate our proposed matching criterion against Zhang's. Our method gave better results in both experiments on real face images.

LDA (linear discriminant analysis) and its variants are popular for image-based classification problems such as face recognition. However, their performance is inherently unstable when the samples are sparse. We propose a new type of discriminant analysis called MMDA (margin-maximization discriminant analysis), which derives features by maximizing the average margin between the classes. The method does not require S/sub W/ (within-class scatter matrix) to be non-singular and well-conditioned as it does not involve its inverse term, and the features can be directly derived from the input space. A computational trick has also been proposed for MMDA to handle high-dimensional data. We conducted intensive tests on ORL and UMIST face databases, and the results show that MMDA is a good replacement of LDA for the sparse sample problem.

Image classification is an important task in computer vision. However, how to assign suitable labels to images is a subjective matter, especially when some images can be categorized into multiple classes simultaneously. Multilabel image classification focuses on the problem that each image can have one or multiple labels. It is known that manually labelling images is time-consuming and expensive. In order to reduce the human effort of labelling images, especially multilabel images, we proposed a multilabel SVM active learning method. We also proposed two selection strategies: Max Loss strategy and Mean Max Loss strategy. Experimental results on both artificial data and real-world images demonstrated the advantage of proposed method.

This paper proposes a multiscale corner detection method for gray level images based on scale-space theory and Plessey operator. The proposed method solves three problems existing in the original Plessey detector. First, it works in the scale-space domain, so it detects corners belonging to different scales instead of a certain scale. Second, only one parameter needs to be set instead of three parameters needed in the original Plessey method. Third, delocalization is a well-known inherent drawback of the Plessey corner operator and it will increase with the scale at which it operates. The proposed algorithm solves the problem by detecting the corners from small scale to large scale, then track back from large scale to small scale. As the delocalization in the smallest scale can be ignored, the proposed method obtain the accurate localization. This proposed multiscale scheme can also be applied to other spatial corner detectors to improve their performances. The simulation results and the application in stereo matching show the improved performance of the proposed method compared with the original Plessey detector and the SUSAN detector

This paper presents a new 3D face model of a specific person constructed from the anatomical perspective. By exploiting the laser range data, a 3D facial mesh precisely representing the skin geometry is reconstructed. Based on the geometric facial mesh, we develop a deformable multi-layer skin model. It takes into account the nonlinear stress-strain relationship and dynamically simulates the non-homogenous behavior of the real skin. The face model also incorporates a set of anatomically-motivated facial muscle actuators and underlying skull structure. Lagrangian mechanics governs the facial motion dynamics, dictating the dynamic deformation of facial skin in response to the muscle contraction.

A method is proposed to register visible and infrared face images. The vanishing-point based approach [J.G. Wang and E. Sung, 2001] is applied to the visible image to determine the 3D pose of the human head. Then the corresponding pose with respect to the infrared camera is computed through the known relationship by calibration between the visible and infrared cameras. By doing so, the skin temperature range within the infrared image can be superimposed over the visible face image. We use the EM strategy to first compute the 3D pose using some initially learned (PCA) model parameters, and then update iteratively the parameters for individual persons and their facial expressions till convergence. The EM technique models data uncertainty using Gaussian mixtures defined over positions and orientation of facial plane. The resulting weighted parameters estimation problem is solved using the Levenberg-Marquardt method. The results on the synthetic data and real images have verified the performance of the approach.

In this paper we propose an individualized 3D facial model constructed from the anatomical perspective for realistic facial expression synthesis. Our model incorporates a physically-based approximation to facial skin tissue, a set of anatomically-motivated facial muscle actuators and underlying skull structure. The skin is modeled as a multi-layer mass-spring lattice by taking into account the nonlinear stress-strain relationship of the real skin. The 3D face model incorporates a skull structure which extends the scope of facial motion and facilitates facial muscle definition. The automatic facial muscle construction is achieved by using an efficient muscle mapping approach that ensures different muscles to be located at the anatomically correct positions between the skin and skull layers. For computational efficiency, we devise a semi-implicit integration method to numerically solve the governing dynamics. The dynamic facial animation algorithm runs at an interactive rate with realistic facial expressions to be synthesized.

In this paper, a new approach is proposed for estimating 3D head pose from a monocular image. The approach assumes the more difficult full perspective projection camera model as against most previous approaches that approximate the non-linear perspective projection via linear affine assumption. Perspective-invariance is used to estimate the head pose from a face image. Our approach employs a general prior knowledge of face structure and the corresponding geometrical constraints provided by the location of a certain vanishing point to determine the pose of human faces. To achieve this, eye-lines, formed from the far and near eye corners, and mouth-line of the mouth corners are assumed parallel in 3D space. Then the vanishing point of these parallel lines found by the intersection of the eye-line and mouth-line in the image can be used to infer the 3D orientation and location of the human face. Perspective invariance of cross ratio and harmonic range are used to locate the vanishing point stably. In order to deal with the variance of the facial model parameters, e.g. ratio between the eye-line and the mouth line, an EM framework is applied to update the parameters iteratively. We use the EM strategy to first compute the 3D pose using some initially learned (PCA) parameters, e.g. ratio and length, then update iteratively the parameters for individual persons and their facial expressions till convergence. The EM technique models data uncertainty as Gaussian defined over positions and orientation of facial plane. The resulting weighted parameters estimation problem is solved using the Levenberg-Marquardt method. The robustness analysis of the algorithm with synthetic data and real face images are included.

In this paper we propose a new hierarchical 3D facial model based on anatomical knowledge that provides high fidelity for realistic facial expression animation. Like real human face, the facial model has a hierarchical biomechanical structure, incorporating a physically-based approximation to facial skin tissue, a set of anatomically-motivated facial muscle actuators and underlying skull structure. The deformable skin model has multi-layer structure to approximate different types of soft tissue. It takes into account the nonlinear stress-strain relationship of the skin and the fact that soft tissue is almost incompressible. Different types of muscle models have been developed to simulate distribution of the muscle force on the skin due to muscle contraction. By the presence of the skull model, our facial model takes advantage of both more accurate facial deformation and the consideration of facial anatomy during the interactive definition of facial muscles. Under the muscular force, the deformation of the facial skin is evaluated using numerical integration of the governing dynamic equations. The dynamic facial animation algorithm runs at interactive rate with flexible and realistic facial expressions to be generated.

A new approach for estimating 3D head pose form a monocular image is proposed. Our approach employs general prior knowledge of face structure and the corresponding geometrical constraints provided by the location of vanishing point to determine pose of human faces. Eye-lines and mouth-line are assumed parallel in 3D space, and the vanishing point formed by the intersection of the eyeline and mouth-line in the image can be used to infer 3D orientation and location of human face. Perspective invariance of cross ratio and harmonic range is used to locate the vanishing point stably. The robustness analysis of the algorithm with synthesis data and real face images are enclosed.

We propose a physically-based approach based on anatomical knowledge for real-time facial expression animation. The facial model incorporates a physically-based approximation to facial skin and a set of anatomically-motivated facial muscles. The skin is modeled by a mass-spring system with nonlinear springs which have a biphasic stress-strain relationship to simulate the elastic dynamics of real facial skin. Facial muscles are modeled as forces deforming the spring mesh. Based on the action units (AUs) of the facial action coding system (FACS), various expressions can be generated by a combination of contractions of a set of facial muscles. Lagrangian mechanics governs the dynamics, dictating the deformation of the facial surface in response to muscle forces. Simulation results show real-time facial deformation as well as realistic expression animation.

A vision system capable of extracting features from a semi-structured environment for vehicle guidance is described in this paper. The system is primarily used for road following via the detection of mud tracks in a tropical environment. The scene captured by a CCD colour camera is digitised into 24-bit colour images with a resolution of 320x240 pixels. Partitioning of the scene into road and non-road areas is based on the results of a colour image segmentation algorithm applied to these images. The RGB colour images from the camera are converted to HSI format. Training samples of road and non-road features of the terrain to be explored, stored in a database, are used to classify blocks of pixels using only the hue information content of the images. A Bayesian classifier in conjunction with a smooth thresholding function is used for the segmentation algorithm on a per block basis. This approach results in the recognition of traversable areas, particularly non-metalled roads. Experimental results have showed that the algorithm is invariant to shadow conditions, i.e. roads were detected under varying light conditions. Due to the soil conditions of the test sites, small puddles of water on the mud tracks are also classified as driveable areas. The system outputs a one bit 2-D map of the image every 200ms. Field results of the proposed approach have shown favourable responses for real-time implementation on an autonomous ground vehicle.

This paper proposes a method of implementing stereovision in existing mobile robot perception systems without having to solve the classical stereo correspondence problem. The proposed method makes use of the fact that tracking of features is an easier task than performing stereo correspondence. A two-view motion stereo system is implemented in which the correspondence between two cameras is found by temporal stereo tracking. The algorithm assumes an initial stereo match of homologous features seen in the two cameras but subsequently the algorithm reliably tracks each feature by use of accurate odometric information, and accurate 3-D information can be derived from each matched feature. The derived set of 3-D data (of points, lines, and regions) are then fed to a fast surface reconstruction algorithm for later stage interpretation. The proposed scheme can be added to existing mobile robot road following systems at little added cost.© (1992) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

theseapproaches yield some percentage of incorrect matches. It should be realised thatany criterion based on image properties will not be robust since the image capturecontains a great deal of "disturbances" in the form of uncontrolled illumination,shadows, occlusions, noise, etc. Of fundamental interest is whether there exist adeterministic property that can be applied in general to validate a match? Ayache,Faugeras and Zhang have presented a trinocular stereovision algorithm to match 2???Dedge segments.

This paper presents a semi—automatic method to calibrate a three—camera stereovision set-up accurately and simply. The algorithm handles both initial calibration and on—line self—calibration and assumes the availability of a reliable set of matched image points. No 3—D world coordinate control points are available to confirm the validity of the self—calibration. A new general algorithm is introduced based on the minimum distance measure which provides the necessary criterion to make self— calibration possible. Convergence is obtained for small changes in camera parameter by optimization of the minimum distances over the 22 parametric space. Ill—conditioning problem is solved by a partitioning method. Comparisons with SYD approach are made. Synthetic tests were performed and the method proposed gave satisfactory results.

Machine Learning methods have been widely used in recent years in many areas, such as object recognition, autonomous vehicle, recovery rate of symptoms and stock prediction. However, it is very difficult task to make a computer learn from experience as human can. Machine learning has been in the field of research for more than five decades since 1950s. It has become a challenging task to create the method that able to make machines can learn faster. Master of Science (Computer Control and Automation)