Real-time classification of variable length multi-attribute motions

Multi-attribute motion data can be generated in many applications/ devices, such as motion capture devices and animations. It can have dozens of attributes, thousands of rows, and even similar motions can have different durations and different speeds at corresponding parts. There are no row-to-row correspondences between data matrices of two motions. To be classified and recognized, multi-attribute motion data of different lengths are reduced to feature vectors by using the properties of singular value decomposition (SVD) of motion data. The reduced feature vectors of similar motions are close to each other, while reduced feature vectors are different from each other if their motions are different. By applying support vector machines (SVM) to the feature vectors, we efficiently classify and recognize real-world multi-attribute motion data. With our data set of more than 300 motions with different lengths and variations, SVM outperforms classification by related similarity measures, in terms of accuracy and CPU time. The performance of our approach shows its feasibility of real-time applications to real-world data. Chuanjun Li is a Ph.D. candidate in Computer Science at the University of Texas at Dallas. His Ph.D. research works primarily on efficient segmentation and recognition of human motion streams, and development of indexing and clustering techniques for the multi-attribute motion data as well as classification of motion data. Dr. Latifur R. Khan has been an Assistant Professor of Computer Science Department at University of Texas at Dallas since September, 2000. He received his Ph.D. and M.S. degree in Computer Science from University of Southern California (USC) in August 2000 and December 1996, respectively. He obtained his B.Sc. degree in Computer Science and Engineering from Bangladesh University of Engineering and Technology, Dhaka, Bangladesh in November 1993. Professor Khan is currently supported by grants from the National Science Foundation (NSF), Texas Instruments, NOKIA, Alcatel, USA and has been awarded the Sun Equipment Grant. Dr. Khan has more than 50 articles, book chapters, and conference papers focusing in the areas of: database systems, multimedia information management, and data mining in bio-informatics and intrusion detection. Professor Khan has also served as a referee for database journals, conferences (e.g., IEEE TKDE, KAIS, ADL, VLDB) and he is currently serving as a program committee member for Eleventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (SIGKDD2005), ACM Fourteenth Conference on Information and Knowledge Management (CIKM 2005), International Conference on Database and Expert Systems Applications DEXA 2005, and International Conference on Cooperative Information Systems (CoopIS 2005), and program chair of ACM SIGKDD International Workshop on Multimedia Data Mining, 2004. Dr. Balakrishnan Prabhakaran is currently with the Department of Computer Science, University of Texas at Dallas. Dr. B. Prabhakaran has been working in the area of multimedia systems: multimedia databases, authoring & presentation, resource management, and scalable web-based multimedia presentation servers. He has published several research papers in prestigious conferences and journals in this area.Dr. Prabhakaran received the NSF CAREER Award FY 2003 for his proposal on Animation Databases. Dr. Prabhakaran has served as an Associate Chair of the ACM Multimedia’2003 (November 2003, California), ACM MM 2000 (November 2000, Los Angeles), and ACM Multimedia’99 conference (Florida, November 1999). He has served as guest-editor (special issue on Multimedia Authoring and Presentation) for ACM Multimedia Systems journal. He is also serving on the editorial board of Multimedia Tools and Applications Journal, Kluwer Academic Publishers. He has also served as program committee member on several multimedia conferences and workshops. Dr. Prabhakaran has presented tutorials in several conferences on topics such as network resource management, adaptive multimedia presentations, and scalable multimedia servers.B. Prabhakaran has served as a visiting research faculty with the Department of Computer Science, University of Maryland, College Park. He also served as a faculty in the Department of Computer Science, National University of Singapore as well as in the Indian Institute of Technology, Madras, India     

Segmentation of blurred objects by classification of isolabel contours

Segmentation of objects with blurred boundaries is an important and challenging problem, especially in the field of medical image analysis. A new approach to segmentation of homogeneous blurred objects in grayscale images is described in this paper. The proposed algorithm is based on building of an isolabel-contour map of the image and classification of closed isolabel contours by the SVM. Each closed isolabel contour is described by the feature vector that can include intensity-based features of the image area enclosed by the contour, as well as geometrical features of the contour shape. The image labeling procedure for construction of the training base becomes very fast and convenient because it is reduced to clicking on isolabel contours delineating the objects of interest on the isolabel-contour map. The proposed algorithm was applied to the problem of brain lesion segmentation in MRI and demonstrated performance figures above 98% on real data, both in sensitivity and in specificity.     

Incorporating prior knowledge in support vector machines for classification: A review

For classification, support vector machines (SVMs) have recently been introduced and quickly became the state of the art. Now, the incorporation of prior knowledge into SVMs is the key element that allows to increase the performance in many applications. This paper gives a review of the current state of research regarding the incorporation of two general types of prior knowledge into SVMs for classification. The particular forms of prior knowledge considered here are presented in two main groups: class-invariance and knowledge on the data. The first one includes invariances to transformations, to permutations and in domains of input space, whereas the second one contains knowledge on unlabeled data, the imbalance of the training set or the quality of the data. The methods are then described and classified into the three categories that have been used in literature: sample methods based on the modification of the training data, kernel methods based on the modification of the kernel and optimization methods based on the modification of the problem formulation. A recent method, developed for support vector regression, considers prior knowledge on arbitrary regions of the input space. It is exposed here when applied to the classification case. A discussion is then conducted to regroup sample and optimization methods under a regularization framework.     

Human motion recognition using support vector machines

We propose a motion recognition strategy that represents each videoclip by a set of filtered images, each of which corresponds to a frame. Using a filtered-image classifier based on support vector machines, we classify a videoclip by applying majority voting over the predicted labels of its filtered images and, for online classification, we identify the most likely type of action at any moment by applying majority voting over the predicted labels of the filtered images within a sliding window. We also define a classification confidence and the associated threshold in both cases, which enable us to identify the existence of an unknown type of motion and, together with the proposed recognition strategy, make it possible to build a real-time motion recognition system that cannot only make classifications in real-time, but also learn new types of motions and recognize them in the future. The proposed strategy is demonstrated on real datasets.     

Single-channel SEMG using wavelet deep belief networks for upper limb motion recognition

Surface electromyography (SEMG) has been widely used in different fields such as human machine interaction and motion recognition. A hybrid classification model based on singular value decomposition (SVD) and wavelet deep belief networks (WDBN) is firstly proposed in this paper, which allows the machine to recognize the single-joint motions of upper limb by using one channel. In this experiment, the three-joint SEMG signals of upper limb are respectively recorded through different two channels, which are employed for subsequent comparison to obtain the best single-channel of each joint. Afterwards, the collected raw signals are enhanced by SVD processing. Wavelet function is applied to replace sigmoid function as activation function for feature learning, and the spectrum signals processed by fast Fourier transform (FFT) are input to WDBN model. The results demonstrate that the recognition rates of three joint movements can be up to 100% by SVD-WDBN method, which is much better than support vector machine (SVM), back propagation (BP) neural network and extreme learning machine (ELM) model. The proposed method makes it more possible to control wearable devices with different single-channel SEMG signals, thereby the work efficiency of smart wearable devices can be improved, as well as the complexity of operations between human and machine can be reduced.     

Mixture control chart patterns recognition using independent component analysis and support vector machine

Effective recognition of control chart patterns (CCPs) is an important issue since abnormal patterns exhibited in control charts can be associated with certain assignable causes which affect the process. Most of the existing studies assume that the observed process data which needs to be recognized are basic types of abnormal CCPs. However, in practical situations, the observed process data could be mixture patterns, which consist of two basic CCPs combined together. In this study, a hybrid scheme using independent component analysis (ICA) and support vector machine (SVM) is proposed for CCPs recognition. The proposed hybrid ICA-SVM scheme initially applies an ICA to the mixture patterns in order to generate independent components (ICs). The hidden basic patterns of the mixture patterns can be discovered in these ICs. The ICs can then serve as the input variables of the SVM for building a CCP recognition model. Experimental results revealed that the proposed scheme is able to effectively recognize mixture control chart patterns and outperform the single SVM models, which did not use an ICA as a preprocessor.     

支持向量机及其在模式识别中的应用

Statistical learning theory(SLT)and support vector machine(SVM) are effective to solve problems of machine learning under the condition of finite samples.It is known that the performance of support vector machine is often better than that of some neural networks in pattern recognition,especially in high dimensional space,and they arewell used in many domains for recognition.This paper at first introduces the basic theory of SLT and SVM,then points out the key problems of SVM and its research situation in recent years,and at last describes some applications of SVM in the field of pattern recognition.     

Online training of support vector classifier

Support vector machine (SVM) provides accurate classification but suffers from a large amount of computation. This paper presents an online support vector classifier (OSVC) for the pattern classification problems that have input data supplied in sequence rather than in batch. The OSVC has been applied to three benchmark problems: Iris data classification, image segmentation and numerical pattern recognition. The results obtained from the wide range of benchmark problems show that the OSVC algorithm has a much faster convergence and results in a smaller number of support vectors for the same quality of pattern classification and a better generalization performance in comparison with the existing algorithms.     

在原始空间用Rosenbrock算法训练线性支持向量机

为了加快并行下降方法（ＣＤ）用于线性支持向量机（ＳＶＭ）时的最终收敛速度,将Ｒｏｓｅｎｂｒｏｃｋ算法（Ｒ）用于线性ＳＶＭ．在内循环,Ｒ 通过解一个单变量子问题来更新狑的一个分量,并同时固定其他分量不变;在外循环,采用Ｇｒａｍ-ｓｃｈｍｉｄｔ过程构建新的搜索方向．实验结果表明,与ＣＤ 相比,Ｒ 加快了最终的收敛,在分类中能更快地获得更高的测试精度．     

Leave one support vector out cross validation for fast estimation of generalization errors

A Support Vector Classifier (SVC) is formulated in terms of a kernel. The bandwidth of the kernel affects the generalization performance of the SVC. This paper presents a Leave One Support Vector Out Cross Validation (LOSVO-CV) algorithm for estimating the optimal bandwidth of the kernel for classification purpose. The proposed algorithm is based on the Leave One Out Cross Validation (LOO-CV) algorithm (Numer. Math. 31 (1979) 377) that was proposed to find the optimal bandwidth but difficult to be implemented due to its large amount of computation. The properties of LOSVO-CV are analyzed in comparison with the LOO-CV. The simulation study demonstrates that the LOSVO-CV is a fast algorithm and it has the same generalization performance optimized by a bootstrap method (Neural Process. Lett. 11 (2000) 51) which can find an optimal bandwidth of the kernel of the SVC. The LOSVO-CV algorithm is able to provide consistent results with different sizes of a benchmark data set which is obtained from the University of California (UCI) repository.     

基于支持向量机的音频分类与分割

音频分类与分割是提取音频结构和内容语义的重要手段,是基于内容的音频、视频检索和分析的基础。支持向量机(SVM)是一种有效的统计学习方法。本文提出了一种基于SVM的音频分类算法。将音频分为5类：静音、噪音、音乐、纯语音和带背景音的语音。在分类的基础上,采用3个平滑规则对分类结果进行平滑。分析了SVM分类嚣的分类性能,同时也评估了本文提出的新的音频特征在SVM分类嚣上的分类效果。实验结果显示,基于SVM的音频分类算法分类效果良好,平滑处理后的音频分割结果比较准确。     

支持向量机与奇异值分解的盲水印算法

为进一步提高水印算法的抗攻击性能,提出了基于支持向量机(Support Vector Machine,SVM)与奇异值分解(Singular Value Decomposition,SVD)的盲水印算法。首先对宿主图像进行DWT变换,将低频子带分成互不重叠的子块;然后利用SVM建立子块的局部相关性模型,根据模型预测结果与对应位置的低频系数值的大小关系产生特征序列,该序列与水印进行异或运算产生特征水印序列,将特征水印序列通过奇偶量化规则嵌入原始图像小波低频子带对应子块的最大奇异值。实验结果表明,该算法不仅具有较好的不可感知性,而且具有较强的抗攻击能力。     

A modified support vector machine and its application to image segmentation

Recently, researchers are focusing more on the study of support vector machine (SVM) due to its useful applications in a number of areas, such as pattern recognition, multimedia, image processing and bioinformatics. One of the main research issues is how to improve the efficiency of the original SVM model, while preventing any deterioration of the classification performance of the model. In this paper, we propose a modified SVM based on the properties of support vectors and a pruning strategy to preserve support vectors, while eliminating redundant training vectors at the same time. The experiments on real images show that (1) our proposed approach can reduce the number of input training vectors, while preserving the support vectors, which leads to a significant reduction in the computational cost while attaining similar levels of accuracy. (2)The approach also works well when applied to image segmentation.     

Model-based segmentation and recognition of dynamic gestures in continuous video streams

Segmentation and recognition of continuous gestures are challenging due to spatio-temporal variations and endpoint localization issues. A novel multi-scale Gesture Model is presented here as a set of 3D spatio-temporal surfaces of a time-varying contour. Three approaches, which differ mainly in endpoint localization, are proposed: the first uses a motion detection strategy and multi-scale search to find the endpoints; the second uses Dynamic Time Warping to roughly locate the endpoints before a fine search is carried out; the last approach is based on Dynamic Programming. Experimental results on two arm and single hand gestures show that all three methods achieve high recognition rates, ranging from 88% to 96% for the two arm test, with the last method performing best.     

Combined VMD-SVM based feature selection method for classification of power quality events

Power quality (PQ) issues have become more important than before due to increased use of sensitive electrical loads. In this paper, a new hybrid algorithm is presented for PQ disturbances detection in electrical power systems. The proposed method is constructed based on four main steps: simulation of PQ events, extraction of features, selection of dominant features, and classification of selected features. By using two powerful signal processing tools, i.e. variational mode decomposition (VMD) and S-transform (ST), some potential features are extracted from different PQ events. VMD as a new tool decomposes signals into different modes and ST also analyzes signals in both time and frequency domains. In order to avoid large dimension of feature vector and obtain a detection scheme with optimum structure, sequential forward selection (SFS) and sequential backward selection (SBS) as wrapper based methods and Gram–Schmidt orthogonalization (GSO) based feature selection method as filter based method are used for elimination of redundant features. In the next step, PQ events are discriminated by support vector machines (SVMs) as classifier core. Obtained results of the extensive tests prove the satisfactory performance of the proposed method in terms of speed and accuracy even in noisy conditions. Moreover, the start and end points of PQ events can be detected with high precision.     

Human motions representation and recognition by directional motion history images

In this paper, we describe a technique for representing and recognizing human motions using directional motion history images. A motion history image is a single human motion image produced by superposing binarized successive motion image frames so that older frames may have smaller weights. It has, however, difficulty that the latest motion overwrites older motions, resulting in inexact motion representation and therefore incorrect recognition. To overcome this difficulty, we propose directional motion history images which describe a motion with respect to four directions of movement, i.e. up, down, right and left, employing optical flow. The directional motion history images are thus a set of four motion history images defined on four optical flow images. Experimental results show that the proposed technique achieves better performance in the recognition of human motions than the existent motion history images. This work was presented in part at the 13th International Symposium on Artificial Life and Robotics, Oita, Japan, January 31–February 2, 2008     

Multi-resolution form of SVD for text-independent speaker recognition

A text-independent speaker recognition system based on multi-resolution singular value decomposition (MSVD) is proposed. The MSVD is applied to the speaker data compression and feature extraction not at the square matrix. Our results have shown that this MSVD introduced better performance than the other Karhunen-Loeve transform with respect to the percentages of recognition.     

Real-time Unsupervised Segmentation of human whole-body motion and its application to humanoid robot acquisition of motion symbols

An interactive loop between motion recognition and motion generation is a fundamental mechanism for humans and humanoid robots. We have been developing an intelligent framework for motion recognition and generation based on symbolizing motion primitives. The motion primitives are encoded into Hidden Markov Models (HMMs), which we call “motion symbols”. However, to determine the motion primitives to use as training data for the HMMs, this framework requires a manual segmentation of human motions. Essentially, a humanoid robot is expected to participate in daily life and must learn many motion symbols to adapt to various situations. For this use, manual segmentation is cumbersome and impractical for humanoid robots. In this study, we propose a novel approach to segmentation, the Real-time Unsupervised Segmentation (RUS) method, which comprises three phases. In the first phase, short human movements are encoded into feature HMMs. Seamless human motion can be converted to a sequence of these feature HMMs. In the second phase, the causality between the feature HMMs is extracted. The causality data make it possible to predict movement from observation. In the third phase, movements having a large prediction uncertainty are designated as the boundaries of motion primitives. In this way, human whole-body motion can be segmented into a sequence of motion primitives. This paper also describes an application of RUS to AUtonomous Symbolization of motion primitives (AUS). Each derived motion primitive is classified into an HMM for a motion symbol, and parameters of the HMMs are optimized by using the motion primitives as training data in competitive learning. The HMMs are gradually optimized in such a way that the HMMs can abstract similar motion primitives. We tested the RUS and AUS frameworks on captured human whole-body motions and demonstrated the validity of the proposed framework.     

图像代数特征在面部表情识别中的应用

探讨了图像代数特征在面部表情识别中的应用,首先对面部表情图像进行了分割,得到眼睛和嘴巴区域,然后分别对眼睛和嘴巴区域提取不变矩和奇异值特征向量,并进行Fisher线性判别分析,最后训练了支持向量机分类器。实验结果表明该方法取得了比较好的识别效果。