Showing papers on "Dynamic time warping published in 2006"

Fast time series classification using numerosity reduction

Abstract: Many algorithms have been proposed for the problem of time series classification However, it is clear that one-nearest-neighbor with Dynamic Time Warping (DTW) distance is exceptionally difficult to beat This approach has one weakness, however; it is computationally too demanding for many realtime applications One way to mitigate this problem is to speed up the DTW calculations Nonetheless, there is a limit to how much this can help In this work, we propose an additional technique, numerosity reduction, to speed up one-nearest-neighbor DTW While the idea of numerosity reduction for nearest-neighbor classifiers has a long history, we show here that we can leverage off an original observation about the relationship between dataset size and DTW constraints to produce an extremely compact dataset with little or no loss in accuracy We test our ideas with a comprehensive set of experiments, and show that it can efficiently produce extremely fast accurate classifiers

LB_Keogh supports exact indexing of shapes under rotation invariance with arbitrary representations and distance measures

Abstract: The matching of two-dimensional shapes is an important problem with applications in domains as diverse as biometrics, industry, medicine and anthropology. The distance measure used must be invariant to many distortions, including scale, offset, noise, partial occlusion, etc. Most of these distortions are relatively easy to handle, either in the representation of the data or in the similarity measure used. However rotation invariance seems to be uniquely difficult. Current approaches typically try to achieve rotation invariance in the representation of the data, at the expense of discrimination ability, or in the distance measure, at the expense of efficiency. In this work we show that we can take the slow but accurate approaches and dramatically speed them up. On real world problems our technique can take current approaches and make them four orders of magnitude faster, without false dismissals. Moreover, our technique can be used with any of the dozens of existing shape representations and with all the most popular distance measures including Euclidean distance, Dynamic Time Warping and Longest Common Subsequence.

Chromatographic alignment of ESI-LC-MS proteomics data sets by ordered bijective interpolated warping.

Abstract: Mass spectrometry proteomics typically relies upon analyzing outcomes of single analyses; however, comparing raw data across multiple experiments should enhance both peptide/protein identification and quantitation In the absence of convincing tandem MS identifications, comparing peptide quantities between experiments (or fractions) requires the chromatographic alignment of MS signals An extension of dynamic time warping (DTW), termed ordered bijective interpolated warping (OBI-Warp), is presented and used to align a variety of electrospray ionization liquid chromatography mass spectrometry (ESI-LC-MS) proteomics data sets An algorithm to produce a bijective (one-to-one) function from DTW output is coupled with piecewise cubic hermite interpolation to produce a smooth warping function Data sets were chosen to represent a broad selection of ESI-LC-MS alignment cases High confidence, overlapping tandem mass spectra are used as standards to optimize and compare alignment parameters We determine that Pea

Indexing Multidimensional Time-Series

Abstract: While most time series data mining research has concentrated on providing solutions for a single distance function, in this work we motivate the need for an index structure that can support multiple distance measures. Our specific area of interest is the efficient retrieval and analysis of similar trajectories. Trajectory datasets are very common in environmental applications, mobility experiments, and video surveillance and are especially important for the discovery of certain biological patterns. Our primary similarity measure is based on the longest common subsequence (LCSS) model that offers enhanced robustness, particularly for noisy data, which are encountered very often in real-world applications. However, our index is able to accommodate other distance measures as well, including the ubiquitous Euclidean distance and the increasingly popular dynamic time warping (DTW). While other researchers have advocated one or other of these similarity measures, a major contribution of our work is the ability to support all these measures without the need to restructure the index. Our framework guarantees no false dismissals and can also be tailored to provide much faster response time at the expense of slightly reduced precision/recall. The experimental results demonstrate that our index can help speed up the computation of expensive similarity measures such as the LCSS and the DTW.

Discovering Characteristic Actions from On-Body Sensor Data

Abstract: We present an approach to activity discovery, the unsupervised identification and modeling of human actions embedded in a larger sensor stream. Activity discovery can be seen as the inverse of the activity recognition problem. Rather than learn models from hand-labeled sequences, we attempt to discover motifs, sets of similar subsequences within the raw sensor stream, without the benefit of labels or manual segmentation. These motifs are statistically unlikely and thus typically correspond to important or characteristic actions within the activity. The problem of activity discovery differs from typical motif discovery, such as locating protein binding sites, because of the nature of time series data representing human activity. For example, in activity data, motifs will tend to be sparsely distributed, vary in length, and may only exhibit intra-motif similarity after appropriate time warping. In this paper, we motivate the activity discovery problem and present our approach for efficient discovery of meaningful actions from sensor data representing human activity. We empirically evaluate the approach on an exercise data set captured by a wrist-mounted, three-axis inertial sensor. Our algorithm successfully discovers motifs that correspond to the real exercises with a recall rate of 96.3% and overall accuracy of 86.7% over six exercises and 864 occurrences.

Similarity search on time series based on threshold queries

Abstract: Similarity search in time series data is required in many application fields. The most prominent work has focused on similarity search considering either complete time series or similarity according to subsequences of time series. For many domains like financial analysis, medicine, environmental meteorology, or environmental observation, the detection of temporal dependencies between different time series is very important. In contrast to traditional approaches which consider the course of the time series for the purpose of matching, coarse trend information about the time series could be sufficient to solve the above mentioned problem. In particular, temporal dependencies in time series can be detected by determining the points of time at which the time series exceeds a specific threshold. In this paper, we introduce the novel concept of threshold queries in time series databases which report those time series exceeding a user-defined query threshold at similar time frames compared to the query time series. We present a new efficient access method which uses the fact that only partial information of the time series is required at query time. The performance of our solution is demonstrated by an extensive experimental evaluation on real world and artificial time series data.

Exploiting the Analogy Between Traces and Signal Processing

Abstract: The main challenge of dynamic analysis is the huge volume of data, making it difficult to extract high level views. Most techniques developed so far adopt a finegrained approach to address this issue. In this paper we introduce a novel approach representing entire traces as signals in time. Drawing this analogy between dynamic analysis and signal processing, we are able to access a rich toolkit of well-established and ready-to-use analysis techniques. As an application of this analogy, we show how to fit a visualization of the complete feature space of a system on one page only: our approach visualizes feature traces as time plots, summarizes the trace signals and uses Dymanic Time Warping to group them by similar features. We apply the approach on a case study, and discuss both common and unique patterns as observed on the visualization.

On-Line Handwritten Character Recognition with 3D Accelerometer

Abstract: We present a pen-style hardware and analysis software for the recognition of handwritten characters. The hardware has a 3-dimensional acceleration sensor, an amplifier, a microcontroller with AD converter and communication port, and does not need any touching screen. The algorithm software includes signal pre-processing, feature extraction, and classifier. Both Hidden Markov Model (HMM) and Dynamic Time Warping (DTW) algorithms are implemented. For the experiments with 10 Arabic numerals the hardware and software system shows very high recognition rates, i.e., 100% and 90.8% for the writer-dependent and writer-independent cases, respectively. Although the database is quite small, it clearly demonstrates the usefulness of the acceleration-based handwritten character recognition system without touching screen or pad.

Automatic splicing for hand and body animations

Abstract: We propose a solution to a new problem in animation research: how to use human motion capture data to create character motion with detailed hand gesticulation without the need for the simultaneous capture of hands and the full-body. Occlusion and a difference in scale make it difficult to capture both the detail of the hand movement and unrestricted full-body motion at the same time. With our method, the two can be captured separately and spliced together seamlessly with little or no user input required. The algorithm relies on a novel distance metric derived from research on gestures and uses a two-pass dynamic time warping algorithm to find correspondence between the hand and full-body motions. In addition, we provide a method for supplying user input, useful to animators who want more control over the integrated animation. We show the power of our technique with a variety of common and highly specialized gesticulation examples.

Time Warped Modified Transform Coding of Audio Signals

Abstract: A spectral representation of an audio signal having consecutive audio frames can be derived more efficiently, when a common time warp is estimated for any two neighbouring frames, such that a following block transform can additionally use the warp information. Thus, window functions required for successful application of an overlap and add procedure during reconstruction can be derived and applied, the window functions already anticipating the re-sampling of the signal due to the time warping. Therefore, the increased efficiency of block-based transform coding of time-warped signals can be used without introducing audible discontinuities.

Fast structuring of large television streams using program guides

Abstract: An original task of structuring and labeling large television streams is tackled in this paper. Emphasis is put on simple and efficient methods to detect precise boundaries of programs. These programs are further analysed and labeled with information coming from a standard television program guide using an improved Dynamic Time Warping algorithm (DTW) and a manually labeled reference video dataset. It is shown that the labeling process yields a very high accuracy and opens the way to many applications. We eventually indicate how the dependency to a manually labeled video dataset can be removed by providing an algorithm for a dynamic update of the reference video dataset.

Template Based Micro-Doppler Signature Classification

Abstract: The micro-Doppler signature of a target is a time varying frequency modulation imparted on the radar echo signal by moving components of the target. Battlefield radar output the baseband signal as audio and soldiers listening on headphones are able to identify the target from its micro-Doppler signature. Automation of this capability is desirable for improved reliability and reduction in classification time. For the first time dynamic time warping (DTW), a speech recognition technique, has been applied to the problem. Its performance has been compared with the common k-nearest neighbour (k-NN) classification method since both approaches utilise a template library.

A Time Warping Based Approach for Video Copy Detection

Abstract: The proliferation of digital video urges the need of video copy detection for content and rights management. An efjcient video copy detection technique should be able to deal with spatiotemporal variations (e.g., changes in brightness or frame rates), and lower down the computation cost. m i l e most studies put more emphases on spatial variations, less effort is made for temporal variations and computation cost. To address the above issues, we propose a time warping based approach for video copy detection. A time warping matching algorithm is used to deal with video temporal variations. To reduce matching times, a fast Pltering method to generate key parries and select candidate clips from video is presented. Our experiments demonstrate promising results of the proposed approach.

Classification of vocalizations of killer whales using dynamic time warping

Abstract: A large number of killer whale sounds have recently been classified perceptually into Call Types. [A. Hodgins-Davis, thesis, Wellesley College (2004)]. The repetition rate of the pulsed component of five or more examples of each call type has been calculated using a modified form of the FFT based comb-filter method. A dissimilarity or distance matrix for these sounds was calculated using dynamic time warping to compare their melodic contours. These distances were transformed into a component space using multidimensional scaling and the resulting points were clustered with a kmeans algorithm. In grouping 57 sounds into 9 call types, a single discrepancy between the perceptual and the automated methods occurred.

Automatic identification of bird calls using Spectral Ensemble Average Voice Prints

Abstract: Automatic identification of bird calls without manual intervention has been a challenging task for meaningful research on the taxonomy and monitoring of bird migrations in ornithology. In this paper we apply several techniques used in speech recognition to the automatic identification of bird calls. A new technique which computes the ensemble average on the FFT spectrum is proposed for identification of bird calls. This ensemble average is computed on the FFT spectrum of each bird and is called the Spectral Ensemble Average Voice Print (SEAV) of that particular bird. The SEAV of various birds are computed and are found to be different when compared to each other. A database of bird calls is created from the available recordings of fifteen bird species. The SEAV is then used for the identification of bird calls from this database. The results of identification using SEAV are then compared against the results derived from common classifiers used in speech recognition like dynamic time warping (DTW), Gaussian mixture modeling (GMM). A one level and two level classifier combination is also tried by combining SEAV classifier with the DTW classifier. The SEAV is computationally less expensive when compared to DTW or the GMM based classifiers while performing better than the DTW technique. Several new possibilities in automatic bird call identification using SEAV are also listed.

Segmental Hidden Markov Models with Random Effects for Waveform Modeling

Abstract: This paper proposes a general probabilistic framework for shape-based modeling and classification of waveform data. A segmental hidden Markov model (HMM) is used to characterize waveform shape and shape variation is captured by adding random effects to the segmental model. The resulting probabilistic framework provides a basis for learning of waveform models from data as well as parsing and recognition of new waveforms. Expectation-maximization (EM) algorithms are derived and investigated for fitting such models to data. In particular, the "expectation conditional maximization either" (ECME) algorithm is shown to provide significantly faster convergence than a standard EM procedure. Experimental results on two real-world data sets demonstrate that the proposed approach leads to improved accuracy in classification and segmentation when compared to alternatives such as Euclidean distance matching, dynamic time warping, and segmental HMMs without random effects.

Identifying Cover Songs from Audio Using Harmonic Representation

Abstract: This extended abstract describes in detail a submission to the task on Audio Cover Song in the Music Information Retrieval eXchange in 2006. The system uses as feature set a chord sequence identified by an HMM trained with audiofrom-symbolic data, and computes a distance between two chord sequence pair using the Dynamic Time Warping algorithm to find the minimum alignment cost. The rational behind the system is that cover songs largely preserve harmonic content even if they vary in other musical attributes such as instrumentation, tempo, key, and/or melody.

Automatic splicing for hand and body animations

Abstract: We propose a solution to a new problem in animation research: how to use human motion capture data to create character motion with detailed hand gesticulation without the need for the simultaneous capture of hands and the full-body. Occlusion and a difference in scale make it difficult to capture both the detail of the hand movement and unrestricted full-body motion at the same time. With our method, the two can be captured separately and spliced together seamlessly with little or no user input required. The algorithm relies on a novel distance metric derived from research on gestures and uses a two-pass dynamic time warping algorithm to find correspondence between the hand and full-body motions. In addition, we provide a method for supplying user input, useful to animators who want more control over the integrated animation. We show the power of our technique with a variety of common and highly specialized gesticulation examples.

Aligning transcripts to automatically segmented handwritten manuscripts

Abstract: Training and evaluation of techniques for handwriting recognition and retrieval is a challenge given that it is difficult to create large ground-truthed datasets. This is especially true for historical handwritten datasets. In many instances the ground truth has to be created by manually transcribing each word, which is a very labor intensive process. Sometimes transcriptions are available for some manuscripts. These transcriptions were created for other purposes and hence correspondence at the word, line, or sentence level may not be available. To be useful for training and evaluation, a word level correspondence must be available between the segmented handwritten word images and the ASCII transcriptions. Creating this correspondence or alignment is challenging because the segmentation is often errorful and the ASCII transcription may also have errors in it. Very little work has been done on the alignment of handwritten data to transcripts. Here, a novel Hidden Markov Model based automatic alignment algorithm is described and tested. The algorithm produces an average alignment accuracy of about 72.8% when aligning whole pages at a time on a set of 70 pages of the George Washington collection. This outperforms a dynamic time warping alignment algorithm by about 12% previously reported in the literature and tested on the same collection.

Biologic verification based on pressure sensor keyboards and classifier fusion techniques

Abstract: This paper presents a novel biologic verification method based on pressure sensor keyboards and classifier fusion techniques. The pressure sensor keyboard is a new product that occurs in the market recently. It produces a pressure sequence when keystroke occurs. The analysis of the pressure sequence should be a novel research area. In this paper, we use the pressure sequence and traditional keystroke dynamics in user authentication. Three methods (global features of pressure sequences, dynamic time warping, and traditional keystroke dynamics) are proposed for the authentication task. We combined the three methods together using a classifier fusion technique at last. Several experiments were performed on a database containing 5000 samples of 100 individuals and the best result were achieved utilizing all the method, obtaining an equal error rate of 1.41%. To make a comparison, the equal error rate is 2.04% when we use only the traditional keystroke dynamics. This approach can be used to improve the usual login-password authentication when the password is no more a secret

Time Warping Invariant Echo State Networks

Abstract: Echo State Networks (ESNs) is a recent simple and powerful approach to training recurrent neural networks (RNNs). In this report we present a modification of ESNs - time warping invariant echo state networks (TWIESNs) that can effectively deal with time warping in dynamic pattern recognition. The standard approach to classify time warped input signals is to align them to candidate pro- totype patterns by a dynamic programming method and use the alignment cost as a classification criterion. In contrast, we feed the original input signal into specifically designed ESNs which intrinsically are invariant to time warping in the input. For this purpose, ESNs with leaky integrator neurons are required, which are here presented for the first time, too. We then explain the TWIESN architecture and demonstrate their functioning on very strongly warped, synthetic data sets.

Comparison of Sequence Discriminant Support Vector Machines for Acoustic Event Classification

Abstract: In a previously reported work, classification techniques based on Support Vector Machines (SVM) showed a good performance in the task of acoustic event classification. SVM are discriminant classifiers, but they cannot easily deal with the dynamic time structure of sounds, since they are constrained to work with fixed-length vectors. Several methods that adapt SVM to sequence processing have been reported in the literature. In this paper, they are reviewed and applied to the classification of 16 types of sounds from the meeting room environment. With our database, we have observed that the dynamic time warping kernels work well for sounds that show a temporal structure, but the best average score is obtained with the Fisher kernel.

Gene Time Eχpression Warper: a tool for alignment, template matching and visualization of gene expression time series

Abstract: Summary: An application tool for alignment, template matching and visualization of gene expression time series is presented. The core algorithm is based on dynamic time warping techniques used in the speech recognition field. These techniques allow for non-linear (elastic) alignment of temporal sequences of feature vectors and consequently enable detection of similar shapes with different phases. Availability: The Java program, examples and a tutorial are available at http://www.psb.ugent.be/cbd/papers/gentxwarper/ Contact: eltsi@psb.ugent.be

Micro-Doppler Signature Classification

Abstract: The micro-Doppler signature of a target is a time varying frequency modulation imparted on the radar echo signal by moving components of the target Battlefield radar output the radar's baseband signal as audio and soldiers listening on headphones are able to identify the target from its micro-Doppler signature Automation of this capability is desirable for improved reliability and reduction in classification time For the first time dynamic time warping (DTW), a speech recognition technique, has been applied to the problem

Discovering patterns in real-valued time series

Abstract: This paper describes an algorithm for discovering variable length patterns in real-valued time series. In contrast to most existing pattern discovery algorithms, ours does not first discretize the data, runs in linear time, and requires constant memory. These properties are obtained by sampling the data stream rather than processing all of the data. Empirical results show that the algorithm performs well on both synthetic and real data when compared to an exhaustive algorithm.

Combining feature selection and DTW for time-varying functional genomics

Abstract: Given temporal high-throughput data defining a two-class functional genomic process, feature selection algorithms may be applied to extract a panel of discriminating gene time series. We aim to identify the main trends of activity through time. A reconstruction method based on stagewise boosting is endowed with a similarity measure based on the dynamic time warping (DTW) algorithm, defining a ranked set of time-series component contributing most to the reconstruction. The approach is applied on synthetic and public microarray data. On the Cardiogenomics PGA Mouse Model of Myocardial Infarction, the approach allows the identification of a time-varying molecular profile of the ventricular remodeling process.

Using visual motifs to classify encrypted traffic

Abstract: In an effort to make robust traffic classification more accessible to human operators, we present visualization techniques for network traffic. Our techniques are based solely on network information that remains intact after application-layer encryption, and so offer a way to visualize traffic "in the dark". Our visualizations clearly illustrate the differences between common application protocols, both in their transient (i.e., time-dependent)and steady-state behavior. We show how these visualizations can be used to assist a human operator to recognize application protocols in unidentified traffic and to verify the results of an automated classifier via visual inspection. In particular, our preliminary results show that we can visually scan almost 45,000 connections in less than one hour and correctly identify known application behaviors. Moreover, using visualizations together with an automated comparison technique based on Dynamic Time Warping of the motifs, we can rapidly develop accurate recognizers for new or previously unknown applications.