There is an explosion of media content, especially user generated content, on the Internet. Efficient search of media data still remains a distant goal. Tagging has been used recently to bridge the "semantic gap" for images. Tagging of video data poses several challenges. The temporal nature of video requires support from the media player for video tagging. Unlike images which can be tagged simultaneously with viewing, tagging of video is typically not done at first viewing: videos are best tagged after viewing them at least once in entirety. This is not a preferred solution since the entire video has to be played again for tagging. In this paper, we propose a compact representation called "tagboard" which reduces the cognitive overload associated with video tagging. Tagboard also facilitates tag propagation. The technique is based on image similarity and dimensionality reduction. The ease of tagging improves metadata creation there by improving social search.

There is an explosion of multimedia content, especially image, on the Internet. Images as pixel arrays suffer from two deficiencies: semantic gap and geometric gap. There have been several attempts like object detection for closing the semantic gap. Recently the notion of community tagging of objects in an image has attracted a lot of attention. There has been lot of research in closing the geometric gap by 3D reconstruction. In this paper, we present 3D Buddy (3dB) -a system for geometrically tagging images for 3D reconstruction.

The widespread use of portable devices has led to the resurgence of speech interfaces. There is a crucial difference between the visual and auditory presentation of text. If text is presented visually, the user can skim and scroll through the content to locate relevant information. Audio output is usually presented sequentially with media player controls. For synthesized text a much more precise control is required. Actually the ability to skim and scroll through speech or auditory output will be crucial to the deployment speech output. Speakr is a speech synthesis system which permits different levels of skimming in speech outputs produced using speech synthesis. Speakr exploits the availability of the underlying text representation. Different aspects of the text representation - typographic conventions, markups, syntax, semantics, etc. - are used to achieve skim and scroll effects.

Mobile ad hoc networks (MANETs) are a very promising next generation networks. Prediction is an essential aspect in the deployment of MANETs. Current work mainly focuses on mobility prediction: prediction of spatio-temporal trajectories of nodes. Web page prefetching based on prediction of user's behavior has been shown to decrease latency in wired networks-current models of prefetching focus on what to prefetch and not when. Since routing information changes with time in MANETs, the expected time of user's access is also important. In this paper we integrate prefetching algorithms with mobility models and personalized to user browsing patterns. This not only reduces latency but also ensures just-in-time prefetching

Microsoft Windows uses the notion of registry to store all configuration information. The registry entries have associations and dependencies. For example, the paths to executables may be relative to some home directories. The registry being designed with faster access as one of the objectives does not explicitly capture these relations. In this paper, we explore a representation that captures the dependencies more explicitly using shared and unifying variables. This representation, called mRegistry exploits the tree-structured hierarchical nature of the registry, is concept-based and obtained in multiple stages. mRegistry captures intra-block, inter-block and ancestor-children dependencies (all leaf entries of a parent key in a registry put together as an entity constitute a block thereby making the block as the only child of the parent). In addition, it learns the generalized concepts of dependencies in the form of rules. We show that mRegistry has several applications: fault diagnosis, prediction, comparison, compression etc.

Wireless LANs (for e.g., IEEE 802.11g) and wireless personal access networks (WPANs) operate in the same 2.4 GHz ISM frequency band. In this paper, we consider the techniques to mitigate the interference caused by WPANs (e.g., Bluetooth) on the IEEE 802.11 wireless LAN. In particular, we consider the downlink where an adjacent Bluetooth device interferes with a WLAN station. We show that a simple frequency diversity scheme makes the IEEE 802.11 WLAN robust to the narrow band interference caused by a Bluetooth device. We also propose an interference cancellation algorithm which estimates the interference and reconstructs the WLAN signal (which is interference free). We compare our scheme with that of the existing architecture. We show that our scheme offers a significant gain compared to the existing architecture. Other techniques like error control coding may also be used for interference mitigation. But, coding does not exploit the inherent frequency diversity gain that we show in this paper. Hence, the performance of our scheme would be much better than what one would expect with coding techniques

Code division multiple access (CDMA) and orthogonal frequency division multiplexing (OFDM) are two techniques that are being commonly used in most wireless communication systems. In this paper we examine the coexistence of the above schemes through a two user communication system, i.e., one user transmits using CDMA and the other transmits using OFDM. Using an OFDM based receiver, we derive the received symbols at the output of each branch of the OFDM demodulator. The symbols transmitted by the CDMA user are then detected using the minimum power distortionless response (MPDR) approach, an array processing technique commonly used in beamforming. Through simulations, we then show that the MPDR method is indeed very efficient in suppressing the OFDM interference

IEEE 802.11 wireless local area networks that span large buildings or campuses must comprise multiple cells, several of which must necessarily be cochannel cells. It can be shown that in such multicell networks, typically, the radio ranges of cochannel cells overlap. The paper studies the saturation throughput performance of two cochannel cells with critical overlap, i.e., their interference ranges completely overlap but no node in either cell can decode any transmissions from the other cell. We identify that the difference between the two MAC parameters, EIFS (extended interframe space) and DIFS (DCF interframe space) is a key issue and incorporate this difference as a parameter into an analytical model. The model yields a fixed point equation that yields an approximation to the saturation throughputs of the two cells. The results from the analysis are validated against ns-2 simulations. We find that with EIFS>DIFS there is substantial temporal unfairness in the channel access between the two cells, but, because of fewer collisions, the critical overlap configuration has higher per cell throughput than if the cells' decoding ranges overlapped.

Auditory scene analysis (ASA) tries to segment an auditory signal (scene) into objects. Most of the intermediate representations currently proposed based on ASA are difficult to compute. We propose auditory strands and blobs as intermediate representations. Auditory blobs are parts of an audio signal which have the same onset. By the principles of computational auditory scene analysis, they belong to the same object. We show how auditory blobs can be extracted and define harmonicity, dynamics, and onset features for auditory blobs. We also demonstrate their application to audio separation.

Features are very important for audio processing. Tasks like speech recognition and instrument identification are based on features. Most low-level features currently used are based on LPC and cepstral analysis. We propose a class of features based on dynamics and harmonicity. In particular, we define the notion of harmonic derivative. The efficacy of the features is demonstrated for music genre classification and instrument family classification. In particular, the features are shown to be cepstrum-equivalent.

The boom in new streaming media applications in the recent past calls for innovative strategies that can support both data and media over the same global IP based Internet in a fair and efficient manner. Several solutions have already been proposed to control the rate of media traffic and to maintain fairness among the heterogeneous traffic flows. We propose a mechanism that solves many of the problems associated with the earlier schemes. A new concept, called "loss sharing", is introduced for 'media rate control' and 'maintenance of fairness'. We show that these two objectives are automatically achieved as desirable side-effects of the recently proposed "TTL routing algorithm". This algorithm adopts a new interpretation of the TTL (time-to-live) field of the IP header. It is shown that rate control by "TTL routing" has certain other desirable features.

Conventional mosaicing techniques convert a video from frame-based representation to scene-based representation, but they usually lack dynamic information so that their mosaic is not complete. In this paper, we present a novel method to detect moving objects in the video sequences, then add them into the static background mosaic to represent the scene completely. This novel algorithm separates static and dynamic information in a video sequence, builds the background mosaic from static part and reconstructs moving objects on the static mosaic. We have implemented our techniques and the experimental results demonstrate the effectiveness of our approach

Earth mover's distance is one of the recently proposed distance measures in multimedia. Though attractive, the earth mover's distance is computationally complex. In this paper, we propose a local earth mover's distance which reduces the size of the linear program used by exploiting locality constraints. The speedup and approximation capabilities of the local earth mover's distance are demonstrated. The local earth mover's distance is then applied to warping face images

Characterizing music dynamics is important for music generation, music analogies, music retrieval, music improvisation etc. Though there has been some works on modeling music, most of them do not consider the temporal or dynamical aspects of music. In this paper we show that it is possible to characterize music dynamics using linear prediction. We provide a technique for music improvisation using this model. The improvisation scheme is then applied to the generation of background music for videos

Wavelength converters in simple wavelength division multiplexed (WDM) networks helps in improving the network blocking performance by reducing the number of blocked demands arising out of the wavelength continuity constraint. However, cost and technological limitations make sparse wavelength conversion networks preferable. In this work, we propose a technique for wavelength converter placement under non-uniform traffic in the network. The algorithm that uses normalized cut based graph partitioning technique is extended for non-uniform traffic. A comparison of the performance under uniform and non-uniform traffic is given to highlight the effect of traffic on optimal placement. The algorithm is attractive due to lower complexity and efficiency.

Objective speech quality measures are very important for evaluating speech codecs and VoIP systems. Recent speech quality measures are based on psychoacoustic facts. In this paper we incorporate ideas from auditory scene analysis in the computation of speech quality measure. Specifically, use notions of harmonicity, dynamics, and onset times. Both relative and absolute metrics are proposed.

This paper addresses the problem of extracting video objects from MPEG compressed video. The only cues used for object segmentation are the motion vectors which are sparse in MPEG. A method for automatically estimating the number of objects and extracting independently moving video objects using motion vectors is presented here. First, the motion vectors are accumulated over a few frames to enhance the motion information, which are further spatially interpolated to get dense motion vectors. The final segmentation, using the dense motion vectors, is obtained by applying the expectation maximization (EM) algorithm. A block-based affine clustering method is proposed for determining the number of appropriate motion models to be used for the EM step and the segmented objects are temporally tracked to obtain the video objects. Finally, a strategy for edge refinement is proposed to extract the precise object boundaries. Illustrative examples are provided to demonstrate the efficacy of the approach. A prominent application of the proposed method is that of object-based coding, which is part of the MPEG-4 standard.

Audio signal source separation is an interesting task performed by humans. In this paper, we present a frequency grouping algorithm based on principles of harmonicity and dynamics: frequency components with a harmonic relation and similar dynamics belong to the same source. The grouping is demonstrated for a variety of sound mixtures.

Wide baseline matching of images is an important problem in multimedia and computer vision. Recently a class of spectral methods have been proposed for tasks like segmentation. In this paper we propose a spectral algorithm for wide baseline matching.

We propose a new multiple description method for wavelet transform coded images. The partitioning of the coefficients at the encoder is combined with concealment at the decoder to produce descriptions. An optimal concealment method for wavelet coefficients is derived based on statistical characteristics. Concealment in the LL subband is treated differently and is based on its smoothness property. We have designed a coder for wireless image transmission which emphasizes the need for source and channel coding interaction based on residual redundancy. Optimal channel rate allocation is derived depending upon source significance and channel state information. Simulations for wireless transmission have shown promising results.

A technique to extract pitch information directly from audio files encoded using MPEG/Audio coding standard is described. The technique works in the compressed domain and requires the MPEG/Audio file to be decoded only partially for extracting the encoded subband samples. The paper first proposes a method for extracting wavelet coefficients from the subband samples. The pitch interval is then estimated from the time interval between two successive maxima of the wavelet coefficients. It is shown that the computational complexity for compressed domain pitch extraction is less than 7% of the computations required for decoding the subband samples and finding the pitch.

In this paper, we explore the convergence of the caching and streaming technologies for Internet multimedia. The paper describes a design for a streaming and caching architecture to be deployed on broadband networks. The basis of the work is the proposed Internet standard, Real Time Streaming Protocol (RTSP), likely to be the de-facto standard for web-based A/V caching and streaming, in the near future. The proxies are all managed by an `Intelligent Agent' or `Broker' - this has been designed as an enhanced RTSP proxy server that maintains the state information that is so essential in streaming of media data. In addition, all the caching algorithms run on the broker. Having an intelligent agent or broker ensures that the `simple' caching servers can be easily embedded into the network. However, RTSP does nor have the right model for doing broker based streaming/caching architecture. The work reported here is an attempt to contribute towards that end.

What are the natural features of handwritten characters and how to arrive at them automatically? We apply independent components analysis on handwritten characters. Independent components analysis extracts the underlying statistically independent signals from a mixture of them. We expect strokes to be the independent components of handwritten characters. Our findings show that stroke-like features emerge as a result of the analysis confirming the above intuition. This finding is significant since it gives automatic procedures for extracting stroke-like features from multilingual character data sets. We use these features for handwritten digit recognition using a very simple classifier. The classifier is chosen to be simple so that the quality of the input feature set can be evaluated. The recognition results indicate that the features arrived at by independent component analysis are useful.

A new technique for scene change detection using neural networks is proposed to operate on compressed video data obtained with only minimal decoding of video. In this paper various types of neural network structures have been trained and tested on different varieties of video clips. The neural networks are shown to be robust for camera activities and object movements. They gave comparatively good results over the conventional techniques. All the algorithms are tested on MPEG-1 video streams.