Abstract: Prototype based classification offers intuitive and sparse models with excellent generalization ability. However, these models usually crucially depend on the underlying Euclidian metric; moreover, online variants likely suffer from the problem of local optima. We here propose a generalization of learning vector quantization with three additional features: (I) it directly integrates neighborhood cooperation, hence is less affected by local optima; (II) the method can be combined with any differentiable similarity measure whereby metric parameters such as relevance factors of the input dimensions can automatically be adapted according to the given data; (III) it obeys a gradient dynamics hence shows very robust behavior, and the chosen objective is related to margin optimization.

Abstract: Natural head motion is important to realistic facial animation and engaging human–computer interactions. In this paper, we present a novel data-driven approach to synthesize appropriate head motion by sampling from trained hidden markov models (HMMs). First, while an actress recited a corpus specifically designed to elicit various emotions, her 3D head motion was captured and further processed to construct a head motion database that included synchronized speech information. Then, an HMM for each discrete head motion representation (derived directly from data using vector quantization) was created by using acoustic prosodic features derived from speech. Finally, first-order Markov models and interpolation techniques were used to smooth the synthesized sequence. Our comparison experiments and novel synthesis results show that synthesized head motions follow the temporal dynamic behavior of real human subjects. Copyright © 2005 John Wiley & Sons, Ltd.

Abstract: System and method for determining a classifier to discriminate between two classes—object or non-object. The classifier may be used by an object detection program to detect presence of a 3D object in a 2D image. The overall classifier is constructed of a sequence of classifiers, where each such classifier is based on a ratio of two graphical probability models. A discreet-valued variable representation at each node in a Bayesian network by a two-stage process of tree-structured vector quantization is discussed. The overall classifier may be part of an object detector program that is trained to automatically detect different types of 3D objects. Computationally efficient statistical methods to evaluate overall classifiers are disclosed. The Bayesian network-based classifier may also be used to determine if two observations belong to the same category.

Abstract: In this paper, we use the recently proposed wet paper codes and introduce a new approach to passive-warden steganography called perturbed quantization. In perturbed quantization, the sender hides data while processing the cover object with an information-reducing operation that involves quantization, such as lossy compression, downsampling, or A/D conversion. The unquantized values of the processed cover object are considered as side information to confine the embedding changes to those unquantized elements whose values are close to the middle of quantization intervals. This choice of the selection channel calls for wet paper codes as they enable communication with non-shared selection channel. Heuristic is presented that indicates that the proposed method provides better steganographic security than current JPEG steganographic methods. This claim is further supported by blind steganalysis of a specific case of perturbed quantization for recompressed JPEG images.

Abstract: We derive a generalization bound for prototype-based classifiers with adaptive metric. The bound depends on the margin of the classifier and is independent of the dimensionality of the data. It holds for classifiers based on the Euclidean metric extended by adaptive relevance terms. In particular, the result holds for relevance learning vector quantization (RLVQ) [4] and generalized relevance learning vector quantization (GRLVQ) [19].

Abstract: A new kind of feature for colour image retrieval based on DCT-domain vector quantisation (VQ) index histograms (DCTVQIH) is proposed. For each colour image in the database, 12 histograms (four for each colour component) are calculated from 12 DCT-VQ index sequences, respectively. The retrieval simulation results show that, compared with the traditional spatial-domain colour-histogram-based features, the proposed features can largely improve the recall and precision performance.

Abstract: A shot-based video content analysis method and system is described for providing automatic recognition of logical story units (LSUs). The method employs vector quantization (VQ) to represent the visual content of a shot, following which a shot clustering algorithm is employed together with automatic determination of merging and splitting events. The method provides an automated way of performing the time-consuming and laborious process of organising and indexing increasingly large video databases such that they can be easily browsed and searched using natural query structures.

Abstract: An efficient audio coding device that quantizes and encodes digital audio signals with a reduced amount of computation. A spatial transform unit subjects samples of a given audio signal to a spatial transform, thus obtaining transform coefficients of the signal. With a representative value selected out of the transform coefficients of each subband, a quantization step size calculator estimates quantization noise and calculates, in an approximative way, a quantization step size of each subband from the estimated quantization noise, as well as from a masking power threshold determined from a psycho-acoustic model of the human auditory system. A quantizer then quantizes the transform coefficients, based on the calculated quantization step sizes, thereby producing quantized values of those coefficients. The quantization step sizes are also used by a scalefactor calculator to calculate common and individual scalefactors. A coder encodes at least one of the quantized values, common scalefactor, and individual scalefactors.

Abstract: This paper presents a novel approach to the classification of power quality disturbances by the employment of Hidden Markov Models. In these models, power quality disturbances are represented by a sequence of consecutive frames. Both the Fourier and Wavelet Transforms are utilized to produce sequence of spectral vectors that can accurately capture the salient characteristics of each disturbance. Vector Quantization is used to assign chain of labels for power quality disturbances utilizing their spectral vectors. From these labels, a separate Hidden Markov Model is developed for each class of the power quality disturbances in the training phase. During the testing stage, the unrecognized disturbance sequence is matched against all the developed Hidden Markov Models. The best-matched model pinpoints the class of the unknown disturbance. Simulation results prove the competence of the proposed algorithm.

Abstract: An important aspect of analog-to-digital conversion is the impact of quantization errors. This paper outlines how finite horizon constrained optimization methods can be utilized to design converters which minimize a weighted measure of the quantization distortion. We propose a novel converter, which can be implemented as a feedback loop. It embeds /spl Sigma//spl Delta/ conversion in a more general setting and typically provides better performance. We also examine the role played by the associated design parameters in ensuring error convergence.

Abstract: In data assimilation for numerical weather prediction, measurements of various observation systems are combined with background data to define initial states for the forecasts. Current and future observation systems, in particular satellite instruments, produce large numbers of measurements with high spatial and temporal density. Such datasets significantly increase the computational costs of the assimilation and, moreover, can violate the assumption of spatially independent observation errors. To ameliorate these problems, we propose two greedy thinning algorithms, which reduce the number of assimilated observations while retaining the essential information content of the data. In the first method, the number of points in the output set is increased iteratively. We use a clustering method with a distance metric that combines spatial distance with difference in observation values. In a second scheme, we iteratively estimate the redundancy of the current observation set and remove the most redundant data points. We evaluate the proposed methods with respect to a geometric error measure and compare them with a uniform sampling scheme. We obtain good representations of the original data with thinnings retaining only a small portion of observations. We also evaluate our thinnings of ATOVS satellite data using the assimilation system of the Deutscher Wetterdienst. Impact of the thinning on the analysed fields and on the subsequent forecasts is discussed. Copyright © 2005 Royal Meteorological Society

Abstract: An electrocardiogram (ECG) data compression scheme is presented using the gain-shape vector quantization. The proposed approach utilizes the fact that ECG signals generally show redundancy among adjacent heartbeats and adjacent samples. An ECG signal is QRS detected and segmented according to the detected fiducial points. The segmented heartbeats are vector quantized, and the residual signals are calculated and encoded using the AREA algorithm. The experimental results show that with the proposed method both visual quality and the objective quality are excellent even in low bit rates. An average PRD of 5.97% at 127 b/s is obtained for the entire 48 records in the MIT-BIH database. The proposed method also outperforms others for the same test dataset.

Abstract: Indices are modified so that the secret data can be hidden into the index-based cover image, and thereby the problem of the stego-image quality degradation occurs. If the stego-image quality degradation problem can be solved enabling the receiver to reconstruct the original indices after extracting the hidden secret data from the index-based stego-image, then the compressed cover image can be used repeatedly by different users. To achieve our goal, in this paper, a reversible data-hiding scheme based on a modified side match vector quantization (SMVQ) technique is proposed. Our experimental results confirm the effectiveness and the reversibility of the proposed scheme.

Abstract: We introduce a technique for dimensionality estimation based on the notion of quantization dimension, which connects the asymptotic optimal quantization error for a probability distribution on a manifold to its intrinsic dimension. The definition of quantization dimension yields a family of estimation algorithms, whose limiting case is equivalent to a recent method based on packing numbers. Using the formalism of high-rate vector quantization, we address issues of statistical consistency and analyze the behavior of our scheme in the presence of noise.

Abstract: Vector quantization (VQ) is an effective and widely known method for low-bit-rate communication of speech and image signals. A common assumption in the design of VQ-based communication systems is that the compressed digital information is transmitted through a perfect channel. Under this assumption, quantization distortion is the only factor in output signal fidelity. Moreover, the assignment of channel symbols to the VQ reconstruction vectors is of no importance. However, under physical channels, errors may be present, causing degradation in overall system performance. In such a case, the effect of channel errors on the coding system performance depends on the index assignment of the reconstruction vectors. The index assignment problem is a special case of the Quadratic Assignment Problem (QAP) and is known to be NP-complete. For a VQ with N reconstruction vectors there are N! possible assignments, meaning that an exhaustive search over all possible assignments is practically impossible. To help the VQ designer, we present in this correspondence lower and upper bounds on the performance of VQ systems under channel errors, over all possible assignments. The bounds coincide with a general bound for the QAP. Nevertheless, the proposed derivation allows us to compare the bounds with published results on VQ index assignment. A related expression for the average performance is also given and discussed. Special cases and numerical examples are given in which the bounds and average performance are compared with index assignments obtained by known algorithms.

Abstract: The utility of the supervised Kohonen self-organizing map was assessed and compared to several statistical methods used in QSAR analysis. The self-organizing map (SOM) describes a family of nonlinear, topology preserving mapping methods with attributes of both vector quantization and clustering that provides visualization options unavailable with other nonlinear methods. In contrast to most chemometric methods, the supervised SOM (sSOM) is shown to be relatively insensitive to noise and feature redundancy. Additionally, sSOMs can make use of descriptors having only nominal linear correlation with the target property. Results herein are contrasted to partial least squares, stepwise multiple linear regression, the genetic functional algorithm, and genetic partial least squares, collectively referred to throughout as the "standard methods". The k-nearest neighbor (kNN) classification method was also performed to provide a direct comparison with a different classification method. The widely studied dihydrofolate reductase (DHFR) inhibition data set of Hansch and Silipo is used to evaluate the ability of sSOMs to classify unknowns as a function of increasing class resolution. The contribution of the sSOM neighborhood kernel to its predictive ability is assessed in two experiments: (1) training with the k-means clustering limit, where the neighborhood radius is zero throughout the training regimen, and (2) training the sSOM until the neighborhood radius is reduced to zero. Results demonstrate that sSOMs provide more accurate predictions than standard linear QSAR methods.

Abstract: With the proliferation of digital multimedia content, such as image, audio, video, and animation, information hiding techniques have attracted more and more research interests. High bitrate information hiding is different from digital watermarking in that it tries to hide relatively large amount of auxiliary information instead of just one or a few verification bits. This research is mainly focused on high bitrate information hiding within digital video content. Channel capacity and the robustness of hidden information against lossy video codec are the two main concerns. In the proposed algorithm, 1 bit is hidden within each 4*4 DCT coefficient block by means of vector quantization. Low-frequency coefficients are chosen for information hiding due to their relatively large amplitudes and the corresponding small step sizes in the quantization matrix. The proposed algorithm is tested under H.264/AVC (advanced video codec), which is the state-of-the-art video coding standard. Experimental results show that the proposed algorithm is very robust and a very high percentage of the hidden information survives H.264/AVC compression. It is also observed that hidden information within different types of frame/slice have different levels of robustness against H.264/AVC compression

Abstract: VQ coding is a powerful technique in digital image compression. Conversional methods such as classic LBG algorithm always generate local optimal codebook. In this paper, we introduce Particle Swarm Optimization (PSO) cluster method to build high quality codebook for image compression. We also set the result of LBG algorithm to initialize global best particle by which it can speed the convergence of PSO. Both image encoding and decoding process are simulated in our experiments. Results show that the algorithm is reliable and the reconstructed images get higher quality to images reconstructed by other methods.

Abstract: PROBLEM TO BE SOLVED: To improve coding efficiency by suitably setting, with respect to input information on moving image coding, a parameter for quantization calculation at the time of quantization. SOLUTION: Various parameters for quantization calculation are set by a parameter setter 103, quantization is performed with the set parameters for quantization calculation by a quantizer 102, a coding cost J is calculated from a generated code amount R and a quantization distortion amount D obtained as a result thereof, and a parameter judgement unit 107 for quantization calculation selects, as a parameter for quantization calculation having highest coding efficiency, a parameter by which the cost J becomes minimum. COPYRIGHT: (C)2006,JPO&NCIPI

Abstract: We present an approach for the supervised online learning of object representations based on a biologically motivated architecture of visual processing We use the output of a recently developed topographical feature hierarchy to provide a view-based representation of three-dimensional objects using a dynamical vector quantization approach For a simple short-term object memory model we demonstrate real-time online learning of 50 complex-shaped objects within three hours Additionally we propose some modifications of learning vector quantization algorithms that are especially adapted to the task of online learning and capable of effectively reducing the representational effort in a transfer from short-term to long-term memory

Abstract: An image data compression device includes a quantization part quantizing image data with a quantization step that varies based on a quantization parameter, a FIFO buffer part buffering quantized data of a plurality of frames, a coded data formation part reading out the quantized data from the FIFO buffer part asynchronously with a writing to the FIFO buffer part and generating coded data by encoding the quantized data, a rate control part controlling a data size of the coded data by changing the quantization step and a frame skip part skipping a generation process of the image data. The rate control part calculates the quantization parameter by using a predicted data size of the coded data of a previous frame which is calculated from a data size of the quantized data of the previous frame. The frame skip part performs the skip process if a frame, in which the quantization parameter becomes larger than a skip threshold, appears consecutively in a number of times which is equal or more than the number of a skip succession threshold.

Abstract: To increase the number of the embedded secrets and to improve the quality of the stego-image in the vector quantization (VQ)-based information hiding scheme, in this paper, we present a novel information-hiding scheme to embed secrets into the side match vector quantization (SMVQ) compressed code First, a host image is partitioned into non-overlapping blocks For these seed blocks of the image, VQ is adopted without hiding secrets Then, for each of the residual blocks, SMVQ or VQ is employed according to the smoothness of the block such that the proper codeword is chosen from the state codebook or the original codebook to compress it Finally, these compressed codes represent not only the host image but also the secret data Experimental results show that the performance of the proposed scheme is better than other VQ-based information hiding scheme in terms of the embedding capacity and the image quality Moreover, in the proposed scheme, the compression rate is better than the compared scheme

Abstract: This paper proposes an encoding method for high-quality, low-delay audio communication that is robust to losses in packetized transmission. Robustness is provided by a multiple description vector quantization (MDVQ) technique that is designed to minimize the mean-squared error (MSE). The key to applying this technique effectively is the use of psycho-acoustically controlled preand post-filters that make the mean-squared quantization error perceptually relevant. Experiments show that the MDVQ-based encoder yields better results-in both MSE and subjective audio quality-than simple alternative coders with the same low delay.

Abstract: This paper studies the problem of quantization of a source that lives on the complex Grassmann manifold. The special structure of the Grassmann manifold and the distortion measures that are defined on it differentiates this problem from the traditional problem of vector quantization in Euclidean spaces. Assuming a uniform source distribution along with a distortion based on chordal distance, codebook design algorithms are mentioned and rate distortion tradeoffs are studied. The expected distortion for such a quantizer is approximately characterized. These results are then applied to the performance analysis of a multiple antenna wireless communication system.

Abstract: This paper describes an automatic real-time video surveillance system, capable of autonomously learning and signaling anomalous activities of moving objects To obtain these capabilities, an improved version of the altruistic vector quantization algorithm (AVQ) is proposed The modified AVQ automatically evaluates the number of trajectory prototypes, and improves the representativeness of the prototypes themselves, so the visual events can be easily and accurately classified Anomalous behaviors are detected if visual trajectories deviate from the self-learned representations of "typical" behaviors The system has been implemented by means of standard PCs and TV cameras, and has been tested in many real outdoor contexts in different conditions (night and day) Currently it is used to monitor the storage areas of British Airways at the airport of Peretola (Florence, Italy), and some access gates of Autostrade per FItalia SpA (the main Italian highways company) If the camera field-of-view is changed, the system automatically re-learns new "typical" behaviors and accurately detects anomalous events