Showing papers on "Codebook published in 1996"

Apparatus and method for analyzing and correlating events in a system using a causality matrix

Abstract: An apparatus and method is provided for efficiently determining the source of problems in a complex system based on observable events. The problem identification process is split into two separate activities of (1) generating efficient codes for problem identification and (2) decoding the problems at runtime. Various embodiments of the invention contemplate creating a causality matrix which relates observable symptoms to likely problems in the system, reducing the causality matrix into a minimal codebook by eliminating redundant or unnecessary information, monitoring the observable symptoms, and decoding problems by comparing the observable symptoms against the minimal codebook using various best-fit approaches. The minimal codebook also identifies those observable symptoms for which the greatest benefit will be gained if they were monitored as compared to others. By defining a distance measure between symptoms and codes in the codebook, the invention can tolerate a loss of symptoms or spurious symptoms without failure. Changing the radius of the codebook allows the ambiguity of problem identification to be adjusted easily. The invention also allows probabilistic and temporal correlations to be monitored.

Codebook constrained Wiener filtering for speech enhancement

Abstract: Speech enhancement using iterative Wiener filtering has been shown to require interframe and intraframe constraints in all-pole parameter estimation We show that a clean speech VQ codebook is more effective in providing intraframe constraints and, hence, better convergence of the iterative filtering scheme Satisfactory speech enhancement results are obtained with a small codebook of 128, and the algorithm is effective for both white noise and pink noise up to 0 dB SNR

Simple universal lossy data compression schemes derived from the Lempel-Ziv algorithm

Abstract: Two universal lossy data compression schemes, one with fixed rate and the other with fixed distortion, are presented, based on the well-known Lempel-Ziv algorithm. In the case of fixed rate R, the universal lossy data compression scheme works as follows: first pick a codebook B/sub n/ consisting of all reproduction sequences of length n whose Lempel-Ziv codeword length is /spl les/nR, and then use B/sub n/ to encode the entire source sequence n-block by n-block. This fixed-rate data compression scheme is universal in the sense that for any stationary, ergodic source or for any individual sequence, the sample distortion performance as n/spl rarr//spl infin/ is given almost surely by the distortion rate function. A similar result is shown in the context of fixed distortion lossy source coding.

Two fast nearest neighbor searching algorithms for image vector quantization

Abstract: In this paper, two efficient codebook searching algorithms for vector quantization (VQ) are presented. The first fast search algorithm utilizes the compactness property of signal energy on transform domain and the geometrical relations between the input vector and every codevector to eliminate those codevectors that have no chance to be the closest codeword of the input vector. It achieves a full search equivalent performance. As compared with other fast methods of the same kind, this algorithm requires the fewest multiplications and the least total times of distortion measurements. Then, a suboptimal searching method, which sacrifices the reconstructed signal quality to speed up the search of nearest neighbor, is presented. This algorithm performs the search process on predefined small subcodebooks instead of the whole codebook for the closest codevector. Experimental results show that this method not only needs less CPU time to encode an image but also encounters less loss of reconstructed signal quality than tree-structured VQ does.

Depth-first algebraic-codebook search for fast coding of speech

Abstract: A codebook is searched in view of encoding a sound signal. This codebook consists of a set of codevectors each of 40 positions and comprising N non-zero-amplitude pulses assignable to predetermined valid positions. To reduce the search complexity, a depth-first search is used which involves a tree structure with levels ordered from 1 through M. A path-building operation takes place at each level whereby a candidate path from the previous level is extended by choosing a predetermined number of new pulses and selecting valid positions for said new pulses in accordance with a given pulse-order rule and a given selection criterion. A path originated at the first level and extended by the path-building operations of subsequent levels determines the respective positions of the N non-zero-amplitude pulse of a candidate codevector. Use of a signal-based pulse-position likelihood estimate during the first few levels enable initial pulse-screening to start the search on favorable conditions. A selection criterion based on maximizing a ratio is used to assess the progress and to choose the best one among competing candidate codevectors.

Fast three‐dimensional data compression of hyperspectral imagery using vector quantization with spectral‐feature‐based binary coding

Abstract: A fast lossy 3-D data compression scheme using vector quantization (VQ) is presented that exploits the spatial and the spectral redundancy in hyperspectral imagery. Hyperspectral imagery may be viewed as a 3-D array of samples in which two dimensions correspond to spatial position and the third to wavelength. Unlike traditional 2-D VQ, where spatial blocks of n3m pixels are taken as vectors, we define one spectrum, corresponding to a profile taken along the wavelength axis, as a vector. This constitution of vectors makes good use of the high corre- lation in the spectral domain and achieves a high compression ratio. It also leads to fast codebook generation and fast codevector matching. A coding scheme for fast vector matching called spectral-feature-based binary coding (SFBBC) is used to encode each spectral vector into a simple and efficient set of binary codes. The generation of the codebook and the matching of codevectors are performed by matching the binary codes produced by the SFBBC. The experiments were carried out using a test hyperspectral data cube from the Compact Airborne Spectro- graphic Imager. Generating a codebook is 39 times faster with the SF- BBC than with conventional VQ, and the data compression is 30 to 40 times faster. Compression ratios greater than 192 : 1 have been achieved with peak signal-to-noise ratios of the reconstructed hyper- spectral sequences exceeding 45.2 dB. © 1996 Society of Photo-Optical Instru- mentation Engineers.

An on-line universal lossy data compression algorithm via continuous codebook refinement. I. Basic results

Abstract: A new on-line universal lossy data compression algorithm is presented. For finite memoryless sources with unknown statistics, its performance asymptotically approaches the fundamental rate distortion limit. The codebook is generated on the fly, and continuously adapted by simple rules. There is no separate codebook training or codebook transmission. Candidate codewords are randomly generated according to an arbitrary and possibly suboptimal distribution. Through a carefully designed "gold washing" or "information-theoretic sieve" mechanism, good codewords and only good codewords are promoted to permanent status with high probability. We also determine the rate at which our algorithm approaches the fundamental limit.

Embedded algebraic vector quantizers (EAVQ) with application to wideband speech coding

Abstract: A new lattice vector quantization scheme, namely embedded algebraic vector quantizers (EAVQ), is proposed. This scheme makes use of spherical subsets of the rotated Gosset lattice RE/sub 8/ to constitute a vector-quantizer codebook. The codebook consists of several sub-codebooks and has an embedded structure. The codewords can be generated using an algebraic method and do not have to be stored. In combination with transform coded excitation (TCX) coding, this quantization technique is applied to 16 kbps wideband speech coding in order to quantize the so-called target signal in the frequency domain. Compared to a stochastic complex vector quantization scheme, EAVQ can achieve better performance and lead to significant savings of memory requirements.

Variable-dimension vector quantization

Abstract: In many signal compression applications, the evolution of the signal over time can be represented by a sequence of random vectors with varying dimensionality. Frequently, the generation of such variable-dimension vectors can be modeled as a random sampling of another signal vector with a large but fixed dimension. Efficient quantization of these variable-dimension vectors is a challenging task and a critical issue in speech coding algorithms based on harmonic spectral modeling. We introduce a simple and effective formulation of the problem and present a novel technique, called variable-dimension vector quantization (VDVQ), where the input variable-dimension vector is directly quantized with a single universal codebook. The application of VDVQ to low bit-rate speech coding demonstrates significant gain in subjective quality as well as in rate-distortion performance over prior indirect methods.

Apparatus and method for encoding/decoding a speech signal using adaptively changing codebook vectors

Abstract: An encoding apparatus in which an input speech signal is divided into blocks and encoded in units of blocks. The encoding apparatus includes an encoding unit for performing CELP encoding having a noise codebook memory containing having codebook vectors generated by clipping Gaussian noise and codebook vectors obtained by learning using the code vectors generated by clipping the Gaussian noise as initial values. The encoding apparatus enables optimum encoding for a variety of speech configurations.

Table-based low-level image classification and compression system

Abstract: A system for classifying image elements comprising means for converting an image into a series of vectors and a hierarchical lookup table that classifies the vectors. The lookup table implements a pre-computed discrete cosine transform (DCT) to enhance classification accuracy. The hierarchical lookup table includes four stages: three of which constitute a preliminary section; the fourth stage constitutes the final section. Each stage has a respective stage table. The method for designing each stage table comprises a codebook design procedure and a table fill-in procedure. Codebook design for the preliminary stages strives to minimize a classification-sensitive proximity measure; codebook design for the final stage attempts to minimize Bayes risk of misclassification. Table fill-in for the first stage involves generating all possible input combinations, concatenating each possible input combination to define a concatenated vector, applying a DCT to convert the address vector to the spatial frequency domain, finding the closest first-stage codebook vector, and assigning to the address the index associated that codebook vector. Table fill-in for subsequent stages involves decoding each possible input combination to obtain spatial frequency domain vectors, applying an inverse DOC to convert the inputs to pixel domain vectors, concatenating the pixel domain vectors to obtain a higher dimension pixel domain vector, applying a DCT to obtain a spatial frequency domain vector, finding the closest same-stage codebook vector, and assigning the codebook vector index to the input combination.

A generalized VQ method for combined compression and estimation

Abstract: In vector quantization, one approximates an input random vector, Y, by choosing from a finite set of values known as the codebook. We consider a more general problem where one may not have direct access to Y but only to some statistically related random vector X. We observe X and would like to generate an approximation to Y from a codebook of candidate vectors. This operation, called generalized vector quantization (GVQ), is essentially that of quantized estimation. An important special case of GVQ is the problem of noisy source coding wherein a quantized approximation of a vector, Y, is obtained from observation of its noise-corrupted version, X. The optimal GVQ encoder has high complexity. We overcome the complexity barrier by optimizing a structurally-constrained encoder. This challenging optimization task is solved via a probabilistic approach, based on deterministic annealing, which overcomes problems of shallow local minima that trap simpler descent methods. We demonstrate the successful application of our method to the coding of noisy sources.

Signal conditioned minimum error rate training for continuous speech recognition

Abstract: Hierarchical signal bias removal (HSBR) signal conditioning (26) uses a codebook (28) constructed from the set of recognition models (34). HSBR signal conditioning and recognition model training may be based on the same set of recognition model parameters and provides a reduction in recognition errors.

Perceptual speech coding using prediction residuals, having harmonic magnitude codebook for voiced and waveform codebook for unvoiced frames

Abstract: A speech encoding method and apparatus for encoding an input speech signal on a block-by-block or frame-by-frame basis wherein short-term prediction residuals are found and then sinusoidal analytic encoding parameters are produced based on those short-term prediction residuals. Perceptually weighted vector quantization is performed for voiced blocks or frames by encoding their sinusoidal frequency or analytic harmonic magnitudes and, in the case of unvoiced blocks or frames, the time waveforms of the unvoiced blocks are encoded.

Adaptive edge-based side-match finite-state classified vector quantization with quadtree map

Abstract: Vector quantization (VQ) is an effective image coding technique at low bit rate. The side-match finite-state vector quantizer (SMVQ) exploits the correlations between neighboring blocks (vectors) to avoid large gray level transition across block boundaries. A new adaptive edge-based side-match finite-state classified vector quantizer (classified FSVQ) with a quadtree map has been proposed. In classified FSVQ, blocks are arranged into two main classes, edge blocks and nonedge blocks, to avoid selecting a wrong state codebook for an input block. In order to improve the image quality, edge vectors are reclassified into 16 classes. Each class uses a master codebook that is different from the codebooks of other classes. In our experiments, results are given and comparisons are made between the new scheme and ordinary SMVQ and VQ coding techniques. As is shown, the improvement over ordinary SMVQ is up to 1.16 dB at nearly the same bit rate, moreover, the improvement over ordinary VQ can be up to 2.08 dB at the same bit rate for the image, Lena. Further, block boundaries and edge degradation are less visible because of the edge-vector classification. Hence, the perceptual image quality of classified FSVQ is better than that of ordinary SMVQ.

Multiresolution joint source-channel coding using embedded constellations for power-constrained time-varying channels

Abstract: We explore joint source-channel coding (JSCC) for time-varying (slow-fading Rayleigh) channels, using a multiresolution (MR) framework for both source coding and transmission (via a novel MR modulation constellation). We tackle the important case of the informed receiver but uninformed transmitter, i.e. where the receiver has access to the channel state information (CSI), but the transmitter does not. We describe an algorithm which jointly optimizes the design of the MR source codebook, the MR constellation, and the decoding strategy of optimally matching the source and signal constellation resolution trees according to the time-varying channel, and show how this leads to improved performance over separately designed source and channel coders.

The futility of square isometries in fractal image compression

Abstract: In fractal image compression an image is partitioned into a set of image blocks, called ranges. The ranges are matched with blocks taken from a codebook of filtered and subsampled domain image blocks up to an affine transformation of intensity values. It is common practise in fractal image compression to include all 8 isometric versions of a codebook block in the codebook. It is reasoned that such enlarged domain pools yield better rate-distortion curves. However, this is not a valid argument supporting the use of isometries. A fair test must compare the performance of the method using a codebook including isometries with that obtained when using a plain codebook of the same size. We have performed such analysis and our results show that codebooks with isometries offer no advantages in terms of fidelity-in contrast to the prevalent belief. A similar study is carried out for the effects of including respectively excluding negative scaling factors in the fractal code.

An 8-kb/s conjugate structure CELP (CS-CELP) speech coder

Abstract: This paper describes a high-quality 8-kb/s speech coder called conjugate structure code-excited linear prediction (CS-CELP) with a 10-ms frame length. To provide a short delay and high quality under both error-free and channel error conditions, it uses three new schemes: line spectrum pair (LSP) quantization using interframe prediction, preselection in the codebook search, and gain vector quantization (VQ) with backward prediction. The LSP parameters are quantized by using multistage VQ with moving-average (MA) prediction. This scheme can operate efficiently with various frequency responses of speech. The preselection of the codebook reduces the computational complexity and improves the robustness to channel errors. The gain VQ with backward prediction can provide a high quality and robustness without transmission of input speech power information. A conjugate structure for both random codebook and gain codebook is introduced to improve the ability to handle random bit errors and to reduce codebook storage memory requirements. Subjective testing indicates that the quality of this coder is equivalent to that of 32-kb/s adaptive differential pulse code modulation (ADPCM) under error-free conditions. Testing has further demonstrated that the coder is robust against random bit errors.

An on-line universal lossy data compression algorithm via continuous codebook refinement. II. Optimality for phi-mixing source models

Abstract: For pt.I see ibid., vol.42, no.3, p.803-21 (1996). Two versions of the gold-washing data compression algorithm, one with codebook innovation interval and the other with finitely many codebook innovations, are considered. The version of the gold-washing algorithm with codebook innovation interval k is a variant of the gold-washing algorithm such that the codebook is innovated once every k+1 source words during the process of encoding the entire source. It is demonstrated that when this version of the gold-washing algorithm is applied to encode a stationary, /spl phi/-mixing source, the expected distortion performance converges to the distortion rate function of the source as the codebook length goes to infinity. Furthermore, if the source to be encoded is a Markov source or a finite-state source, then the corresponding sample distortion performance converges almost surely to the distortion rate function. The version of the gold-washing algorithm with finitely many codebook innovations is a variant of the gold-washing algorithm in which after finitely many codebook innovations, the codebook is held fixed and reused to encode the forthcoming source sequence block by block. Similar results are shown for this version of the gold-washing algorithm. In addition, the convergence speed of the algorithm is discussed.

Progressive vector quantization on a massively parallel SIMD machine with application to multispectral image data

Abstract: This correspondence discusses a progressive vector quantization (VQ) compression approach, which decomposes image data into a number of levels using full-search VQ. The final level is losslessly compressed, enabling lossless reconstruction. The computational difficulties are addressed by implementation on a massively parallel SIMD machine. We demonstrate progressive VQ on multispectral imagery obtained from the advanced very high resolution radiometer (AVHRR) and other earth-observation image data, and investigate the tradeoffs in selecting the number of decomposition levels and codebook training method.

Codebook organization to enhance maximum a posteriori detection of progressive transmission of vector quantized images over noisy channels

Abstract: We describe a new way to organize a full-search vector quantization codebook so that images encoded with it can be sent progressively and have resilience to channel noise. The codebook organization guarantees that the most significant bits (MSBs) of the codeword index are most important to the overall image quality and are highly correlated. Simulations show that the effective channel error rates of the MSBs can be substantially lowered by implementing a maximum a posteriori (MAP) detector similar to one suggested by Phamdo and Farvardin (see IEEE Trans. Inform. Theory, vol.40, no.1, p.156-193, 1994). The performance of the scheme is close to that of pseudo-gray coding at lower bit error rates and outperforms it at higher error rates. No extra bits are used for channel error correction.

Multi-stage target recognition using modular vector quantizers and multilayer perceptrons

Abstract: An automatic target recognition (ATR) classifier is proposed that uses modularly cascaded vector quantizers (VQs) and multilayer perceptrons (MLPs). A dedicated VQ codebook is constructed for each target class at a specific range of aspects, which is trained with the K-means algorithm and a modified learning vector quantization (LVQ) algorithm. Each final codebook is expected to give the lowest mean squared error (MSE) for its correct target class at a given range of aspects. These MSEs are then processed by an array of window MLPs and a target MLP consecutively. In the spatial domain, target recognition rates of 90.3 and 65.3 percent are achieved for moderately and highly cluttered test sets, respectively. Using the wavelet decomposition with an adaptive and independent codebook per sub-band, the VQs alone have produced recognition rates of 98.7 and 69.0 percent on more challenging training and test sets, respectively.

Pitch-synchronous speech coding by applying multiple analysis to select and align a plurality of types of code vectors

Abstract: A speech coder using a pitch synchronous innovation code excited linear prediction (PSI-CELP) speech coding system. The speech coder is capable of representing a portion which is not sufficiently represented by an adaptive codebook in a periodic portion of input speech and capable of improving the quality of reproduced speech. The periodicity corresponds to the pitch cycle of input speech by preliminarily reproducing speech from simple impulse trains. The speech coder depending on the particular embodiment includes an adaptive code book, a fixed code book, a noise code book, and a pulse codebook. A pulse code book stores a plurality of types of codevectors corresponding to pitch waveforms of voiced sounds. At the time of coding input speech, the pulse code book is searched.

Voice coder for coding voice signal with code-excited linear prediction coding

Abstract: A voice coder for coding a speech signal at a low bit rate with high speech quality and improved efficiency for gain quantization according to code-excited linear prediction (CELP) coding. The voice coder has an LPC (linear prediction coding) analyzer for calculating LPC coefficients, a parameter quantizer for quantizing the LPC coefficients to output a quantized code CL, an adaptive codebook, a long-term predicting circuit for searching the adaptive codebook to determine a delay code CD and an adaptive code vector, an excitation codebook, an excitation codebook searching circuit for determining an optimum quantized code CS and an excitation vector, and a gain codebook searching circuit for outputting a gain code CG by determining quantized gains representing quantized vectors of gains of the adaptive code vector and the excitation vector. The gain codebook searching circuit has a plurality of gain codebooks each for storing quantized gains corresponding to one of searching ranges divided by predetermined ranges with respect to the value of a searching parameter, and gain codebook selector for selecting one of the gain codebooks depending on the value of the searching parameter. The gain code CG is determined by using the gain codebook selected by the gain codebook selector.

Constrained and recursive hierarchical table-lookup vector quantization

Abstract: This paper presents techniques for the design of generic constrained and recursive vector quantizer encoders implemented by table-lookups. These vector quantizers include entropy-constrained VQ, tree structured VQ, classified VQ, product VQ, mean-removed VQ, multi-stage VQ, hierarchical VQ, nonlinear interpolative VQ, predictive VQ and weighted universal VQ. Our algorithms combine these different VQ structures with hierarchical table-lookup vector quantization. Thus the full-search encoder in the different VQ structures is replaced by a table-lookup encoder, which approximates the search, but the codebook structure and decoder are the same. In these table-lookup encoders, input vectors to the encoders are used directly as addresses in code tables to choose the codewords. In order to preserve manageable table sizes for large dimension VQs, we use hierarchical structures to quantize the vector successively in stages. Since both the encoder and decoder are implemented by table-lookups, there are no arithmetic computations required in the final system implementation. To further improve the subjective quality of the compressed images we use block transform based table-lookup vector quantizers with subjective distortion measures. There is no need to perform the forward or reverse transforms as they are implemented in the tables.

A new VLSI architecture for full-search vector quantization

Abstract: This paper presents a new systolic architecture that can be used to realize the full-search vector quantization (VQ) encoder for high-speed applications. The architecture possesses the features of regularity and modularity, and is thus very suitable for VLSI implementation. For a codebook of size N and dimension k, the VQ encoder has an area complexity of O(N), a time complexity of O(k), and I/O bandwidth of O(k). It reaches a compromise between the hardware cost and speed performance as compared to existing systolic/regular VQ encoders. At the current state of VLSI technology, the proposed system can easily be realized in a single chip for most practical applications. In addition, it provides flexibility in changing the codebook contents and extending the codebook size, where the latter is achieved simply by cascading some identical basic chips. With 0.8 /spl mu/m CMOS technology to implement the proposed VQ encoder for the case of N=256, K=16, and an input data wordlength of 8 bit, the chip requires a die size of about 5.5/spl times/8.9 mm/sup 2/ and is able for processing 6.25 M data vectors (or 100 M data samples) every second. These features show that the proposed architecture is attractive for use in high-speed image/video applications.

VQ-enhanced fractal image compression

Abstract: A novel hybrid scheme combining fractal image compression with mean-removed shape-gain vector quantization is presented. The scheme uses a small set of VQ codebook blocks as a block classifier for the domain blocks and as an alternative means of coding when able to provide a satisfying distortion. Our scheme is shown to improve the performance of conventional fractal coding in all its aspects. The rate-distortion curve is ameliorated, and both the encoding and the decoding are faster.

Reallocation of GLA codevectors for evading local minimum

Abstract: The performance of the Generalized Lloyd Algorithm (GLA) is improved by reallocating the codevectors every time GLA reaches a local optimum. This is done by splitting the largest partition and by merging two small neighboring partitions; thus preserving the size of the codebook. The whole procedure is repeated until no improvement is achieved.

Wideband re-synthesis of narrowband CELP-coded speech using multiband excitation model

Abstract: In this paper, a method for improving the quality of narrowband CELP-coded speech is present. The approach is to reduce the hoarse voice in CELP-coded speech by enhancing the pitch periodicity in the reproduction signal and also to reduce the muffing characteristics of narrowband speech by regenerating the highband components of speech spectra from the reproduction signal. In the proposed method, multiband excitation (MBE) analysis is performed on the reproduction speech signal from a CELP decoder and the pitch periodicity is enhanced by resynthesizing the speech signal using a harmonic synthesizer according to the MBE model. The highband magnitude spectra are regenerated by matching to lowband spectra using a trained wideband spectral codebook. Information about the voiced/unvoiced (V/UV) excitation in the highband are derived from a training procedure and then stored alongside with the wideband spectral codebook so that they can be recovered by indexing to the codebook using the matched lowband index. Simulation results indicate that the quality of the wideband resynthesized speech is significantly improved over the narrowband CELP-coded speech.