Finite-state Vector Quantization Research Articles

Image compression using finite-state vector quantization with derailment compensation

Presents an image compression method using a finite-state vector quantizer (FSVQ) with derailment compensation. The derailment problem in conventional FSVQ systems is discussed first. Then a novel scheme is presented to overcome the problem. In this scheme, each state is partitioned into K Voronoi regions. The corresponding region codebook is designed individually. A simple initial classifier based on a mean codebook is also presented to determine the state and the region for an input vector. The results show that the derailment problem has been completely attacked. Experimental results indicate that the new FSVQ system is very computationally efficient and achieves good picture quality at an average rate of 0.33 b per pixel (bpp).< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Variable-rate finite-state vector quantization and applications to speech and image coding

Two variable rate versions of the finite-state vector quantizer introduced by J. Foster, M.O. Dunham, and R.M. Gray (FDG) (1985) are presented, with the possibility of using structured as well as unstructured state codebooks. In the first scheme, the state codebook sizes can be different for different states, implying different rate distribution among the states. In the second scheme, in addition to this flexibility, pruned tree-structured vector quantizers are used as the state quantizers, i.e., each of the state quantizers can be a variable-rate encoder. For encoding sampled speech data, both schemes perform significantly better than the fixed-rate FDG scheme, and the second scheme gives the best performance of all. The 2-D extension of the above schemes is considered, and two low-bit rate image coding systems based on them are described.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Image coding using variable-rate side-match finite-state vector quantization

Future B-ISDN (broadband integrated services digital network) users will be able to send various kinds of information, such as voice, data, and image, over the same network and send information only when necessary. It has been recognized that variable-rate encoding techniques are more suitable than fixed-rate techniques for encoding images in a B-ISDN environment. A new variable-rate side-match finite-state vector quantization with a block classifier (CSMVQ) algorithm is described. In an ordinary fixed-rate SMVQ, the size of the state codebook is fixed. In the CSMVQ algorithm presented, the size of the state codebook is changed according to the characteristics of the current vector which can be predicted by a block classifier. In experiments, the improvement over SMVQ was up to 1.761 dB at a lower bit rate. Moreover, the improvement over VQ can be up to 3 dB at nearly the same bit rate.

Side match and overlap match vector quantizers for images

A class of vector quantizers with memory that are known as finite state vector quantizers (FSVQs) in the image coding framework is investigated. Two FSVQ designs, namely side match vector quantizers (SMVQs) and overlap match vector quantizers (OMVQs), are introduced. These designs take advantage of the 2-D spatial contiguity of pixel vectors as well as the high spatial correlation of pixels in typical gray-level images. SMVQ and OMVQ try to minimize the granular noise that causes visible pixel block boundaries in ordinary VQ. For 512 by 512 gray-level images, SMVQ and OMVQ can achieve communication quality reproduction at an average of 1/2 b/pixel per image frame, and acceptable quality reproduction. Because block boundaries are less visible, the perceived improvement in quality over ordinary VQ is even greater. Owing to the structure of SMVQ and OMVQ, simple variable length noiseless codes can achieve as much as 60% bit rate reduction over fixed-length noiseless codes.

Image sequence coding using adaptive finite-state vector quantization

A coding algorithm must have the ability to adapt to changing image characteristics for image sequences. An adaptive finite-state vector quantization (FSVQ) in which the bit rate and the encoding time can be reduced is described. In order to improve the image quality and avoid producing a wrong state for an input vector, a threshold is used in FSVQ to decide whether to switch to a full searching VQ. The codebook is conditionally replenished according to a distortion threshold at a later time to reflect the local statistics of the current frame. After the codebook is replenished, one can quickly reconstruct the state codebooks of FSVQ using the state codebook selection algorithm. In the experiments, the improvement over the static SMVQ is up to 2.40 dB at nearly the same bit rate and the encoding time is only one-ninth the time required by the static SMVQ. Moreover, the improvement over the static VQ is up to 2.91 dB, and the encoding time is only three-fifths the time required by the static VQ for the image sequence 'Claire'. >

Tandem transcoding without distortion accumulation for vector quantization

Transcoding algorithms that eliminate distortion accumulation due to tandem transcodings between memoryless, finite-state, and predictive vector quantization and pulse code modulation (PCM) are presented. The algorithms can be implemented using table lookups for memoryless and finite-state vector quantization, whereas predictive vector quantization requires online calculations. Computer simulations indicate a 6 dB improvement in the case of 16 kb/s predictive vector quantizers, 48 kb/s PCM, and four tandems for speech.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

On the design of connectionist vector quantizers

The Kohonen self-organizing feature map has been widely used for the design of connectionist vector quantizers (VQ). One of the features of the Kohonen algorithm is that the weight update gain sequence η ( m) is a decreasing function of the number of iterations, and if incorrectly chosen can lead to very long training times. Here we derive the time-optimal gain sequence and demonstrate its efficacy for a number of cases. The performance is demonstrated using a Gauss-Markov source and compared with a VQ designed using the LBG algorithm. It is demonstrated that the new method is time optimal and that its performance tends to that of the VQ with the LBG algorithm. The special case of a connectionist VQ for linear predictive or auto-regressive data is analysed. The optimal design is derived for the Itakura distance measures. It is again compared with the LBG design. Two connectionist models for finite-state vector quantizers (FSVQ) are proposed. These models learn not only the distribution but also the conditional distribution of a source. The connectionist FSVQs exploit the redundancy between the vectors (frames) of a highly correlated source, and further improve the rate-distortion performance. Their architectures and learning algorithms are presented. To evaluate the performance of the proposed quantizers, comparisons between quantized and unquantized speech spectra are presented. The connectionist VQ and FSVQ for linear predictive data are finally applied to a multipulse linear predictive speech coder using data from the TIMIT database. Comparisons are made of waveforms and rate-distortion functions.

Two-stage multirate coding of color images

Various procedures for coding color images have been described in the literature. Several of these have been based on subband coding, including some early results by the authors of this paper and Bruder, in which vector quantization and ADPCM were utilized effectively to encode the subband images. High quality coded images were achieved at bit-rates of approximately 0.6 bit/pixel for 512 x 512 color images. This paper introduces an improved system which evolved as an attempt to maintain the quality of the subband coder at lower bit-rates. The approach is based on a three component multirate representation of the image, composed of a chrominance sub-image pair, a medium-to-low frequency luminance image, and a high frequency luminance residual. Finite State Vector Quantization and Differential PCM are used to code these components. In addition, the paper discusses subband coding, vector quantization and predictive schemes, and provides some comparisons at low bit-rates. Simulation results indicate that noticeable subjective quality improvement can be achieved using the two-stage multirate scheme relative to the other coders that were considered.

AN EFFICIENT SPEECH RECOGNITION SYSTEM FOR THE INITIALS OF MANDARIN SYLLABLES

In a long-term research project, the recognition of Mandarin speech for very large vocabulary and unlimited text is considered. Its first stage goal is to recognize the Mandarin syllables. In a previous paper, an initial/final two-phase recognition approach to recognize these very confusing syllables was proposed, in which each syllable is divided into initial and final parts and recognized separately, and efficient recognition techniques for the finals were proposed and discussed. This paper serves as a continuation and proposes an efficient system to recognize the Mandarin initials. In this system, a classification procedure is first used to categorize the unknown initials into two groups C1 and C2; different approaches are then separately applied and independently optimized to recognize C1 and C2. It is found that Finite State Vector Quantization (FSVQ) is very useful, whose two modified versions, Modified FSVQ (MFSVQ) and the Second Order FSVQ (SOFSVQ), can provide the best recognition performance for C1 and C2 by carefully adjusting a design parameter called characteristic interval. Experimental results show that a recognition rate of 94.1% to 94.7% can be achieved using this system. Such a design is accomplished by carefully considering the special characteristics of Mandarin syllables and initials.

Statistical methods in multi-speaker automatic speech recognition

Automatic speech recognition and understanding (ASR) plays an important role in the framework of man-machine communication. Substantial industrial developments are at present in progress in this area. However, after 40 years or so of efforts several fundamental questions remain open. This paper is concerned with a comparative study of four different methods for multi-speaker word recognition: (i) clustering of acoustic templates, (ii) comparison with a finite state automaton, (iii) dynamic programming and vector quantization, (iv) stochastic Markov sources. In order to make things comparable, the four methods were tested with the same material made up of the ten digits (0 to 9) pronounced four times by 60 different speakers (30 males and 30 females). We will distinguish in our experiments between multi-speaker systems (capable of recognizing words pronounced by speakers that have been used during the training phase of the system) and speaker-independent systems (capable of recognizing words pronounced by speakers totally unknown to the system). Half of the corpus (15 male and 15 female) were used for training, and the remaining part for test.

Recursive subband image coding with adaptive prediction and finite state vector quantization

This paper discusses a method for subband image coding which employs a number of novel techniques and concepts. Recursive exact reconstruction filters are used for analysis and synthesis to achieve computational savings. Issues related to the filter bank design criteria and the impact of these filters on the subjective quality of decoded images are discussed. A new coding system is introduced which combines the advantages of analysis-by-synthesis, linear prediction and finite state vector quantization to obtain higher compression ratios. This method is used for encoding the lowest frequency subband and is shown to yield improvement over the commonly used methods of differential pulse code modulation and adaptive differential pulse code modulation. A multi-codebook vector quantization strategy is introduced for coding the high frequency subbands which enables inter-band correlation to be exploited. Experimental results and comparisons with previous work are presented which show the utility of this approach.

Image Compression Based On Vector Quantization With Finite Memory

Image compression using memoryless vector quantization (VQ), in which small blocks (vectors) of pixels are independently encoded, has been demonstrated to be an effective technique for achieving bit rates above 0.6 bits per pixel (bpp). To maintain the same quality at lower rates, it is necessary to exploit spatial redundancy over a larger region of pixels than is possible with memoryless VQ. This can be achieved by incorporating memory of previously encoded blocks into the encoding of each successive input block. Finite-state vector quantization (FSVQ) employs a finite number of states, which summarize key information about previously encoded vectors, to select one of a family of codebooks to encode each input vector. In this paper, we review the basic ideas of VQ and extend the finite-state concept to image compression. We introduce a novel for-mulation of the state and state-transition rule that uses a perceptually based edge classifier. We also examine the use of interpolation in conjunction with VQ with finite memory. Coding results are presented for monochrome images in the bit-rate range of 0.24 to 0.32 bpp. The results achieved with finite memory are comparable to those of memoryless VQ at 0.6 bpp and show that there are significant gains to be obtained by enhancing the basic VQ approach with interblock memory.

Simulation of Vector Trellis Encoding Systems

Most tree or trellis encoding data compression systems use decoders which form scalar outputs as a (possibly nonlinear) function of the contents of a shift register containing received channel symbols. We here develop design algorithms for trellis encoding systems having decoders not constrained to have such a scalar sliding-block code structure. In particular, we consider using the decoder of a finite-state vector quantizer together with a vector trellis search. Simulation results are presented for vector trellis encoding systems for Gauss-Markov sources, sampled speech data, and LPC speech data.

Finite-state vector quantization for waveform coding

A finite-state vector quantizer is a finite-state machine used for data compression: Each successive source vector is encoded into a codeword using a minimum distortion rule, and into a code book, depending on the encoder state. The current state and the selected codeword then determine the next encoder state. A finite-state vector quantizer is capable of making better use of the memory in a source than is an ordinary memoryless vector quantizer of the same dimension or blocklength. Design techniques are introduced for finite-state vector quantizers that combine ad hoc algorithms with an algorithm for the design of memoryless vector quantizers. Finite-state vector quantizers are designed and simulated for Gauss-Markov sources and sampled speech data, and the resulting performance and storage requirements are compared with ordinary memoryless vector quantization.

An Algorithm for the Design of Labeled-Transition Finite-State Vector Quantizers

A finite-state vector quantizer (FSVQ) is a switched vector quantizer where the sequence of quantizers selected by the encoder can be tracked by the decoder. It can be viewed as an adaptive vector quantizer with backward estimation, a vector generalization of an AQB system. Recently a family of algorithms for the design of FSVQ's for waveform coding application has been introduced. These algorithms first design an initial set of vector quantizers together with a next-state function giving the rule by which the next quantizer is selected. The codebooks of this initial FSVQ are then iteratively improved by a natural extension of the usual memoryless vector quantizer design algorithm. The next-state function, however, is not modified from its initial form. In this paper we present two extensions of the FSVQ design algorithms. First, the algorithm for FSVQ design for waveform coders is extended to FSVQ design of linear predictive coded (LPC) speech parameter vectors using an Itakura-Saito distortion measure. Second, we introduce a new technique for the iterative improvement of the next-state function based on an algorithm from adaptive stochastic automata theory. The design algorithms are simulated for an LPC FSVQ and the results are compared with each other and to ordinary memoryless vector quantization. Several open problems suggested by the simulation results are presented.