Advances in residual vector quantization: a review

Barnes, Christopher F.; Rizvi, Syed A.; Nasrabadi, Nasser M.

doi:10.1109/83.480761

Cited by 84 publications

(44 citation statements)

References 132 publications

(75 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Codes that can be used for multiresolution source description (whether or not that was the initial intention of their design) include treestructured scalar and vector quantizers [23]- [25], multistage or residual vector quantizers [26], [27], multiresolution transform codes (see, for example, [28], and the references therein), multiresolution trellis source codes [29], and multiresolution source codes combining wavelets or other frequency decompositions with zero-trees or other embedded codes [30]- [32]. We here focus on only the most closely related vector quantizers.…”

Section: Optimality Criteria For Code Designmentioning

confidence: 99%

Multiresolution Vector Quantization

Effros

Dugatkin

2004

IEEE Trans. Inform. Theory

View full text Add to dashboard Cite

Abstract-Multiresolution source codes are data compression algorithms yielding embedded source descriptions. The decoder of a multiresolution code can build a source reproduction by decoding the embedded bit stream in part or in whole. All decoding procedures start at the beginning of the binary source description and decode some fraction of that string. Decoding a small portion of the binary string gives a low-resolution reproduction; decoding more yields a higher resolution reproduction; and so on. Multiresolution vector quantizers are block multiresolution source codes. This paper introduces algorithms for designing fixed-and variable-rate multiresolution vector quantizers. Experiments on synthetic data demonstrate performance close to the theoretical performance limit. Experiments on natural images demonstrate performance improvements of up to 8 dB over tree-structured vector quantizers. Some of the lessons learned through multiresolution vector quantizer design lend insight into the design of more sophisticated multiresolution codes.Index Terms-Embedded source code design, fixed rate, multiuser, network, progressive transmission, successive refinement, variable rate.

show abstract

Section: Optimality Criteria For Code Designmentioning

confidence: 99%

Multiresolution Vector Quantization

Effros

Dugatkin

2004

IEEE Trans. Inform. Theory

View full text Add to dashboard Cite

show abstract

“…The methods proposed that satisfy this demand are the sequential searchdirect sum codebook multistage VQ (residual VQ) [2] and the lattice VQ [3]. The problem of the former method is that quantization losses are caused by mismatches in the distributions between the stages.…”

Section: Introductionmentioning

confidence: 99%

Performance improvement of vector quantizer with reflection group for uniform distribution on hyperspace

Yamane

Morikawa

Mae

et al. 2006

Electron Comm Jpn Pt III

View full text Add to dashboard Cite

SUMMARYA two-stage vector quantizer (kaleidoscope vector quantizer: KVQ) using multidimensional symmetry which has a reflection group is proposed as a high-speed vector quantizer (VQ) for vectors following a uniform distribution on a hypersphere. The first-stage VQ of this method is a lattice quantizer on a hypersphere based on a reflection group and plays a central role in achieving higher dimensions and higher rates. The last-stage VQ is a full-search vector quantizer and plays the roles of repartitioning and reforming the Voronoi regions and improving the quantization properties. In this method, the code vectors are assigned only inside the first-stage Voronoi regions. In this paper, a unified-region KVQ is proposed as a method for improving the quantization characteristic in the high dimensions of KVQ. This method also assigns the code vectors of the last-stage VQ to the boundary surfaces of the Voronoi regions of the first-stage VQ. By using this method, the multiple Voronoi regions in the first-stage VQ are merged and repartitioned, and the degree of freedom in the code vector assignment is increased. Simulation tests were conducted on Gaussian vectors and showed that the quantization characteristics equivalent to the full-search VQ were obtained in the range of 16 dimensions and a rate of about 2.5 bits/sample.

show abstract

“…However, if the direct sum codebook size (sum of each stage codebook size) becomes large, the memory associated with the codetable becomes excessive. Several techniques have been presented to control the memory cost, for example, the entropyconstrained RVQ (EC-RVQ), predictive RVQ (PRVQ), and finite-state RVQ (FSRVQ) [4][5][6][7][8]. However, the design and implementation of the techniques are very complicated.…”

Section: Introductionmentioning

confidence: 99%

“…The third stage quantizes the second stage error output to provide a further refinement and so on. The error vector is also called residue; thus the MSVQ is sometimes referred to as residual VQ (RVQ) [4][5][6][7][8]. The output of each stage quantizer is a quantization index.…”

Section: Introductionmentioning

confidence: 99%