The direct implementation of the conventional vector quantization codec requires unfeasibly large-sized codebooks as the block length increases. This paper discusses a systematic approach for constructing vector quantization codec with large block length which can be compared with that of the transform coding techniques. Here we propose a class of Recursive Vector Quantizer (RVQ) which recursively encodes a given large-dimensional input vector into a series of indices of reproduction vectors derived from a small-sized and small-dimensional codebook. This codebook is referred to as a wavelet codebook. Note that a single codebook will be used repeatedly in every stage of the hierarchical quadtree decomposition of input vectors. For this construction of the RVQ system, the mean value of each input vector is extracted and then encoded separately. The side information, which represents how the wavelet vectors are combined for reproducing the replica of the input vector, can be efficiently encoded by using the binary tree code. We also give a design example of a 64-dimensional RVQ using a four-dimensional tree search vector quantizer as a wavelet quantizer. The results of computer simulation show effectiveness of the RVQ for video signals. For example, the signal-to-noise ratio of 37.6 dB is obtained at the rate of 1.27 bits per pixel for the image data
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copy
Yoshio YAMADA, Saburo TAZAKI, "Recursive Vector Quantization for Monochrome Video Signals" in IEICE TRANSACTIONS on Information,
vol. E74-D, no. 2, pp. 399-405, February 1991, doi: .
Abstract: The direct implementation of the conventional vector quantization codec requires unfeasibly large-sized codebooks as the block length increases. This paper discusses a systematic approach for constructing vector quantization codec with large block length which can be compared with that of the transform coding techniques. Here we propose a class of Recursive Vector Quantizer (RVQ) which recursively encodes a given large-dimensional input vector into a series of indices of reproduction vectors derived from a small-sized and small-dimensional codebook. This codebook is referred to as a wavelet codebook. Note that a single codebook will be used repeatedly in every stage of the hierarchical quadtree decomposition of input vectors. For this construction of the RVQ system, the mean value of each input vector is extracted and then encoded separately. The side information, which represents how the wavelet vectors are combined for reproducing the replica of the input vector, can be efficiently encoded by using the binary tree code. We also give a design example of a 64-dimensional RVQ using a four-dimensional tree search vector quantizer as a wavelet quantizer. The results of computer simulation show effectiveness of the RVQ for video signals. For example, the signal-to-noise ratio of 37.6 dB is obtained at the rate of 1.27 bits per pixel for the image data
URL: https://global.ieice.org/en_transactions/information/10.1587/e74-d_2_399/_p
Copy
@ARTICLE{e74-d_2_399,
author={Yoshio YAMADA, Saburo TAZAKI, },
journal={IEICE TRANSACTIONS on Information},
title={Recursive Vector Quantization for Monochrome Video Signals},
year={1991},
volume={E74-D},
number={2},
pages={399-405},
abstract={The direct implementation of the conventional vector quantization codec requires unfeasibly large-sized codebooks as the block length increases. This paper discusses a systematic approach for constructing vector quantization codec with large block length which can be compared with that of the transform coding techniques. Here we propose a class of Recursive Vector Quantizer (RVQ) which recursively encodes a given large-dimensional input vector into a series of indices of reproduction vectors derived from a small-sized and small-dimensional codebook. This codebook is referred to as a wavelet codebook. Note that a single codebook will be used repeatedly in every stage of the hierarchical quadtree decomposition of input vectors. For this construction of the RVQ system, the mean value of each input vector is extracted and then encoded separately. The side information, which represents how the wavelet vectors are combined for reproducing the replica of the input vector, can be efficiently encoded by using the binary tree code. We also give a design example of a 64-dimensional RVQ using a four-dimensional tree search vector quantizer as a wavelet quantizer. The results of computer simulation show effectiveness of the RVQ for video signals. For example, the signal-to-noise ratio of 37.6 dB is obtained at the rate of 1.27 bits per pixel for the image data
keywords={},
doi={},
ISSN={},
month={February},}
Copy
TY - JOUR
TI - Recursive Vector Quantization for Monochrome Video Signals
T2 - IEICE TRANSACTIONS on Information
SP - 399
EP - 405
AU - Yoshio YAMADA
AU - Saburo TAZAKI
PY - 1991
DO -
JO - IEICE TRANSACTIONS on Information
SN -
VL - E74-D
IS - 2
JA - IEICE TRANSACTIONS on Information
Y1 - February 1991
AB - The direct implementation of the conventional vector quantization codec requires unfeasibly large-sized codebooks as the block length increases. This paper discusses a systematic approach for constructing vector quantization codec with large block length which can be compared with that of the transform coding techniques. Here we propose a class of Recursive Vector Quantizer (RVQ) which recursively encodes a given large-dimensional input vector into a series of indices of reproduction vectors derived from a small-sized and small-dimensional codebook. This codebook is referred to as a wavelet codebook. Note that a single codebook will be used repeatedly in every stage of the hierarchical quadtree decomposition of input vectors. For this construction of the RVQ system, the mean value of each input vector is extracted and then encoded separately. The side information, which represents how the wavelet vectors are combined for reproducing the replica of the input vector, can be efficiently encoded by using the binary tree code. We also give a design example of a 64-dimensional RVQ using a four-dimensional tree search vector quantizer as a wavelet quantizer. The results of computer simulation show effectiveness of the RVQ for video signals. For example, the signal-to-noise ratio of 37.6 dB is obtained at the rate of 1.27 bits per pixel for the image data
ER -