Quantization of wavelet coefficients for image compression

Aziz Mohammed, Khalid Sayood

Research output: Contribution to journalConference article

Abstract

The use of wavelets and multiresolution analysis is becoming increasingly popular for image compression. In this paper we examine several different approaches to the quantization of wavelet coefficients. A standard approach in subband coding is to use DPCM to encode the lowest band while the higher bands are quantized using a scalar quantizer for each band or a vector quantizer. We implement these schemes using a variety of quantizer including pdf optimized quantizers and recursively indexed scalar quantizers (RISQ). We then incorporate a threshold operation to prevent the removal of perceptually important information. We show that there is a both subjective and objective improvements in performance when we use the RISQ and the perceptual thresholds. The objective performance measure shows a consistent two to three dB improvement over a wide range of rates. Finally we use a recursively indexed vector quantizer (RIVQ) to encode the wavelet coefficients. The RIVQ can operate at relatively high rates and is therefore particularly suited for quantizing the coefficients in the lowest band.

Original languageEnglish (US)
Number of pages1
JournalData Compression Conference Proceedings
StatePublished - Jan 1 1995
EventProceedings of the 5th Data Compression Conference - Snowbird, UT, USA
Duration: Mar 28 1995Mar 30 1995

Fingerprint

Quantization (signal)
Image compression
Multiresolution analysis
Wavelet analysis

ASJC Scopus subject areas

  • Hardware and Architecture
  • Electrical and Electronic Engineering

Cite this

Quantization of wavelet coefficients for image compression. / Mohammed, Aziz; Sayood, Khalid.

In: Data Compression Conference Proceedings, 01.01.1995.

Research output: Contribution to journalConference article

@article{ccd2caf2ef074851a17d4ef6dfc5c593,
title = "Quantization of wavelet coefficients for image compression",
abstract = "The use of wavelets and multiresolution analysis is becoming increasingly popular for image compression. In this paper we examine several different approaches to the quantization of wavelet coefficients. A standard approach in subband coding is to use DPCM to encode the lowest band while the higher bands are quantized using a scalar quantizer for each band or a vector quantizer. We implement these schemes using a variety of quantizer including pdf optimized quantizers and recursively indexed scalar quantizers (RISQ). We then incorporate a threshold operation to prevent the removal of perceptually important information. We show that there is a both subjective and objective improvements in performance when we use the RISQ and the perceptual thresholds. The objective performance measure shows a consistent two to three dB improvement over a wide range of rates. Finally we use a recursively indexed vector quantizer (RIVQ) to encode the wavelet coefficients. The RIVQ can operate at relatively high rates and is therefore particularly suited for quantizing the coefficients in the lowest band.",
author = "Aziz Mohammed and Khalid Sayood",
year = "1995",
month = "1",
day = "1",
language = "English (US)",
journal = "Proceedings of the Data Compression Conference",
issn = "1068-0314",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - JOUR

T1 - Quantization of wavelet coefficients for image compression

AU - Mohammed, Aziz

AU - Sayood, Khalid

PY - 1995/1/1

Y1 - 1995/1/1

N2 - The use of wavelets and multiresolution analysis is becoming increasingly popular for image compression. In this paper we examine several different approaches to the quantization of wavelet coefficients. A standard approach in subband coding is to use DPCM to encode the lowest band while the higher bands are quantized using a scalar quantizer for each band or a vector quantizer. We implement these schemes using a variety of quantizer including pdf optimized quantizers and recursively indexed scalar quantizers (RISQ). We then incorporate a threshold operation to prevent the removal of perceptually important information. We show that there is a both subjective and objective improvements in performance when we use the RISQ and the perceptual thresholds. The objective performance measure shows a consistent two to three dB improvement over a wide range of rates. Finally we use a recursively indexed vector quantizer (RIVQ) to encode the wavelet coefficients. The RIVQ can operate at relatively high rates and is therefore particularly suited for quantizing the coefficients in the lowest band.

AB - The use of wavelets and multiresolution analysis is becoming increasingly popular for image compression. In this paper we examine several different approaches to the quantization of wavelet coefficients. A standard approach in subband coding is to use DPCM to encode the lowest band while the higher bands are quantized using a scalar quantizer for each band or a vector quantizer. We implement these schemes using a variety of quantizer including pdf optimized quantizers and recursively indexed scalar quantizers (RISQ). We then incorporate a threshold operation to prevent the removal of perceptually important information. We show that there is a both subjective and objective improvements in performance when we use the RISQ and the perceptual thresholds. The objective performance measure shows a consistent two to three dB improvement over a wide range of rates. Finally we use a recursively indexed vector quantizer (RIVQ) to encode the wavelet coefficients. The RIVQ can operate at relatively high rates and is therefore particularly suited for quantizing the coefficients in the lowest band.

UR - http://www.scopus.com/inward/record.url?scp=0029225938&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0029225938&partnerID=8YFLogxK

M3 - Conference article

JO - Proceedings of the Data Compression Conference

JF - Proceedings of the Data Compression Conference

SN - 1068-0314

ER -