
CSE300 Multimedia Systems: Student Work
Literature Survey on Wavelet Compression of Still Color Images
 Orthonormal Wavelets With Balanced Uncertainty
Monro, D.M., Bassil, B.E. and Dickson, G.J.
Proc. IEEE ICIP 1996, Vol. 2, pp. 581584
This paper tries to determine the "best" wavelet by using the psychovisual evaluation of selected images. The wavelets under consideration are modified to trade off between time and frequency uncertainty. The authors note that by selecting a nearly balanced ratio between the two produces the best subjective results for image compression.
 A Fast Wavelet Based KarhunenLoeve Transform
Greenshields, I.R. and J.A. Rosiene
Pattern Recognition, 31, 7 1998
The KarhuhenLoeve Transform is computationally expensive, but provides the best basis system for representing information. To avoid this complexity, many "fast" algorithms make potentially incorrect assumptions about the statistics of the data. In this paper, the authors take advantage of some of the processing done by the wavelet decomposition to avoid the high cost of computing the full KLT.
 Optimized Wavelets for Fingerprint Compression
Sherlock, B.G. and Monro, D.M.
Proc. IEEE ICASSP 1996, Vol. 3, pp. 14471450
This paper discusses how orthonormal wavelets can be optimized for finger printing. The basic technique is an iterative process of analysis and synthesis followed by rms error comparison. Psychovisual testing was performed which supported the rms error between the original and recovered images.
 SpaceFrequency Balance in Biorthogonal Wavelets
D.M. Monro and B.G. Sherlock
Proc. IEEE ICIP 1997, Vol. 1, pp. 624627
Wavelets can be optimized by deciding between space and frequency uncertainty. The authors derive a metric that suggests the optimal tradeoff point. The goal is a fast iterative process that can produce a FIR filter bank that will satisfy their metric.
 On the Space of Orthonormal Wavelets
B.G. Sherlock and D.M. Monro
IEEE Transactions on Signal Processing, June 1998, Vol. 46, No. 6, pp. 17161720
This paper demonstrates how one can compute othornomal wavelets recursively. The goal is to be able to produce all wavelets in a given twodimensional space without repetition so that they may be tested exhaustively. Source code is provided in this paper that will produce the FIR coefficients for any even length filter bank.
 Approximate Continuous Wavelet Transform with an Application to Noise Reduction
J. M. Lewis and C. S. Burrus
Proceedings of ICASSP 1998
This paper looks at the characteristics of the wavelets that may be selected when computing the DWT. In particular, the Daubechies family of wavelets are examined and the coefficients are adapted. The modifications to the coefficients that the author makes are for noise reduction and edge detection.
 Smooth Biorthogonal Wavelets for Applications in Image Compression
J. E. Odegard and C. S. Burrus
Workshop, Loen, Norway, September 1996
This paper looks at a new class of biorthogonal wavelet filters. The authors show the standard equation that must be solved in order to generate a biorthogonal wavelet. They then show by variation of the vanishing moments, that a smoother filter may be created.
 Discrete Finite Variation: A New Measure of Smoothness for the Design of Wavelet Basis
J. E. Odegard and C. S. Burrus
Proceedings of ICASSP, Atlanta, GA, May 1996
This paper recognizes that smooth biorthogonal wavelets are important for image compression. Since artifacts are introduced primarily by the non smooth portions of the wavelet it is important to make the wavelets as smooth as possible. The authors of this paper propose a metric in an effort to determine the "best" wavelet.
 Design of Linear Phase Cosine Modulated Filter Banks for Subband Image Compression
J. E. Odegard, R. A. Gopinath, C. S. Burrus
Rice University CML Technical Report CML TR9406
The authors of this paper focus on the three basic artifacts with image compression, blocking, blurring, and ringing. Shorter or longer length filter banks are discussed as tradeoffs between the two common wavelet artifacts, blurring and ringing. This paper provides only a quick overview to research that is further discussed below.
 WaveletBased Signal Processing Using Hidden Markov Models
M. S. Crouse, R. D. Nowak, and R. G. Baraniuk
To appear in IEEE Transactions on Signal Processing (Special Issue on Wavelets and Filterbanks), 1998
In this paper, the authors discuss the clustering and persistence characteristics of wavelets. They look at the coefficients that result from wavelets as being modeled by Guassian and nonGaussian statistics. The concept of a Hidden Markov Model (HMM) allows the authors to further exploit the interdependency between coefficients than the simple quadtree structure that Shapiro originally recognized.
 Improved Wavelet Denoising via Empirical Wiener Filtering
S. Ghael, A. M. Sayeed, and R. G. Baraniuk
Proc. SPIE, San Diego, July 1997
This paper looks at wavelets in the signal processing domain. Some of the goals with this use of wavelets include removing noise from signals. The example the authors offer is a noisy Doppler test signal which they recover by wavelet filtering.
 Wavelet Based SAR Speckle Reduction and Image Compression
J. E. Odegard, H. Guo, M. Lang, C. S. Burrus, R. O. Wells Jr., L. M. Novak and M. Hiett
SPIE Symposium on OE/Aerospace Sensing and Dual Use Photonics, Algorithm for Synthetic Aperture Radar Imagery II, 1721 April 1995, Orlando, FL., Also a Rice University Tech. Report CML TR9508
This paper looks at compressing and enhancing SAR images. Typical images are huge and need to be transmitted back down to Earth, so compression is essential. The authors note that after wavelet compression, the resulting image will be despeckled and therefore more highly correlated. With this higher correlation, it is possible to further compress the image before transmitting.
 WaveletBased PostProcessing of Low Bit Rate Transform Coded Images
R. Gopinath, M. Lang, H. Guo, J. E. Odegard
IEEE Proc. ICIP, Nov. 1994, Also a Rice University CML Technical Report CML TR9415
JPEG decompression may introduce blocking artifacts into the resulting image if the bit rate is too low. To correct for this, a wavelet filter may be passed over the image after processing to help smooth out the artifacts. This paper focuses on the problems with under and overcompensating resulting in blocky or overly smoothed images.
 Enhancement of Decompressed Images at Low Bit Rates
R. Gopinath, M. Lang, H. Guo, J. E. Odegard
SPIE, Mathematical Imaging: Wavelet Applications in Signal Processing and Image Processing II, Vol. 2303, Jul. 1994, Also a Rice University CML Technical Report CML TR9405
In this paper, the authors examine the problem of blocking in the DCT. They review a variety of techniques to remove artifacts from the image including a wavelet technique. This postprocessing effect on decompressed JPEG images helps reduce the blocking effect while maintaining high definition lines in the image.
 A Simple Scheme for Adapting TimeFrequency Representations
D. L. Jones, R. G. Baraniuk
IEEE Transactions on Signal Processing, Vol. 42, No. 12, pp. 35303535, December 1994
The Heisenberg uncertainty between time and frequency can be optimally reduced through extensive knowledge of the signal to be encoded. However, this is computationally prohibitive, so efficient methods of decided where to trade off are needed. The authors of this paper propose a solution that uses a single free parameter to facilitate the analysis of the tradeoffs.
Last Modified: December 13, 1998

Barry E. Mapen
