Chair: Xinhua Zhang, University of Missouri, USA
Marie Chabert, Toulouse (France)
Jean-Yves Tourneret, Toulouse (France)
Francis Castani, Toulouse (France)
This work applies the Continuous Wavelet Transform (CWT) to Abrupt Change (AC) detection in zeromean multiplicative noise. An AC signature can be defined because of CWT translation invariance property. The problem is then signature detection in the time-scale domain. A contrast criterion is defined as a second-order measure of performance. This contrast depends on the first and second order moments of the multiplicative process CWT. An optimal wavelet maximizing the contrast is derived for an ideal step in white multiplicative noise. In this fundamental case, the Signal to Noise Ratios (SNR) in the time domain and in the time-scale domain are compared. A scale exists above which SNR is larger on the CWT maxima curve than in the time domain. An asymptotically optimal wavelet is derived for smoothed AC.
Y. Li, Memorial University of Newfoundland (Canada)
C. Moloney, Memorial University of Newfoundland (Canada)
Speckle noise is one characteristic of Synthetic Aperture Radar (SAR) images. The primary goal of speckle smoothing of SAR images is to reduce the speckle noise without sacrificing information content. Various speckle filters have been devised to smooth speckle in the spatial domain. In this paper, we perform speckle reduction in the wavelet domain. A hierarchical correlation is defined which takes into account both the inter- and intra-band correlation among wavelet coefficients. According to this definition, the correlation values at edge positions are larger than those for nonedge positions. We use this correlation map to distinguish edge coefficients from noise coefficients and thus perform selective soft-thresholding on the wavelet coefficients. The proposed method is applied to airborne SAR images and the results are compared with Donoho's original soft-thresholding and the well-known Lee multiplicative speckle filter. Test results show that this method can substantially smooth noise while preserving major edge structures in images.
Sreela Sasi, Wayne State University (U.S.A.)
Loren Schwiebert, Wayne State University (U.S.A.)
Jatinder Bedi, Wayne State University (U.S.A.)
This paper presents a novel method for handwritten character recognition using wavelet packet transform and fuzzy logic. This method exploits the time-frequency localization and compression capability of wavelet packet transform, using the best basis algorithm to enhance the accuracy of recognition at the pixel level and the computational capability of fuzzy logic with linguistic variables, which is a universal approximator if it uses enough rules. The best basis algorithm automatically adapts the transform to best match the characteristics of the signal, as well as minimize the additive cost function. The wavelet packet transform of the handwritten characters are taken using the best basis algorithm. The standard deviation of the spread of the coefficients in each multi- resolution level axe computed, which forms the characteristic features for the characters. These features are given as input to the fuzzy logic character recognition system, where these are fuzzified, analyzed, and the corresponding. characters are given as output using IF ... THEN rules. This method is more efficient for handwritten character recognition than energy sorted wavelet transform of character images, since it contains only a few edges in the image. Simulation of four multi-resolution levels for each character is done using symmlet and results show that they have better accuracy than the methods using only fuzzy logic.
Bing-Bing Chai, University of Missouri-Columbia (U.S.A.)
Jozsef Vass, University of Missouri-Columbia (U.S.A.)
Xinhua Zhuang, University of Missouri-Columbia (U.S.A.)
Recent success in wavelet image coding is mainly attributed to recognition of the importance of data organization and representation. Several very competitive wavelet coders have been developed, namely, Shapiro's embedded zerotree wavelets (EZW), Servetto et al.'s morphological representation of wavelet data (MRWD), and Said and Pearlman's set partitioning in hierarchical trees (SPIHT). In this paper, we develop a novel wavelet image coder called significance-linked connected component analysis (SLCCA) of wavelet coefficients that exploits both within-subband clustering of significant coefficients and cross-subband dependency in significant fields. Extensive computer experiments show that the proposed SLCCA outperforms all three aforementioned wavelet coders. For example, for the "Barbara" image, at 0.50 bpp SLCCA outperforms EM and SPIHT by 1.75 dB and 0.89 dB in PSNR respectively. It is also observed that SLCCA works extremely well for images with large texture regions. For eight typical 256 x 256 grayscale texture images compressed at 0.40 bpp, SLCCA outperforms SPIHT by 0.32 dB0.70 dB. This outstanding performance is achieved without any optimal bit allocation procedure. Thus both the encoding and decoding procedures are fast.
M. Pawlak, University of Manitoba (Canada)
Z. Hasiewicz, Technical University of Wroclaw (Poland)
This paper deals with the problem of reconstruction of nonlinearities in a certain class of nonlinear systems of composite structure from their input-output observations when a prior information about the system is poor, thus excluding the standard parametric approach to the problem. The multiresolution idea, being the fundamental concept of modern wavelet theory, is adopted and is applied to construct nonparametric identification techniques of nonlinear characteristics. The pointwise convergence properties of the proposed identification algorithms are established.
Anestis Karasaridis, University of Toronto (Canada)
Dimitrios Hatzinakos, University of Toronto (Canada)
In this paper, we present and discuss measurements of Webcasting and aggregated Web traffic in our research group's local area network. The Webcasting traffic is multiplexed to simulate the effect of having many clients running simultaneously webcasting software in the background. The multiplexed Webcasting and the aggregated Web traffic appear to be asymptotically self-similar. The a-stable self-similar stochastic process, originally proposed in  to model aggregated Ethernet LAN and WAN traffic, is applied to the new measurements and the results, implications and extensions are discussed.