Chair: Doina Petrescu, Tampere University of Technology, Finland
Omer N. Gerek, Bilkent University (Turkey)
A. Enis Cetin, Bilkent University (Turkey)
Efficient compression of binary textual images is an important issue in document archiving. Most popular textual image compression methods exploit the fact that document images are composed of repeating character images. By determining the locations of the characters, a textual image can be efficiently coded. The image can then be reconstructed from a character library according to the character locations. In this paper, new nonlinear subband decomposition structures are presented and they are used in a textual image compression scheme in which the character libraries and locations are determined in the subband domain.
Luc Van Eycken, Katholieke Universiteit Leuven (Belgium)
In this paper, we'll describe the results of the Human Capital & Mobility cooperation network "Model based Analysis of Video Information" (MAVI). This network focused its cooperation on non-linear image processing, feature extraction and image sequence segmentation, and model based video compression techniques. To complement the results, shown in this paper, a list of joint publications with more scientific details is also provided.
Doina Petrescu, Tampere University of Technology (Finland)
Moncef Gabbouj, Tampere University of Technology (Finland)
This paper proposes adaptive Boolean filters for realizing the prediction stage in lossless compression of greylevel images. The adaptation of Boolean filters is performed using local information about the presence and orientation of edges (high local amplitude variations) of the pixels inside the prediction mask. It is shown that adaptive predictors improve the performance of optimal Boolean predictors, and they outperform other recently proposed nonlinear predictors: the median adaptive predictor [21 and the gradient adjusted predictor [91. Finally, the gradient adaptive Boolean predictor is included in a lossless compression scheme, and the compression results are presented.
Olympia Lilly Bakalis, Los Alamos National Laboratory (U.S.A.)
Bryan J. Travis, Los Alamos National Laboratory (U.S.A.)
We investigate the use of Iterated Function Systems (IFS) for modeling and compressing 2 dimensional fractal images by exploring solutions of the Inverse IFS Problem: Given a fractal image we are looking for parameters in 24 dimensions for a small set of affine maps and their associated probabilities which constitute the IFS. Upon iteration the IFS solution produces an attractor with the characteristics which describe the image under consideration. We use a Genetic Algorithm (GA) and a Neural Network (NN) scheme which simulates the IFS. A sample cross section of the error hypersurface within the "Mandelbrot set" in the parameter space of a 3-map IFS family is shown. The solution obtained with the GA, in the 18 or 24 dimensional parameter space, meet the desired specifications which describe the original image to within the given discretization. Applications of the inverse problem are aimed towards prediction of second order phase transitions, where scale invariance and power laws are encountered, as well as towards image compression.
J. L. Paredes, University of Delaware (U.S.A.)
G. R. Arce, University of Delaware (U.S.A.)
N. C. Gallagher, University of Delaware (U.S.A.)
In this paper a nonlinear signal decomposition for image compression is presented. It uses a pyramid multiresolution scheme which is similar to that found using wavelet subband decompositions. The self-similarity across the different scales of the nonlinear signal decomposition is then exploited by the SPIHT algorithm which is modified to match the new signal decomposition. The nonlinear decomposition produces better edge preserving and less blocking artifacts in the reconstructed image than traditional wavelet decomposition, specially at very low bit rates.
Sofia Tsekeridou, Aristotle University of Thessaloniki (Greece)
Ioannis Pitas, Aristotle University of Thessaloniki (Greece)
The MPEG-2 compression algorithm, due to the use of variable length coding, is very sensitive to channel disturbances. A single bit error during transmission leads to noticeable degradation of the decoded sequence quality in that part or entire slice information is lost until the next resynchronization point is reached. Error concealment (EC) methods are employed at the decoder side to deal with this problem. A novel concealment scheme is proposed in this paper that uses spatial or, when available, temporal information to reconstruct the corrupted frames. The concealment strategy is embedded in the MPEG-2 decoder model to avoid further concealment of propagation errors. The proposed scheme is proved to be of better performance compared to that achieved by other error concealment (EC) methods. Furthermore, the quality of the concealed sequence seems to ameliorate with time.