Context Clustering in Lossless Compression of Gray-Scale Image.
ABSTRACT We consider and evaluate the context clustering method for lossless image compression based on the existing LOCO-I algorithm
used in JPEG-LS — the latest lossless image compression standard. We employ the LOCO-I Medpredictor to enroll the error pixels.
The contexts are defined by calculating gradient of current pixels. The three directional gradients are quantized with different
codebook size (7, 9, 19) respectively. The error pixels are then corrected and encoded by the clustered-contexts. A main advantage
of using the context clustering method is that it can eliminate the storage of probability vector. An adaptive arithmetic
encoder is also introduced to yield a higher compression rate.
Conference Paper: Bootstrapping sequential monte carlo tracking[Show abstract] [Hide abstract]
ABSTRACT: Sequential Monte Carlo (SMC) methods have in recent years been applied to handle some of the problems inherent to model-based tracking. In this paper we suggest to apply bootstrapping to reduce the required number of particles in SMC tracking. By bootstrapping is meant to track reliable low-level image features and use them to bootstrap the high-level model-based tracking. The concept of bootstrapped SMC tracking is exemplified by monocular tracking of the 3D pose of a human arm with the position of the hand in the image as the bootstrapping information. Tests suggest that both bootstrapping is a sound strategies and an improvement over standard SMC-methods.Proceedings of the 13th Scandinavian conference on Image analysis; 06/2003