NEURAL-NETWORK APPROACHES TO IMAGE COMPRESSION

This paper presents a tutorial overview of neural networks as signal p rocessing tools for image compression. They are well suited to the pro blem of image compression due to their massively parallel and distribu ted architecture. Their characteristics are analogous to some of the f eatures of our own visual system, which allow us to process visual inf ormation with much ease. For example, multilayer perceptions can be us ed as nonlinear predictors in differential pulse-code modulation (DPCM ). Such predictors have been shown to increase the predictive gain rel ative to a linear predictor. Another active area of research is in the application of Hebbian learning to the extraction of principal compon ents, which are the basis vectors for the optimal linear Karhunen-Loev e transform (KLT). These learning algorithms are iterative, have some computational advantages over standard eigendecomposition techniques, and can be made to adapt to changes in the input signal. Yet another m odel, the self-organizing feature map (SOFM), has been used with a gre at deal of success in the design of codebooks for vector quantization (VQ). The resulting codebooks are less sensitive to initial conditions than the standard LBG algorithm, and the topological ordering of the entries can be exploited to further increasing coding efficiency and r educe computational complexity.