NEURAL-NETWORK ARCHITECTURES FOR VECTOR PREDICTION

A vector predictor is an integral part of a predictive vector quantiza tion (PVQ) coding scheme. However, the performance of a classical line ar vector predictor is limited by its ability to exploit only the line ar correlation between the blocks. Furthermore, its performance deteri orates as the vector dimension (block size) is increased, especially w hen predicting blocks that contain edge information. However, a nonlin ear predictor exploits the higher-order correlations among the neighbo ring blocks, and can predict edge blocks with increased accuracy. Beca use the conventional techniques for designing a nonlinear predictor ar e extremely complex and suboptimal due to the absence of a suitable mo del for the source data, it is necessary to investigate new procedures in order to design nonlinear vector predictors. In this paper, we hav e investigated several neural network architectures that can be used t o implement a nonlinear vector predictor, including the Multilayer Per ceptron, the Functional Link network, and the Radial Basis Function ne twork. We also evaluated and compared the performance of these neural network predictors with that of a linear vector predictor. Our experim ental results show that a neural network predictor can predict the blo cks containing edges with a higher accuracy than a linear predictor. H owever, the performance of a neural network predictor is comparable to that of a linear predictor for predicting the stationary and shade bl ocks.