Graph-Based Transform Based on Neural Networks for Intra-Prediction of Imaging Data

Abstract

This paper introduces a novel class of Graph-based Transform based on neural networks (GBT-NN) within the context of block-based predictive transform coding of imaging data. To reduce the signalling overhead required to reconstruct the data after transformation, the proposed GBT-NN predicts the graph information needed to compute the inverse transform via a neural network. Evaluation results on several video frames and medical images, in terms of the percentage of energy preserved by a sub-set of transform coefficients and the mean squared error of the reconstructed data, show that the GBT-NN can outperform the DCT and DST, which are widely used in modern video codecs.