Implementation of Irregular Meshes for the Sparse Representation of Multidimensional Signals

Abstract

The paper is dedicated to development of effective tools of multidimensional digital signal processing on irregular meshes. ANN-based method of irregular mesh generation for intra-frame video coding is developed. The method described is based on artificial neural network implementation. Different architectures and types of artificial neural networks are compared. The training and testing sequences generation problem is discussed. The aim of the irregular mesh coverage of the two-dimensional signal (frame) is to decrease computational cost for the further motion detection between frames. The benefit of the artificial neural network usage is the relatively low computational cost of the mesh generation in comparison with analogous. The implementation of the irregular meshes for the correlation analysis between signals is discussed. Examples of the utilization of the irregular mesh-based FIR filtering for the open-boundary problem numerical solution are presented. Generalized results obtained may be used for pattern recognition, data compression, multidimensional look-up table interpolation.