OPTIMIZED COMPRESSION METHOD FOR MULTISPECTRAL DIGITAL IMAGES OF PROJECTION NATURE

Abstract

A method for compressing digital images of a projection nature obtained in an arbitrary amount of spectral radiation ranges - a carrier of specific information or under various positional conditions of formation, optimized according to the criteria for preserving the specified levels of signal image energy and their structural similarity, is proposed.

The method uses the representation of a set of such images in the form of a single multidimensional geometric object, which is described by a two-dimensional data array ordered by raster and spectral intervals and includes preprocessing by means of singular decomposition of the specified array. Actual compression is carried out by means of decomposition the constituent images, represented by maximum singular numbers, according to discrete orthonormed functional bases, zeroing out part of the decomposition coefficients and subsequent reconstruction of the digital brightness codes of images. The determination of the thresholds for zeroing the decomposition coefficients of images digital levels is formulated in the form of a two-criteria optimization problem of obtaining the given ratios of signal energies of the compressed and initial images and indices of their structural similarity.

The implementation of the proposed methods includes the following steps: pairwise orthogonalization of distributions of digital brightness codes of images in different spectral channels based on singular decompositions; preservation of the constituent distributions of brightness with maximum singular numbers; compression of orthogonalized representations according to the specified criteria; reconstructing the brightness codes of compressed images of spectral channels by a functional transformation reversed by the one used at the stage of decomposition of brightness distributions of directly recorded multispectral images.

Comparison of various discretized functional bases as the basis for compression showed the greatest efficiency in the specified criteria of the discrete functional basis of Haar.

Key words: multispectral image, orthogonalization, signal energy, structural similarity index, orthonormed basis.