High-Dimensional Data Reduction, Image Inpainting and their Astronomical Applications

Abstract

Technological advances are revolutionizing multispectral astrophysics as well as the detection and study of transient sources. This new era of multitemporal and multispectral data sets demands new ways of data representation, processing and management thus making data dimension reduction instrumental in efficient data organization, retrieval, analysis and information visualization. Other astrophysical applications of data dimension reduction which require new paradigms of data analysis include knowledge discovery, cluster analysis, feature extraction and object classification, de-correlating data elements, discovering meaningful patterns and finding essential representation of correlated variables that form a manifold (e.g. the manifold of galaxies), tagging astronomical images, multiscale analysis synchronized across all available wavelengths, denoising, etc. The second part of this paper is dedicated to a new, active area of image processing: image inpainting that consists of automated methods for filling in missing or damaged regions in images. Inpainting has multiple astronomical applications including restoring images corrupted by instrument artifacts, removing undesirable objects like bright stars and their halos, sky estimating, and pre-processing for the Fourier or wavelet transforms. Applications of high-dimensional data reduction and mitigation of instrument artifacts are demonstrated on images taken by the Spitzer Space Telescope.