ARTEFACT DETECTION IN ASTROPHYSICAL IMAGE DATA USING INDEPENDENT COMPONENT ANALYSIS

ABSTRACTThis paper is the first reported application of ICA on astro-physical image data. When studying far-out galaxies froma series of consequent telescope images, there are severalsources for artefacts that influence all the images such ascameranoise, atmosphericfluctuationsanddisturbances,andstars in our own galaxy. For this problem, the linear ICAmodel holds very accurately, because the independence ofsuch artefacts is guaranteed. Using image data on the M31Galaxy, it is shown that several clear artefacts can be de-tected and recognized based on their temporal pixel lumi-nosity profiles and independent component images. Oncethese are removed, it is possible to concentrate on the realphysical events like gravitational lensing. ICA might pro-vide a very useful preprocessing for the large amounts ofavailable telescope image data.1. INTRODUCTIONIn modern astrophysics, one of the main research directionsis understanding the dark matter in the universe. Possi-ble candidates include compact objects such as small blackholes, dwarf stars, or planets. When such an object passesnear the line of sight of a star, the luminosity of the star willincrease – an effect called gravitational lensing, predictedby the general theory of relativity.In studying other galaxies than our own, individual starscannot be resolved, but a whole group of unresolved starsis registered in a single pixel element of a telescope CCDcamera. In a new technique called pixel lensing (see [1]),the pixel luminosity variations over time are monitored, andusingthese time series thelensing eventscanyet be detectedeven in the case of unresolved stars.A problem in the analysis of the images and luminosityvariations is the presence of artefacts. One of the possibleartefacts are the resolvedor individualstars between the far-out galaxy and the camera, which emerge sharply from the