Effective Stray Light Correction for EUV Images Paul Shearer with Richard A. Frazin, Anna C. Gilbert, and Alfred O. Hero III The Stray Light Problem • Stray light (for this talk): the long range scattering component of the PSF • Stray light makes bright features dimmer, dim features MUCH brighter, hindering quantitative science • coronal holes, prominence cavities, offlimb are all dominated by stray light in EUVI The Solution: Blind Deconvolution u1 u2 h Blind deconvolution (BDC) Given observations f, find trueemission images u and PSF h. • BDC appears underdetermined, but can be solved if u and h are constrained enough. f1 f2 • Challenge: find model that is solvable + contains true PSF/images Previous Work est. stray light % deconv obs 171 Å (under)estimating stray light by deconvolution • current EUVI PSFs: from SolarSoft’s euvi_psf.pro utility • suggests large corrections to faint regions • doesn’t remove all stray light (evidence – Moon transit) • Simple parametric PSF model cannot account for complex mirror scatter Our BDC Model Image model each image ui must be: • positive • zero on any known transit object • zero in vignetted corners (?) PSF model h must be: • positive • normalized to 1 • roughly monotonic inward gradient model BDC on Synthetic Data • Problem: such an unusually flexible model may overfit and/or be impossible to solve in practical problems. • Question: can it work on realistic images? Our experiments say yes! Synthetic Data Experiment • u = series of EUVI obs. with Moon and off-limb set to 0 • h = power-law decaying PSF, roughly approximates 171 Å mirror scattering • f = convolve u and h, add poisson noise Results recovered image error is at worst 5%, comparable to off-limb SNR; negligible on disk. PSF well recovered without parametric formula. BDC on EUVI-B 171 Å log10 f log10 u estimated stray light % log10 PSF BDC: Effects on Features • some faint features become up to 70% dimmer • up to double stray light removal of euvi_psf.pro o b s e r deconvolved v e d euvi_psf.pro deconv Conclusions and Ongoing Work • Our PSF removes almost all stray light from EUVI-B 171 time series – except far off limb – must fix boundary conditions • Next steps: – refine PSF wings by improving boundary conditions (how?); work on full images – estimate cores for high res work (e.g. loops) – estimate uncertainty with cross-validation and independent modeling approaches – repeat for EUVI A/B, all 4 bands – apply to AIA, which has much better transits – all PSFs will be publically available for correction use Questions • Are any model assumptions incorrect? • Are there any additional constraints we could incorporate on images or PSF? is there any help from experimental data? Applications • differential emission measure studies • coronal loop widths, background subtraction • ???