Variational Attenuation Correction of Two-View Confocal Microscopic Recordings

Abstract: We present an approach to recover attenuation-free intensities of a thick sample, that is imaged by a standard confocal microscope from two views (top and bottom). A variational approach simultaneously estimates the local signal attenuation and the real attenuation-free intensity at each position. Compared to earlier work we introduce a refined image formation model, that models photo-bleaching and photon noise using Poisson image statistics. We examine the effects of different regularization methods on the absorption field (Tikhonov-Miller, Total Variation, and sparsity) and the benefit of a constrained optimization in comparison to an orthogonal subspace projection. We quantify the efficacy of the approach on synthetically generated samples, and show its general applicability on two real biological applications, namely the recordings of zebrafish embryos and Arabidopsis thaliana root tips.