Restauration d'images de la rétine corrigées par optique adaptative

Abstract : High resolution retinal imaging is hampered by eye's aberrations. The measurement and correction of these aberrations is made possible with an adaptive optics (AO) system. A retinal imaging bench is developped by Observatoire de Paris and is currently used in Hôpital des XV-XX in Paris. In wide field imaging, the object under examination (the retina) is three-dimensional and each recorded
image contains information on the object's volume. Furthermore, the AO correction is always partial and a dedicated deconvolution method is necessary to increase the images resolution and to separate numericaly the different object plans. A deconvolution method need an accurate knowledge of the point spread function (PSF) of the instrument and the tuning of several parameters named hyper parameters. In this framework, we developped two 3D deconvolution methods. The first one use the PSF, supposed
well known. The second method, which is a 3D phase diversity extension, estimates the aberrations jointly with the object. Furthermore, we developped an unsuppervised hyper parameters estimation method which is compatible with an effectiveness use of our 3D deconvolution method by non expert people.
The performance of all these methods are shown on simulated retinal images and experimental data. The experimental data become from a 3D images optical bench developped at ONERA during this phD thesis.
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00412618
Contributor : Guillaume Chenegros <>
Submitted on : Wednesday, September 2, 2009 - 12:04:40 PM
Last modification on : Tuesday, March 26, 2019 - 2:24:02 PM
Long-term archiving on: Tuesday, October 16, 2012 - 10:11:02 AM

Identifiers

  • HAL Id : tel-00412618, version 1

Collections

Citation

Guillaume Chenegros. Restauration d'images de la rétine corrigées par optique adaptative. Astrophysique [astro-ph]. Université Paris-Diderot - Paris VII, 2008. Français. ⟨tel-00412618⟩

Share

Metrics

Record views

311

Files downloads

1739