Ghosting reduction in adaptive non-uniformity correction of infrared focal-plane array image sequences

ESTEBAN MAURICIO VERA ROJAS, I. Sergio Torres

Research output: Contribution to conferencePaperpeer-review

12 Scopus citations

Abstract

Non-uniformity correction is a critical task for achieving higher performances in modern infrared imaging systems. Lately, special interest has been given to a scene-based adaptive non-uniformity correction approach based on a neural network with a steepest descent learning rule. However, low motion and some scene artifacts such as edges usually cause the production of ghosting-like artifacts over the output images. We assume that such ghosting is mainly produced by the use of fixed learning rates. In this work we propose the addition of two adaptive learning rate strategies to minimize the presence of ghosting artifacts. The first proposal relies in the fact that a momentum term can accelerate and stabilize a learning process, which in this case could lead to a possible ghosting reduction as well. As an alternative to the momentum, we also propose the use of a variable learning rate related to the local variance of the input image. The both proposed improvements are tested in infrared image sequences with simulated non-uniformity. Results demonstrate that the proposed approaches are helpful in increasing the quality of the corrected image sequences. Nonetheless, only the variable learning rate seems to reduce in a better way the unwanted ghosting artifacts.

Original languageEnglish
Pages1001-1004
Number of pages4
StatePublished - 17 Dec 2003
EventProceedings: 2003 International Conference on Image Processing, ICIP-2003 - Barcelona, Spain
Duration: 14 Sep 200317 Sep 2003

Conference

ConferenceProceedings: 2003 International Conference on Image Processing, ICIP-2003
CountrySpain
CityBarcelona
Period14/09/0317/09/03

Fingerprint Dive into the research topics of 'Ghosting reduction in adaptive non-uniformity correction of infrared focal-plane array image sequences'. Together they form a unique fingerprint.

Cite this