Production of high-resolution reference polarization images from real world scenes - Archive ouverte HAL
Communication Dans Un Congrès Année : 2023

Production of high-resolution reference polarization images from real world scenes

Résumé

A few polarization image datasets depicting real world scenes have been reported. Some of them are available on an open-data basis. Some databases contain color images, often with color bands reconstructed from a sensor equipped with a Bayer filter. Unfortunately, even if these real-world images depict a variety of objects and situations and have a good overall quality (ie the spectral bands and the various polarization channels are or can be registered, noise is reduced), they often have a low definition (smaller than 1 Mp), a low bit depth and are captured with a large lens aperture, resulting in very band-limited images. Moreover, the demosaicing procedure used to reconstruct the various color bands has a smoothing effect, reducing their resolution. This latter point proves detrimental when it comes to use these images as references for demosaicing algorithms, especially for RGB images: since each channel combining polarization direction and spectral band is very sparse in the base mosaic pattern, artifacts likely to appear are considerably underestimated with band-limited images. In this work, we review existing polarization image databases, focus on non-mosaiced datasets and propose a technique to produce HD polarization images with superior quality.
Fichier principal
Vignette du fichier
126900B.pdf (23.21 Mo) Télécharger le fichier
Origine Fichiers éditeurs autorisés sur une archive ouverte

Dates et versions

hal-04231694 , version 1 (06-10-2023)

Identifiants

Citer

Bigué Laurent, Alban Foulonneau, Pierre-Jean Lapray. Production of high-resolution reference polarization images from real world scenes. Polarization Science and Remote Sensing XI, SPIE, Aug 2023, San Diego, United States. pp.126900B, ⟨10.1117/12.2677421⟩. ⟨hal-04231694⟩
23 Consultations
15 Téléchargements

Altmetric

Partager

More