Authors:
Matteo Pedone
1
;
Janne Heikkilä
1
;
Jarno Nikkanen
2
;
Leena Lepistö
2
and
Timo Kaikumaa
2
Affiliations:
1
University of Oulu, Finland
;
2
Nokia Research Center, Finland
Keyword(s):
Denoise, Demosaic, Evaluation, State-of-the-art, Perceptual quality assessment, Artifacts, Degradation, RAW images, Real data.
Related
Ontology
Subjects/Areas/Topics:
Computer Vision, Visualization and Computer Graphics
;
Enhancement and Restoration
;
Image Formation and Preprocessing
;
Image Quality
Abstract:
In this paper we present a performance evaluation of different state-of-the-art denoising method, applied to RAW images in Bayer pattern format. Several measures for assessing objective quality are considered. We also propose, a novel and straightforward extension to the SSIM-Index that handles color information. The evaluation is divided in two stages: first an entire set of images is artificially degraded and then restored with the considered denoising/demosaicking methods. The second stage involved a subjective evaluation with real noisy RAW images. We observed that the resulting qualities of the considered denoising methods are in agreement between the two different evaluation stages, and the best performing algorithms are easily identified. Moreover, the proposed extension of the SSIM-Index proved to behave more consistently in respect to the artifacts introduced by the denoising algorithms, and its outcome was always in fair accordance with the subjective perceived quality.