4DCAF: A temporal approach for denoising hyperspectral image sequences. (December 2017)

• Record Type: Journal Article
• Authors: Priego, Blanca; Duro, Richard J.; Chanussot, Jocelyn
• Is Part Of: Pattern recognition. Volume 72(2017:Dec.)
• Page Start: 433
• Page End: 445
• Publication Date: 2017-12
• Subjects: Hyperspectral -- Temporal denoising -- Cellular automata -- 4DCAF; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.07.023
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: Adaptation of cellular automata (CA) based structures for the denoising of hyperspectral images sequences. The adaptation of the CA to properly perform the denoising considering a particular type of noise model is carried out by an evolutionary technique. Versatility for adjusting the rule set of the CA filtering structure in an automatic manner. Potential of the method for efficient parallel implementations. Performances are very good when compared to alternative state of the art filtering strategies. Abstract: As a consequence of the fast development of sensor technology in the last decade, it is now possible to acquire sequences of hyperspectral images at reasonable frame rates. However, these sequences may be significantly corrupted by noise, especially when the spectral coverage of the data reaches the thermal domain. While there is an abundant literature on denoising of (standard) video sequences or denoising of (still) hyperspectral images, very little has been published on denoising hyperspectral sequences. This paper presents a novel denoising method for actual hyperspectral sequences. The approach is based on spatio-spectral-temporal cellular automata-based filtering. It presents several advantages, especially the fact that the cellular automaton used is able to contemplate information concerning the type of noise present through the use of specific sequences to tune the algorithm. It also considers temporal information by means of a spatio-temporal neighborhood when processing each pixel of the sequence. The proposed method outperforms several state-of-the-art algorithms on both simulated and real sequences.