4DCAF: A Temporal Approach for Denoising Hyperspectral Image Sequences

Abstract

Due to the rapid advancement of sensor technology over the past decade, it is now feasible to capture sequences of hyperspectral images at acceptable frame rates. Nonetheless, these sequences can be heavily affected by noise, particularly when the data spans the thermal spectrum. Although there is extensive research on denoising standard video sequences and still hyperspectral images, there is limited work on denoising hyperspectral sequences. This paper introduces a new denoising technique specifically designed for actual hyperspectral sequences. The method utilizes spatio-spectral-temporal cellular automata-based filtering, offering several advantages. Notably, the cellular automaton employed can incorporate information about the type of noise present by using specific sequences to fine-tune the algorithm. Additionally, it accounts for temporal information through a spatio-temporal neighborhood when processing each pixel in the sequence. The proposed method surpasses several leading algorithms on both simulated and real sequences.