Tutorial 2 2018
Model Based Hyperspectral Image Denoising
Abstract : The received radiance at the hyperspectral sensor is degraded by sensor noises which include Johnson noise, quantization noise, and photon noise. These noises usually corrupt the spectral bands by varying degrees and degrade the efficiency of the hyperspectral image (HSI) analysis techniques. As a result, they are often discarded from the hyperspectral data before any further processing. Alternatively, hyperspectral denoising can be considered as a preprocessing step in HSI analysis to improve the signal to noise ratio and recover the corrupted bands. In this talk, I will first give an introduction about HSI denoising which includes HSI Modeling, HSI Denoising Criteria, HSI Noise Assumptions, and HSI Denoising Challenges. I will then give an overview on HSI denoising techniques categorized in four main groups including, 3D Model- Based and 3D Filtering Approaches, Spectral and Spatial-Spectral Penalty-Based Approaches, Low-Rank Modelbased Approaches, and Approaches Making the Mixed Noise Assumption. I will also show experimental results of a few HSI denoising method applied on simulated and real HSI datasets. Additionally, I will consider HSI denoising as a preprocessing step for HSI classification and will discuss the advantage of utilizing the denoising algorithms to improve the classification accuracies. Finally, I will give a summary on the evolution of HSI denoising and discuss the future challenges in HSI denoising. I will end my talk by introducing a Matlab toolbox on HSI denoising which has been recently provided online related to our recent review paper together with Paul Scheunders, Pedram Ghamisi, Giorgio Licciardi, and Jocelyn Chanussot entitled, “Noise Reduction in Hyperspectral Imagery: Overview and Application”.