Bands In Hyperspectral Image Noise Removal Method

Hyperspectral remote sensing is widely used in features detection, target identification, and other military or civilian fields. However, due to many factors, hyperspectral remote sensing images are usually seriously corrupted by stripe noises which bring huge inconvenience to the sequent image processing tasks. Therefore, we must remove stripe noises firstly.The basic objective of this paper is to effectively remove the stripe noises while preserving the original image details as much as possible. In this study, destriping methods are researched and the main contents are as follows:1. Analysised the two works of the imaging spectrometer, summarized the mechanism of production of stripe noises, researched the stripe degradation model and stripe-degenerated the analog image.2. Researched the existing typical methods such as histogram matching, moment matching, fourier transform method and wavelet transform method. Comparative analysed these methods’ scopes of application, denoising effects and shortcomings by simulation experiment.3. Be aimed at the problem that image after moment matching has "Banding effect ", this thesis proposed an improved algorithm of moment matching based on mean value filter. It takes the filtered mean sequence as the reference mean sequence, and can avoid "Band effect" while effectively removing the stripe noises. It has a wide range of adaptability.4. Be aimed at the objective that preserving the original image details as much as possible, this thesis proposed an new method with self-adaptative stripe noise detection based on different degree of pollution on image. It can effectively remove the stripe noises while preserving the original image details, and has a wide range of adaptability.5. Be aimed at the problem that image after moment matching has "Discrete-intensified effect ", this thesis proposed an improved algorithm of moment matching based on image divided. It divides the original image based on gray uniformity, and can avoid "Discrete-intensified effect" while effectively removing the stripe noises. It is only for particular image.The experiments of the analog stripe image and the real stripe image indicate these three new algorithms could both effectively eliminate stripe noises and preserve the original image details. All of them have a well performance in destriping hyperspectral remote sensing images.