Study And Application On Destriping Methods Of HSI Images

In the process of generation and transmission, remote sensing images are affected by noise and the quality of images is worse. Band noise is a very common phenomenon in images of many space-borne multi-sensor, airborne multi-sensor and single sensor spectrometer. Band noise seriously affects the interpretation of remote sensing data and the extraction of information. It can not play its due role. Many scholars have already made a lot of band noise reduction algorithms, but many of these algorithms are targeted to a certain particular band noise images. So the method has some deficiencies in the effectiveness of eliminating strips, simplicity, universal and self-adaptability of algorithm. This paper research on the algorithm of eliminating strips of hyper-spectral images. Some specific methods were researched. Improved algorithms of Moment Matching based on smoothing filter were put forward. They can recover approximately mean distribution caused by radiation strength and effectively eliminate the strips. This paper compared to image quality before and after destriping.The main contribution and the innovations of this paper are as follows:1) An improved algorithm of Moment Matching based on mean value filter was put forward. It can recover approximately mean distribution caused by radiation strength through combining Moment Matching and mean value filter.This algorithm was no requirement for feature type and gray distribution of images.2) An improved algorithm of Moment Matching based on polynomial fitting filter was put forward. The proposed algorithm uses the column average and variance which was processed with polynomial fitting filter instead of the average and standard deviation of the reference image in traditional moment matching algorithm.Data of dramatic and moderate changes all can get good smoothing effect.3) An improved algorithm of Moment Matching based on sliding window filter was put forward. The proposed algorithm uses the mean value of column average and variance in sliding window instead of the average and standard deviation of the center in the window. This method generally does not cause distortion and recovers radiation distribution well.Images destriped are very clear.4) The standard of image quality evaluation was established.Through the evaluation of the capacity of retaining characteristic information of the original image and de-noising effect, finding the best algorithm of the improved algorithms.The best method is recommended as the hyper-spectral images preprocessing methods. Considering the specials of HJ-1-A satellite HSI data, the best improved algorithms of Moment Matching was applied to eliminate the strips.5) In order to conveniently eliminate strips of a large number of hyper-spectral images, the paper designed the hyper-spectral image band noise reduction batch algorithm, and it was realized with the IDL.