Vertical stripe correction in Hyperion image using wavelet transformation and singular value decomposition (SVD). Issue 10 (21st July 2022)

• Record Type: Journal Article
• Authors: Dixit, Ankur; Agarwal, Shefali
• Is Part Of: Geocarto international. Volume 37:Issue 10(2022)
• Page Start: 3033
• Page End: 3050
• Publication Date: 2022-07-21
• Subjects: Hyperion -- hyperspectral -- wavelet -- SVD -- destriping; Remote sensing -- Periodicals; Geographic information systems -- Periodicals; Geology -- Periodicals; Cartography -- Periodicals; 621.3678
• DOI: 10.1080/10106049.2020.1844312
• Languages: English
• ISSNs: 1010-6049

Abstract

Abstract: Vertical striping are quite common in Hyperion images. The removal of these stripes turns out to be an important step of noise correction along with atmospheric and other artifact correction. Often these corrections are applied on radiometric values in spatial domain or fequency domain. We have proposed a novel approach to correct these vertical stripes using wavelet transformation and singular value decomposition. In the first step, we proposed stripe identification based on the Eigen ratio of first and second Eigen values of vertical components (decomposed using wavelet transformation). This ratio is supposed to be very high for striped images and minimal for stripe free images. This is because of the contribution to first Eigen value of high frequency stripes. These high frequency stripes contributes tremendously to the first Eigen because of their dominance in vertical component. We have also proposed a threshold function to identify the bands that needs to be corrected. After the stripes identification and correction algorithm is appllied on respective bands, it is observed that proposed method successfully removed striping noise with high accuracy. Upon comparing the results of proposed method with various other methods, it is found that proposed method have an edge and performing better than rest of the methods. To assess the results, SNR value was calculated for all methods and found highest for the proposed method except Local Threshold approach. Local Threshold have better SNR probably because of intact noise (even in visual inspection but also present in frequency plot of vertical wavelet component) in these methods.