Automatic local phenology simulation for landsat TM image

• Published in: 2012 IEEE International Conference on Information Science and Technology pp. 898 - 901
• Main Authors: Zhengke Gui, Jianbo Liu, Fu Chen
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.03.2012
• Subjects: Clouds; Correlation; Image color analysis; Predictive models; Reflection; Remote sensing; Sparse matrices
• ISBN: 9781457703430; 1457703432
• ISSN: 2164-4357
• DOI: 10.1109/ICIST.2012.6221778

Partial cloud removal from remote sensing images composed of three sequential steps: accurate cloud and cloud shadow detection of the remote sensing image and corresponding cloud mask generation, phenology simulation for adjacent temporal images, fusion for blending artificial effects of composite image. Phenology simulation predicts what the surface features would look like in fields beneath clouds. With the assumption that surface features have subtle changes between images acquiring interval, some articles proposed phenology simulation methods to solve this problem based on color transfer in lαβ color space. In this paper, we proposed an optimizing algorithm using multiband information of remote sensing image. Our method is on basis of a simple premise: the same kind of surface feature has consistent reflection in all bands, especially in local areas. We formalize the premise using Gaussian probability-density function, on basis of which, a large sparse symmetric positive definite matrix is built to calculate the pixels' value underneath clouds with the information of neighborhood surface features. Some comparison experiments have been presented to demonstrate the effectiveness of our complete and sophisticated approach.