FFT Selective and Adaptive Filtering for Removal of Systematic Noise in ETM+ Imageodesy Images

The images of imageodesy derived from Landsat-7 Enhanced Thematic Mapper Plus (ETM+) image pairs across an earthquake event exhibit severe horizontal and vertical striping noise patterns that obscure the desired information relating to earthquake-induced terrain displacements. The periodic noise patterns were analyzed to reveal that the principal cause of these errors was due to the two-way across-track scanning mechanism of the ETM+ instrument in relation to the minor shift of the satellite orbits between two acquisitions. We first focused on the design of selective filters pinpointing the noise frequencies of horizontal striping based on frequency-domain analysis, via fast Fourier transform (FFT). As a result, the horizontal striping patterns have been successfully eliminated, but meanwhile the systematic varying frequency nature of the more complicated wavy vertical striping patterns is better revealed. This has led to the successful design of an adaptive FFT frequency filtering mechanism based on the function characterizing the relationship between angular speed of a scanner imaging system and its corresponding line speed on the curved surface of the Earth. Finally, the horizontal and vertical striping patterns have been successfully eliminated by semiautomatic selective and adaptive filtering procedures without subduing the key information of coseismic displacement