Interferometric Phase Noise Filtering Based on Adaptive Optimized Wavelet Packets Transform

Interferometric synthetic aperture radar (InSAR) has been used widely in investigation of surface deformation, such as earthquake, volcano, ground subsistence and landslides. But there are still some obstacles standing in the stage of the application. One of those is the phase noise in the interferogra. In particular, there are much noise in the inteferograms derived in steep topographic area with dense vegetation cover, such as the Three Gorges Area in China. This paper addresses the problem of phase noise filtering in InSAR. One difficulty of phase noise filtering of SAR interferometry is to reduce phase noise as much as possible while maintaining the phase information at the same time. The use of wavelet transform filtering is easy to ignore the high-frequency information, and make the image detail slur. One of the merits of wavelet packets transform is that most detail information in each frequency band will be analyzed. But over subtle decomposition in the high frequency bands, where noise is dominant, is apt to treat phase noise in those bands as signal. This would lead to erroneous result. In this paper, a phase noise filtering method based on adaptive optimized wavelet packets transform, or optimized tree-structured wavelet transform, is given out. By checking the correlation of wavelet coefficients in each wavelet scale, we decide whether each wavelet component in this scale should be decomposed further or not. According to such an adaptively constructed wavelet packets tree, complex phase image is decomposed. For the purpose of keeping the phase information, wavelet transform is executed in complex domain, and different threshold values are computed at each wavelet scale by using the intensity of their wavelet coefficients. Moreover, we have used an improved thresholding process method to try to overcome the disadvantages by both hard-thresholding and soft-thresholding methods.