Destriping remotely sensed data using anisotropic diffusion in wavelet domain

Abstract

Due to differences of each detector's response to the same radiant signal, stripe noise exists in various remotely sensed data, which degrades accuracy of the imagery and brings difficulty in subsequent analysis such as quantitative inversion, terrain classification and object detection. Wavelet shrinkage denoising algorithm can remove the stripe effectively because it suppresses the noise through hard or soft thresholding function. However, the main defect lies in blurred edges of ground objects since it always over reduces the wavelet coefficients. In order to preserve and restore the edges to the full extent, this paper adopted anisotropic diffusion method into wavelet domain. An anisotropic diffusion filtering process was applied to the high frequency wavelet coefficients which eradicated the stripe while preserving the ground object edges. Experimental results showed that the proposed method in this paper outperformed the traditional wavelet shrinkage denoising method both in visual effects and several image quality indexes.