Cloud removal for single images based on dual tree complex wavelet transform

Abstract

Clouds can obscure ground information during remote sensing imaging. Cloud removal technology for a single image becomes significant when no images containing cloud-free regions are available. After the fundamental principle of dual tree complex wavelet transform (DTCWT) was reviewed, and the frequency relationships between clouds and ground objects in remote sensing images were analyzed, a novel algorithm to remove clouds from a single remote sensing image was proposed. The algorithm divided the cloud-contaminated image into low level high frequency sub-bands, high level high frequency sub-bands and low frequency sub-band by DTCWT. The low level high frequency sub-bands were filtered to enhance the ground object information by Laplacian filtering. The other two types of sub-bands were processed to remove cloud by applying the method of cloud layer coefficient weighting (CLCW). Image processing experiments were implemented. Their results were analyzed. It proved the Laplacian contributes to enhancing ground object information adaptively. CLCW has the ability to remove clouds while preserving the ground object information outside the cloud cover. The proposed algorithm is greatly superior to algorithms based on traditional wavelet transform and the wavelet threshold theory.