An Operational Superresolution Approach for Multi-Temporal and Multi-Angle Remotely Sensed Imagery

In this paper we propose an operational superresolution (SR) approach for multi-temporal and multi-angle remote sensing imagery. The method consists of two stages: registration and reconstruction. In the registration stage a hybrid patch-based registration scheme that can account for local geometric distortion and photometric disparity is proposed. Obstacles like clouds or cloud shadows are detected as part of the registration process. For the reconstruction stage a SR reconstruction model composed of the L1 norm data fidelity and total variation (TV) regularization is defined, with its reconstruction object function being efficiently solved by the steepest descent method. Other SR methods can be easily incorporated in the proposed framework as well. The proposed algorithms are tested with multi-temporal and multi-angle WorldView-2 imagery. Experimental results demonstrate the effectiveness of the proposed approach.