Abstract:

Since the amount of data in satellite cloud images is huge, acquisition, transmission, and storage of satellite cloud images are costly. To reduce the amount of data, compressed sensing (CS) reconstruction for satellite cloud images is presented based on residual estimation. An aliasing-free directional multi-scale transform (AFDMT) is proposed and applied to the sparse representation of CS. With a random projection technique, the projected Landweber algorithm with a smoothing operation is introduced into reconstruction of the block-based CS recovery. High quality cloud images can be obtained by estimating the residual between the current cloud image and the previous cloud image. Experimental results show that AFDMT is applicable to the sparse representation of CS. The residual estimation scheme can improve quality of the reconstructed cloud images in terms of visual observation and objective evaluation. For example, at a measurement rate of 0.5, peak signal to noise ratio of the reconstructed image is increased by more than 0.5 dB in average compared with direct reconstruction methods.

Key words:  cloud image reconstruction, compressed sensing, residual estimation, directional multi-scale transform