Compensation for optical remote sensing image compression based on distortion sensitivity

High-resolution optical remote sensing images are prone to serious local distortion after high compression, whose targets and textures are abundant and complex. Most of the current researches do not focus on subjective image quality, and this will easily lead to over-compensation. In order to reduce local distortion, the correlations between SSIM (Structural similarity) component functions and MOS (Mean opinion score) were analyzed on an optical remote sensing compression distortion image database, and a distortion sensitivity model for remote sensing image compression was proposed. Then, this model was utilized to design a compensation approach, and applied to an embedded wavelet image coder. This approach could locate the distortion sensitivity areas and compress the distortion values to reserved space at encoder, and compensate these values into reconstructed image at decoder. Experiment results show that this approach can enhance the visibility and identification of remote sensing objects in the distorted sensitive areas, reduce serious local distortions, and improve the overall image quality.