Compression of Hyperspectral Images Using Superpixel Based and Error-Corrected Sparse Representation


Ertem A., KARACA A. C. , GÜLLÜ M. K.

27th Signal Processing and Communications Applications Conference (SIU), Sivas, Türkiye, 24 - 26 Nisan 2019 identifier identifier

Özet

Performance of the superpixel based SSASR method used in hyperspectral image compression is noticeably higher than the methods in the literature. However, the SSASR method gives low signal-to-noise ratio values if there are few pixels that differ spectrally in the image. It is aimed to overcome this problem with the proposed method. In the proposed method, first of all, the pixels that are significantly distorted in the reconstructed data are determined. After that, a new dictionary and sparse coefficients are defined for these pixels. In the experiments, the proposed method is compared with SSASR and the other state-of-the-art methods in terms of different quality metrics. It is represented that the the proposed M-SSASR method provides superior performance compared to the other methods for all these metrics.