|
This work introduces a novel fast hyperspectral image deblurring and denoising approach tailored to archaeological applications of remote sensing. Hyperspectral data recorded by means of airborne or satelliteborne sensors can be used to detect buried archaeological deposits as the latter have a localised impact on the physical and chemical properties of the soil and the vegetation located above them, contributing to make them structurally different from the surrounding elements. By processing and analysing hyperspectral images, archaeological photo-interpreters can detect subtle changes in the properties of ground elements that can be attributed to the presence of subterranean archaeological sites. Hyperspectral imagery, while rich in content as far as the spectral characteristics of ground elements, often lacks in spatial resolution and contains blurring degradation and noise, prominent especially in some spectral regions. The influence of blur and noise highly effects not only the quality of the visual appearance of the represented objects and compromises the interpretation process, but impacts also further processing of imagery, limiting consequently the detection of targets of interest. The methodology here presented is based on the low-rank properties of hyperspectral images and fully exploits a sparse hyperspectral data representation linked with the self-similarity characteristics of image patches (small image parts). The restoration procedure additionally includes a bend-dependent formulation of blurring degradation. The preliminary results show high performances and reduced computational complexity, and that the proposed approach is able to cope with Gaussian and Poisson noise and band-dependent blur. By removing severe noise and blur, the accurate detection and interpretation of buried structures in different shapes and sizes is thus improved. The proposed approach significantly increases the number of hyperspectral bands that can be used for further image processing and analysis, providing new avenues for features of interest discoveries in bands where they normally obscured by noise.
|
