Year: 2013
Type: conference-paper
Abstract: Hyperspectral images (HSI) are a collection of hundreds of images which have been acquired simultaneously in narrow and adjacent spectral bands.Aimed at meeting the needs of real-time process of hyperspectral data, the development of compressive techniques before the transmission and storage becomes critical.Recently, Compressed Sampling (CS), which exploits the sparsity of signals, has been allowed to reconstruct signals with fewer measurements than the traditional Nyquist sampling approach.In order to make use of the spectral correlation and spatial correlation simultaneously in the compressed sampling process, in this paper we developed a new three-dimensional compressed sampling (3DCS) method to reduce the sampling rate.In 3DCS, the three-dimensional circulant sampling model is presented, which samples the hyperspectral images with a random convolution process and a band-varying subsampling.In addition, an efficient reconstruct algorithm called three-dimensional total variation (3DTV) by exploiting its spatial and spectral correlation is used for this 3DCS with guaranteed convergence.The experiment results demonstrate that the superiority of our proposed 3DCS over 2DCS which only makes uses the spatial correlation is in terms of high recovery accuracy with respect to the sampling rate.And the reconstruct image using our proposed 3DCS is much better than the traditional 2DCS.
Authors: Li Wang, Yan Feng
Institution: Northwestern Polytechnical University
Language: English
Subfield: Computational Mechanics
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Award: G438222186
OpenAlex ID: W1973674670