Hyperspectral image compressed sensing model based on the collaborative sparsity of the intra-frame and inter-band

Abstract

This paper proposes a new hyperspectral image compressed sensing model (hyperspectral image collaborative sparsity measure, HICoSM) based on the collaborative sparsity of the intra-frame and inter-band, which utilizes the strong correlation of the inter-band. The sparsity mining of the proposed model includes three aspects:the local smoothness of the intra-frame; the nonlocal self-similarity of the inter-band that measures the detail information of the texture and the edge; and the correlation prediction of the inter-band, which specifically utilizes the prior band via the least squares linear prediction method to predict the current band, and then calculates the best predicting residual to realize the inter-band sparsity measurement. Further, we describe the numerical procedure of the proposed model. A large number of simulation experiments show that the HICoSM can make an effective sparsity measurement for all the intra-frames. Simultaneously, it can sufficiently mine and measure the sparsity of the inter-band of a hyperspectral image and effectively improve the decoding quality of each band in the CS recovery stage.