Joint trace/TV norm minimization: A new efficient approach for spectral compressive imaging

摘要: In this paper we propose a novel and efficient model for compressed sensing of hyperspectral images. A large-size image can be subsampled by retaining only 3% its original size, yet robustly recovered using the new approach present here. Our reconstruction is based on minimizing convex functional which penalizes both trace norm TV data matrix. Thus, solution tends to have simultaneous low-rank piecewise smooth structure: two important priors explaining underlying correlation structure such data. Through simulations will show our significantly enhances conventional compression rate-distortion tradeoffs. particular, in strong undersampling regimes method outperforms standard denoising recovery scheme more than 17dB MSE.