Enhanced Compressive Imaging Approach Based on Multi-measurement and Dynamic Clustering

In noisy environments, signal reconstruction can be converted into the issue of bound constrained quadratic programming which can be resolved by the regularization programming algorithm, but the reconstruction quality depends heavily on the regularization parameter. Without any apriori knowledge of noise, the Generalized Cross-Validation (GCV) algorithm provides a suitable way for estimation. But in low Signal-to-Noise Ratio (SNR) conditions, it is difficult for GCV to guarantee perfect convergence at the global optimum, which results in the Signal-to-Clutter Ratio (SCR) of the reconstructed image declining and targets missing. For robust reconstruction in low SNR conditions, the enhanced compressive imaging approach based on Multi-Measurement and Dynamic Clustering (MMDC) is proposed in this paper. First, it extracts randomly the original measured data by multiple times. Second, it receives the image series by CS processing. Finally, it implements robust reconstruction by clustering the image series with DC algorithm. Both the simulated and experimental results indicate that MMDC not only improves the reconstruction quality, but also receives effective clutter suppression. Due to the heavy computation of GCV and the insensitivity of MMDC to estimation error, the MMDC based on a simplified GCV algorithm is also proposed in this paper.