Content-Based Compressive Sensing

Compressive sensing (CS)provides an alternative to Shannon/Nyquist sampling theorem by performing signal acquisition and compression simultaneously when the signals are sparse or compressible in certain basis. However, the classical random CS strategy heavily relies on the sparsity of signals under acquisition and fails to reconstruct the signals when they are not sparse. In this paper, we propose an approach to sensing signals based on their content information to reduce the sensing rate and get rid of the sparsity requirement. Experimental results demonstrate the effectiveness of the proposed method for both sparse and non-sparse images.