Using Content Knowledge to Improve Reconstruction Performance by Semantic Compressive Sensing

Abstract

Compressive sensing (CS) provides a method to perform sampling and compression simultaneously when a signal is sparse or compressible. The conventional CS is established on the prior knowledge of sparsity and cannot reconstruct the signals if they fail to be represented explicitly in a sparse form. Semantic CS is a new approach for sensing a signal based on its content knowledge, which can reduce the sensing rate and the sparsity constraint. However, the effects of content knowledge on the reconstruction quality is still not clear. In this paper, we provide an approach for designing a sensing matrix for semantic CS and experimentally study how content knowledge affects the reconstruction performance of the proposed CS.