Multi-objects classification via optimized compressive sensing projection

The theory of compressive sensing (CS) enables the reconstruction of a sparse signal from highly compressed data. However, in many applications, we are ultimately interested in information retrieval rather than signal reconstruction. In this paper, we study the problem of multi-objects classification in compressive sensing systems. Theoretical error bounds are derived based on the analysis of classical compressive classification. The optimal projection matrix design problem is studied and an algorithm is derived to solve the corresponding problem. Application in the identification of license plate numbers is considered and simulation results show that the projection measurement obtained using the proposed algorithm significantly improve the classification performance in terms of classification error rate.