Compressed sensing recovery of multi-view video sequences by targeted database

Abstract

Compressed sensing (CS) is a technique that enables signal reconstruction at sub-Nyquist rate and has been widely used for fast reconstruction of multi-view video sequences (MVS) in the surveillance application. In this paper, we propose compressed-sensing recovery of MVS exploiting the model of structural group sparse representation (SGSR) along with a targeted database. SGSR groups similar patches together coupled with learning the adaptive basis from the similar groups, which gets sparser representation and thereby performs better in CS recovery. A targeted database can be easily obtained from the MVS due to their abundant prior information and the database can help us obtain more accurate similar patches, which further improve the performance with SGSR. Considering images as compressible signals rather than sparse signals, we design a filtering to retain the details of images. Simulation results show that the proposed algorithm outperforms existing reconstruction algorithms.