Multi-view image compression for Visual Sensor Network based on Block Compressive Sensing and multi-phase joint decoding

In this paper, a multi-view image compression framework for the Visual Sensor Network (VSN) is proposed that involve the use of Block-based Compressive Sensing (BCS) and multi-phase joint decoding. In the proposed framework, one of the sensor nodes (encoder) is configured to serve as the reference node, whereas the others as non-reference nodes. The images captured by the reference and non-reference nodes are referred as I_R and I_NR respectively. They are encoded independently using the BCS to produce two measurements that will be transmitted to the host workstation (decoder). In this case, I_NR is always encoded at a lower bitrate when compared to I_R, because the idea is to improve the reconstruction of I_NR with the help of IR. After the host workstation receives the two measurements, independent decoding is performed first, and then image registration is applied to project I_R onto the same plane of I_NR. The projected I_R is then fused with I_NR using wavelets. Subsequently, the difference between the measurement of the fused image and the measurement of I_NR is calculated. The difference is then decoded and added to If to produce the final improved version of I_NR. The simulation results show that the proposed framework is able to improve the quality of I_NR by 1dB to ~3dB at lower bitrates, when compared to the conventional BCS.