Optimized Measurements Coding for Compressive Sensing Reconstruction Network

Compressive Sensing (CS) is an emerging technology which can encode the original signal into several incoherent linear measurements and reconstruct the entire signal from a few measurements. Different from former coding schemes whose distortion mainly comes from the quantizer, the distortion is both related to quantization and measurement rate (MR) in CS based coding schemes. In this paper, we present an end-to-end image compression system based on CS. The presented system mainly integrates the conventional compressive sensing coding and the reconstruction with dead-zone quantization. We propose an optimized measurements coding scheme for our CS reconstruction network. We design the system parameters, including the choice of sensing matrix, the trade-off between quantization and MR, and the reconstruction network. Furthermore, the effective method can jointly control the quantization step and MR to achieve near optimal quality at any given bit rate. Therefore, our method can achieve a better balance between reconstruction quality and storage space.