High-speed architecture for image reconstruction based on compressive sensing

Compressive sensing (CS) is a superior signal sampling strategy that combines sampling and compression. CS-based imaging systems include sampling and reconstruction stages. Currently, the complex task of image reconstruction has only been implemented in software, which can only achieve very limited speed. This paper proposes a high-speed hardware architecture for the reconstruction of compressively-sensed images. The reconstruction algorithm based on the split Bregman method, which solves the ℓ₁ minimization problem, is first simplified to reduce hardware complexity. Then an efficient partial parallel hardware architecture is developed to implement the modified algorithm. With moderate silicon area, the proposed architecture can reconstruct a 128 × 128 image in 3.82×10^-2 seconds, which is over 100 times faster than software implementations.