Implementation of Imaging Compressive Sensing Algorithms on Mobile Handset Devices

Abstract

Compressive Sensing (CS) is a remarkable framework that efficiently senses a signal taking a set of random projections from the underlying signal. Using the random projections, a CS reconstruction algorithm is then used to reconstruct the initial signal. Extensive efforts have been made in CS to determine the minimum number of required random projections and to design efficient optimization algorithms for correct signal reconstruction. In practice, the huge number of operations required for these reconstruction algorithms have restricted CS techniques to be implemented on high performance computational architectures, such as personal computers, servers, and Graphical Processing Units (GPU). This work determines the computational requirements to implement CS techniques on a limited memory mobile device. The results show the computational time and the energy consumption of two CS image reconstruction algorithms on a mobile device as a function of the size and sparsity of the underlying image. Results in the quality of the images recovered in smartphones show a Peak Signal to Noise Ratio of about 39 dB. Regarding the energy consumption, both greedy algorithms dissipated the same energy during the compression/reconstruction process.