Information-quantitative evaluation of linear computational imaging and application in ghost imaging

Current evaluation of imaging systems mainly relies on comparing the imaging results to the ground truth. However, the ground truth is usually unavailable in practical scenarios. In this paper, we propose a framework for assessing the capabilities of linear computational imaging processes, where the mapping between object and measurements can be modeled as linear. This framework utilizes Bayesian approach to estimate the amount of acquired information from each sampling, independent of the imaging results and ground truth. We demonstrated our framework based on ghost imaging, a typical linear computational imaging system. This-method facilitates image quality to be improved to the Cramér–Rao bound. Furthermore, an adaptive design of the imaging procedure is also developed and integrated into this dynamic evaluation framework to improve the ability to capture information. The proposed framework exhibits intrinsic generalizability to linear computational imaging systems, and its theoretical structure allows for potential extensions to other computational imaging paradigms.