Chinese researchers have introduced a new in-depth research-based computer visualization (CV) paradigm that examines a physical model and takes its data into account. This allowed them to achieve the most advanced single pixel imaging (SPI) results.
This research was carried out by SITU Guohai of the Shanghai Institute of Optics and Precision Mechanics (SIOM) of the Chinese Academy of Sciences (CAS). The results were published Dec. 13 in the journal Photonics Research.
KV techniques are mainly focused on the development of coding techniques and decoding algorithms. To design a traditional strategy encoding system (a direct physical model), an expert first uses his knowledge and then uses model-based optimization algorithms for decoding. A recent data-driven strategy developed an encoding mode and an inverse decoding model using training data. This significantly improves image quality and performance. However, its practical application is limited by difficulties in data collection, generalization and interpretation.
In this study, the team proposed a new paradigm for CI that makes full use of training data and a physical model. The physical model was reportedly used to eliminate artifacts due to the generalization problem of the train network model.
Their proposed method can provide the best image resolution among typical SPI algorithms including linear correlation and compression detection at a sampling rate of only 6.25%.
They also built a SPI-LiDAR (Laser Detection and Ranging) system and tested the proposed method in outdoor experiments. A tower located approximately 570 m away from the LiDAR system was successfully reconstructed using the proposed method, which shows that it has great potential for remote sensing.
This research was supported by the Sino-German Center Zentrum für Wissenschaftsforderung, the National Natural Science Foundation of China, and the CAS Advanced Scientific Research Program.
Source: https://bit.ly/3pc3YZb
