RestoreNet-Plus: Image restoration via deep learning in optical synthetic aperture imaging system. (November 2021)

• Record Type: Journal Article
• Authors: Tang, Ju; Wu, Ji; Wang, Kaiqiang; Ren, Zhenbo; Wu, Xiaoyan; Hu, Liusen; Di, Jianglei; Liu, Guodong; Zhao, Jianlin
• Is Part Of: Optics and lasers in engineering. Volume 146(2021)
• Publication Date: 2021-11
• Subjects: (110.4850) Optical transfer functions -- (110.3010) Image reconstruction techniques -- (200.4260) Neural networks; Lasers in engineering -- Periodicals; Optical measurements -- Periodicals; Optics -- Periodicals; Lasers en ingénierie -- Périodiques; Mesures optiques -- Périodiques; Optique -- Périodiques; 621.36605
• DOI: 10.1016/j.optlaseng.2021.106707
• Languages: English
• ISSNs: 0143-8166

Abstract

Highlights: The blur is a problem impacting the image quality of optical synthetic aperture imaging (OSAI) system and requiring image restoration techniques for deconvolution. However, most traditional image deconvolution algorithms are non-blind methods and also highly susceptible to the unknown noise. The turbulence perturbation and the co-phase error are both factors that may affect the recovery performance. As for the image restoration under noises and turbulence correction error, a further optimized network named as RestoreNet-Plus is designed. After further optimization, RestoreNet-Plus has advantages as below: Blind restoration ability and generalization capacity: RestoreNet-Plus could remove the blur better in the blind way facing different targets through OSAI system; Anti-noise ability: It could beat other methods under different levels noises, including Wiener filter, Hyper-Laplacian prior algorithm, Lucy-Richardson algorithm, Blind Deconvolution algorithm and RestoreNet. Anti-turbulence correction error ability: RestoreNet-Plus could maintain good Blind restoration ability under the influence of turbulence correction error in imaging. Fast for applying: RestoreNet-Plus performances the best in the image restoration of OSAI system under noises and turbulence correction error with only 54 ms time costs. Abstract: The synthetic aperture technology can improve the resolution effectively in the optical imaging system. In fact, the imaging blur, turbulence aberration and noise can affect the imaging quality of optical synthetic aperture imaging system seriously. Several non-blind methods are applied generally to recover the degraded maps with the prior information. However, the restoration effect is not stable enough and satisfactory. As a data-driven approach, the deep learning framework possesses advantages in solving this problem. In this paper we propose an improved network, RestoreNet-Plus, for the image restoration of optical synthetic aperture imaging system. After the proofs of numerical simulation and experiment results, RestoreNet-Plus is a better alternative compared with other methods, owing to its better restoration ability, strong denoising ability and capacity for turbulence correction error.