A real-time restoring method for infrared images degraded by high-speed airflow

Abstract

Restoring images degraded by low speed airflow such as atmospheric turbulence was studied by many researchers a long time ago, and many methods were proposed. However, those methods proposed for low speed airflow generally can't meet this urgency mainly in two aspects: first, those methods are usually time-consuming, which fail time requisition; second, those deconvolution models designed for low speed airflow may not suit the case of high speed airflow, which leads to bad restoring effect. Because existent blind deconvolution methods are not competent for real-time restoring infrared images degraded by high speed airflow, a fast restoring method for that application is researched in this paper. Both the PSF estimation and the object estimation processes are constructed to improve the algorithm's efficiency. The simplified Weiner filter is adopted to fast estimate objects given PSF. For traditional methods, many computation methods, including iterative and non-iterative ones, can be used to resolve that problem. But we are more interested in non-iterative methods because of application background of our work, in which the speed of restoring algorithms is also an important factor that we must be concerned about. Therefore we choose here inverse-filtering method to estimate the object and use FFT to accelerate the computation. Gaussian-like function is used to approximate the PSF of degradation of high speed airflow. There are many papers on blur identification and was summarized well. However, those methods are either iterative, leading to slow estimation, or not feasible for Gaussian-like PSF. Therefore We figure out a new a frequency domain scheme to estimate the parameter of PSF quickly, and use both simulation and wind tunnel experiment infrared images to test its validation. Finally we compare our algorithm with other three blind algorithms, that is the Rechardson-Lucy method (RL), the maximum likelihood method (ML) and the primary component analysis method (PCA), and the results show that our algorithm not only gives much better result, but also consumes much less time.