| Title |
Bidirectional Point Spread Function Estimation based on Deep Learning for Satellite Imaging |
| Authors |
김한솔(Hansol Kim) ; 이해근(Haegeun Lee) ; 박종은(JongEun Park) ; 최병인(Byungin Choi) ; 이태영(Tae-Young Lee) ; 안종식(JongSik Ahn) ; 강문기(Moon Gi Kang) |
| DOI |
https://doi.org/10.5573/ieie.2022.59.9.84 |
| Keywords |
Satellite image; Point spread function estimation; Superresolution; Generative adversarial network; Regularization |
| Abstract |
Images acquired by satellite imaging system usually suffer from various types of degradation due to physical limitations. It is crucial that the point spread function is estimated accurately to restore the high resolution image from the low resolution image. In this paper, we propose a deep learning-based method that estimates a point spread function of satellite image bidirectionally. During the first stage of estimation, generative adversarial network which includes deep linear network is optimized for generating point spread function. At the second stage of estimation, the cost function is regularized by the gradient distribution of satellite image based on natural image prior. The optimal solution that minimizes cost function, which is the reconstructed high resolution image, performs as constraint of loss function to evaluate generated point spread function. Experimental results show that estimated point spread function of the proposed method is more accurate compared with the one of conventional methods. Also, we demonstrate that the high resolution satellite image could be reconstructed from the low resolution one using estimated point spread function. |