科研之友
微信
新浪微博
Facebook
分享链接
摘要
Modern single lens reflex lenses typically consist of up to two dozen individual optical elements, the complexity of which is necessary to compensate for geometric and chromatic aberrations. This paper adopts one single lens to capture images instead of complex lenses, and computational photography technique is employed to remove corresponding imaging artifacts. We initially estimate the space-variant point spread function of the single lens by combining l(1)/l(2) image and sparse kernel priors. A fast non-blind deconvolution method with hyper-Laplacian prior is then performed to recover clear image. Experimental results show that the proposed method is at par with state-of-the-art non-blind deconvolution approaches, especially with regards its speed that is much faster than that of existing methods.
- 出版日期2015
- 单位中国人民解放军国防科学技术大学; 中国空气动力研究与发展中心
