Bokeh Rendering from Defocus Estimation

Xianrui Luo, Juewen Peng, Ke Xian, Zijin Wu, Zhiguo Cao

Abstract: In this paper, we study realistic bokeh rendering from a single all-in-focus image. Existing computational bokeh rendering methods generate bokeh effects by adding a simple flat background blur. As a result, the rendering results are different from the real bokeh on DSLR cameras. To address this issue, we propose a multi-stage network to learn shallow depth-of-field from a single bokeh-free image. In particular, our network consists of four modules: defocus estimation, radiance, rendering, and upsampling. The four modules are trained on different sizes to learn global features as well as local details around the boundaries of in-focus objects. Experimental results show that our approach is capable of rendering a pleasing distinctive bokeh effect in complex scenes.