Research On Image Bokeh Based On Video Image Sequence

With the development of the times, more and more people start to use mobile devices as cell-phone to take pictures and videos. However, because of the limitation of hardware conditions, the pictures taken by mobile devices usually cannot as much good as SLR cameras, one example is that SLR cameras can take high quality images with bokeh effect. In this paper, we introduce a depth reconstruction algorithm based on a batch of video frames to imitate the bokeh effect. This algorithm implement bokeh effect based on a depth map and the depth levels.In this paper, we use some key frames in a video to reconstruct the depth of the scene. First of all, in consideration of the images taken by cell-phone may exist focus failure which can leads to image blur, we reconstruct the key frames to make sure all of them are sharp enough. We use the sharp image patches to reconstruct the original blur patch and use Markov Random Field to keep the space consistency of the result image. The we use the sharp images to do depth reconstruction. We use structure from motion to recover the 3D structure and recover a depth map based on plane sweep, then we smooth the depth map in an image pyramid by MRF and guided-filter. Finally, we blur the image region chosen by the users and the depth map as background. We display the result by UI.We record the real scene by a cell-phone, and use the video image sequences to test and verify our algorithm, the results show that our depth reconstruction algorithm can recover the depth of the scene perfectly, and the out-of-focus reconstruction can also make the result more robust, and get a better bokeh result.