Conventional approaches to image de-fencing use multiple adjacent frames forsegmentation of fences in the reference image and are limited to restoringimages of static scenes only. In this paper, we propose a de-fencing algorithmfor images of dynamic scenes using an occlusion-aware optical flow method. Wedivide the problem of image de-fencing into the tasks of automated fencesegmentation from a single image, motion estimation under known occlusions andfusion of data from multiple frames of a captured video of the scene.Specifically, we use a pre-trained convolutional neural network to segmentfence pixels from a single image. The knowledge of spatial locations of fencesis used to subsequently estimate optical flow in the occluded frames of thevideo for the final data fusion step. We cast the fence removal problem in anoptimization framework by modeling the formation of the degraded observations.The inverse problem is solved using fast iterative shrinkage thresholdingalgorithm (FISTA). Experimental results show the effectiveness of proposedalgorithm.
展开▼
