Segment-Based Depth Estimation in Light Field Using Graph Cut

In this paper, we present a depth-extracting method on the scenes of 4D light fields. The method is based on image segmentation and epipolar plane images. We extract disparity map and reliability map from the original image of 4D light fields. Then this information is applied to image segmentation, in which a large number of planes are produced, so that the disparity map which is consist of pixels can be transferred to the disparity map which is consist of planes. In the resulting optimization problem, graph-cut technique is used to assign a corresponding disparity plane to each segment. Our method is tested on a number of synthetic and real-world examples captured with a light field camera, and compared to ground truth where available. Furthermore, an approach to optimize the method to reduce the running time is also proposed.