Fast EPI based depth for plenoptic cameras

We present a depth estimation method for light field cameras by making use of Epipolar Plane Image (EPI) representations of the microlens sub images. Light field raw image structure has several sub images and generally depth is estimated using multi baseline techniques or object labeling schemes. Both these approaches are quite complex. EPI representation of the sub images when applied to the multi-baseline framework provides outputs comparable to that of the multi-baseline approach along with significant reduction in complexity (50%). Fast depth estimation will become an important requirement as plenoptic cameras gets into the main stream. The proposed method has been tested on a couple of datasets captured with different light field camera setups and was found to perform as well as the multi-baseline approach in terms of quality of the depth map generated but taking only half of the computational time.