Plenoptic image segmentation with region-based graph cut approach

Seongjin Park, DoHyung Kim, Seong-Jun Bae, Jae Woo Kim, Ho Wook Jang

Abstract: Plenoptic images captured from plenoptic cameras provide angular information as well as conventional spatial information of the scene. Using the angular information, plenoptic imaging can generate multiview images. In this paper, we developed approaches for object segmentation on 4D hyper volume represented with the multiview images. Our approaches are based on graph cut algorithms and applied connected component labeling to divide pre-segmented regions to reduce computation time. The region-based graph cut segmentation is composed of region graph creation, weight computation, and graph cut optimization on the pre-segmented regions. The experiments demonstrated that the proposed method produced more accurate results comparing to the results from conventional graph cut methods.

External IDs:dblp:conf/gcce/ParkKBKJ17