Gibbs sampling for 2D cane structure extraction from images

Abstract

In this paper we are interested in recovering 2D tree structure of vines from binary images. We propose a bottom-up approach that firstly segments an input image into cane parts, and second infer their connectivity by using Gibbs Sampling. Our approach is similar to previous work on vine structure inference [1], but instead of the use of heuristics for connecting cane parts, our method uses Gibbs sampling which has been successfully used in similar computer vision tasks [2]. We show comparative results against [1], and we provide directions on how this work could be extended in the future.