Abstract
This thesis takes as a starting point a segmentation frame-work for visual scenes based on Markov Random Fields. The approach integrates several cues and considers the scenein 3D, improving the hypotheses in an iterative manner.
In this thesis, we substitute the hardware previouslyused for a Kinect, a device capable of creating high qualitydepth maps of the environment.
The main interest of this work is to study the capa-bilities that the exploitation of said map can offer to thesegmentation framework.
Specifically, we focus on the provision of an alterna-tive attention mechanism to select fixations points for thesegmentation, as well as two sets of cues. All of it basedexclusively in the depth data retrieved from the Kinect.
The first set of cues is based in the detection of edgesin said depth data, while the second set relies on the lackof depth information in specific areas