Residual spaces based on component trees: theory and applications

Abstract

This thesis presents a characterization (i.e., definitions, properties and algorithms) of residual spaces obtained from spaces of primitives based on component trees. Residual spaces are hierarchical structures constructed from image regions from which we can perform image analysis efficiently. For that, we can analyze residual regions by means of its maximum residual values leading to the called maximum residual operators. Although such operators extract relevant information, they do not take into account the hierarchy of the residual spaces, which means that they may extract residues from undesirable regions. In another point of view, in this thesis we present a novel approach to analyze residual spaces through a hierarchical structure called resid- ual tree. From this structure, we extract attribute vectors to build a machine learning model which gives a matching value between ground truth regions and residual tree nodes (or regions). After, from the selected residual tree nodes, we present a new approach to choose the best residual nodes. Finally, we show that it is a solution to the residual space analysis problem. In order to evaluate our new approach, some experiments were carried out with a plant dataset and results report the state-of-the-art performance in plant detection and segmentation.