Segmentação Semi-supervisionada de Imagens através de Dinâmicas Coletivas em Redes Complexas

Abstract

Image segmentation is a technique that divides the image into regions of interest, such as objects in a landscape. Image segmentation algorithms present variations in their types of learning, including unsupervised, supervised, and semi-supervised. In the context of interactive segmentation, the challenge is to segment objects from the background with the help of initial labels provided by a user. Superpixels are unsupervised segmentation algorithms used as pre-segmentation for various image problems, such as classification and segmentation. Complex networks are graphs with non-trivial structures used to represent certain data domains, such as regions of an image and their neighborhoods. Collective dynamics in a complex network refer to the emergent and interactive behavior of various elements or actors within an interconnected and complex network, where the actions of one element can influence the actions of others. In this work, we propose a semi-supervised image segmentation algorithm that combines the techniques of superpixels, complex networks, and collective dynamics. The method was evaluated under various conditions using the GrabCut database for interactive segmentation. Our method proved to be as good as the best interactive segmentation methods based on graphs and in the precision metric it surpassed all compared methods.