Shape analysis using complex networks

Abstract

Nowadays, image analysis is one of the most important tasks carried out by computers. The basic information used for image analysis are color, texture and shape. 2D shape can be understood as a set of connected dots that presents a specific distribution pattern in the image captured by sensors. There are in literature several methods for 2D shape description, including methods based on complex networks. The term complex network refers to a graph or a network that shows a non-trivial topographic structure and has characteristics that do not appear in simple networks, such as the coefficient of clustering, the number of edges that are connected to a vertex, the network resilience, and so on. Recently, we have proposed a novel 2D shape descriptor method called HTS (Hough Transform Statistics), which uses Hough space statistics in order to characterize 2D shapes. HTS, besides presenting good shape recognition rates, has linear complexity, which is a very important characteristic for image retrieval in very large image databases. In this current research project we intend to study 2D shape recognition methods based on complex networks found in literature, to investigate alternatives to represent the Hough space as complex networks, and, finally, to propose new 2D shape descriptors extracted from such complex networks, extending the HTS. The 2D shape description methods found in literature and the novel methods proposed in this work will be compared by using some classical measures, like precision, recall and multi-scale separability, on some public shape databases, such as Kimia and MPEG-7.