Determinist walks on complex networks applied in computer vision

Wesley Nunes Gonçalves

Instituto de Ciências Matemáticas e de Computação - Universidade de São Paulo, 2010

Complex networks have received a growing interest in several areas of knowledge. This growth is mainly due to its flexibility in modeling and simulating topological structures that appear in our daily life. In most cases, complex networks characterization are based on basic measurements such as average degree, hierarchical degree, clustering coefficient, among others. Many of the measures are correlated, resulting in redundancy. This dissertation proposes the use of deterministic walks as a robust and efficient complex network measurement. In this measurement, walks are initiated by explorers starting from each vertex and then, informations are extracted on these walks. Experiments were performed on artificial complex networks and network modeling texture images. In artificial network recognition, the proposed method was applied to four theoretical complex network models: random, small-world, free-scale and geographical networks. In texture recognition, the method was evaluated in synthetic and real (texture of leaves) databases. In both applications, the method achieved excellent results compared with the state of the art methods

keywords: Complex networks,Computer vision,Determinist tourist walk,Texture recognition