Geometric Analysis of Textured 3D Scenes via Locally Scaled Point Processes

• Activity Number: 173
• Type: Topic Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Thordis Thorarinsdottir*+ and Eva-Maria Didden and Alex Lenkoski and Christoph Schnoerr
• Companies: Norwegian Computing Center and Heidelberg University and Norwegian Computing Center and Heidelberg University
• Keywords: point process ; image analysis ; image texture ; local scaling

Abstract

Image texture with a regular structure may often be described by a marked point process where only the marks -- the image texture -- are observed, while the underlying point pattern is latent. The goal of the inference is usually to describe the distribution of the latent point pattern. In particular, for 2D texture images with depth effects, the aim is to describe the underlying 3D properties of the texture. We introduce a statistical framework based on locally scaled point processes to extract 3D information from 2D images with irregular texture. Our framework locally accounts for imperfect image processing related to texture element detection, and globally for a moderately irregular random spatial arrangement of elements forming the texture, based on explicit and non-restrictive mathematical assumptions.