Versatile Multidimensional Pattern Analysis for Automated Facial Modeling and Architecture Parsing

This thesis presents a methodology that demonstrates how 2D image processing techniques can be applied to provide solutions for 3D models. Moreover, for the research sets the aim of evaluating the feasibility and effectiveness of this methodology by its implementation in two specified areas, namely, 3D facial mesh alignment and objects recognition in CAD floor plans, respectively. Regarding the former, an image processing method called optical flow is applied in order to help the registration of 3D meshes. Additionally, the novel algorithm is evaluated by comparing its performance with that of the-state-of-art techniques. In relation to the second application, the aspect of component recognition in CAD floor plans, image processing based automatic system for analysis and labelling floor plans drawn in Computer-aided Design (CAD) form is covered. Experiments comparing the proposed system against other mature systems are evaluated, along with the outcomes of a comprehensive user survey on the novel system being presented.