Sketching 3D faces

O.Gunnarsson, "Sketching 3D faces", PhD, 2010. (Supervisor: Dr Steve Maddock). [pdf]

PhD Abstract: Faces can be modelled using a number of techniques. Existing faces can be transferred to a digital form using equipment such as laser scanners. New faces can be constructed using commercial modelling tools or using specialist software such as parameterised based systems. This thesis presents a technique for modelling and posing 3D face models using an intuitive sketch-based approach. In contrast to existing sketch-based systems that manipulate faces, it gives the user an unhindered and natural sketching experience as there is no need to sketch helper strokes, incrementally alter parameters, or define which features are being sketched in order to aid the modelling/posing process. A Non-Photorealistic Rendering (NPR) technique creates a sketch-like representation of the 3D faces. Hidden Markov Models (HMMs) map a sequence of sketched points to the NPR data which is used in a maximum likelihood framework to generate a face model based on the observed data. It is shown that this method can produce a range of novel 3D faces for the purposes of modelling new facial features, as well as quickly posing existing features to create key poses in an animation sequence. Two user evaluations are presented here. The first evaluation verifies that the use of contour data of a face can be used to reconstruct the full face mesh while preserving the facial features. The second evaluation demonstrates how this sketch-based technique can be used to recreate faces from description by novice users.