In this thesis, we introduce new methods for analysing three dimensional surface texture using high resolution normal fields and apply these to the detection and assessment of skin conditions in human faces, specifically wrinkles, pores and acne. The thesis is part of a project sponsored by Unilever with an interest in applying the outcome of the research to facial skin care product development. This explains our focus on facial skin conditions. The main contributions of this thesis are the introduction of three methods of extracting texture descriptors from high resolution surface orientation fields, a comparative study of two-dimensional and three-dimensional skin texture analysis and the collection of an extensive dataset of high resolution 3D facial scans presenting various skin conditions. The dataset includes human rating judgements on the presence of certain skin conditions. Computer aided skin condition assessment has been mostly addressed using two dimensional texture analysis techniques on skin images or coarse geometrical features extracted from the skin’s three dimensional macro structures. While the first trend ignores the three dimensional nature characterising most of the skin conditions, the latter mainly deals with geometrical features that are not fine enough to capture skin structures at the meso and micro scales. Advances in three dimensional surface imaging during the last few decades brings the possibility of capturing human skin’s fine geometrical structures and re- flectance properties with unprecedented quality and resolution (down to the level ii of the pores). The methods proposed in this work aim at leveraging these advances and revisit the formulation of texture analysis as a three dimensional problem. For data collection we set up a Lightstage to capture high resolution facial normal fields along with reflectance properties. The collected data are photo-realistically rendered and presented to the general public for annotations indicating the presence of the studied skin conditions. These constitute the ground truth on which we apply the proposed methods and learn models for detecting and assessing facial skin conditions. We also demonstrate that some of these three dimensional surface texture descriptors can be extended to synthesize highly detailed skin structures and simulate the studied skin condition on normal faces.
|Date of Award
|12 Dec 2016
|Bernie Tiddeman (Supervisor) & Hannah Dee (Supervisor)