White Rose University Consortium logo
University of Leeds logo University of Sheffield logo York University logo

A Biophysically-based Skin Reflectance Model for Face Analysis.

Alotaibi, Sarah (2019) A Biophysically-based Skin Reflectance Model for Face Analysis. PhD thesis, University of York.

[img] Text
SarahAlotaibi_PhDThesis.pdf.pdf - Examined Thesis (PDF)
Restricted until 1 May 2025.

Request a copy


Human faces are the most distinctive feature of human appearance. Humans are extremely perceptive to recognise people and ingest the biophysical and emotional state of people. This thesis presents research aimed to use a biophysical, spectral skin colouration model to constrain the skin colours of the human face suitable for appearance face modelling tasks. More specifically, we use two free histological parameters: the melanin and haemoglobin to describe the characteristics of skin spectral reflectance. We start by focusing on developing a biophysical spectral reflectance skin model of light interaction with skin that is suitable for face analysis tasks. The model is able to represent a wide variety of skin colour types and biophysical phenomena. The model is sufficiently simple to allow skin parameters to be estimated from the image of a face as well as efficient so that it is suitable for statistical modelling and deep learning neural networks. The parameters of this model define biophysical properties of the face. Next, we propose a hybrid of biophysical skin colouration model and statistical modelling. We present methods for fitting the skin model to data captured in a lightstage and then build our hybrid model on a sample of such registered data. Then, we demonstrate a novel approach that combines a dichromatic reflectance model and our biophysical skin model to decompose multispectral human face images into its intrinsic properties: the biophysical layers: the melanin and haemoglobin, the diffuse shading and specular reflection. We show that constraining the albedo colour with our biophysical reflectance model gives a high-quality intrinsic face images estimation. In our final contribution, we take advantage of deep learning and propose a deep convolutional neural network that learns to decompose a single uncontrolled face image into biophysical parameter maps, diffuse and specular shading maps. Furthermore, we estimate the spectral power distribution of the illuminant and the spectral sensitivity of the camera.

Item Type: Thesis (PhD)
Academic Units: The University of York > Computer Science (York)
Depositing User: Mrs Sarah Alotaibi
Date Deposited: 22 May 2020 15:03
Last Modified: 27 May 2020 09:46
URI: http://etheses.whiterose.ac.uk/id/eprint/26375

Please use the 'Request a copy' link(s) above to request this thesis. This will be sent directly to someone who may authorise access.
You can contact us about this thesis. If you need to make a general enquiry, please see the Contact us page.

Actions (repository staff only: login required)