Towards Automation and Human Assessment of Objective Skin Quantification

Alarifi, Jhan Saad A (2022) Towards Automation and Human Assessment of Objective Skin Quantification. Doctoral thesis (PhD), Manchester Metropolitan University.

Preview

Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (26MB) | Preview

Abstract

The goal of this study is to provide an objective criterion for computerised skin quality assessment. Humans have been impacted by a variety of face features. Utilising eye-tracking technology assists to get a better understanding of human visual behaviour, this research examined the influence of face characteristics on the quantification of skin evaluation and age estimation. The results revealed that when facial features are apparent, individuals do well in age estimation. Also, this research attempts to examine the performance and perception of machine learning algorithms for various skin attributes. Comparison of the traditional machine learning technique to deep learning approaches. Support Vector Machine (SVM) and Convolutional Neural Networks (CNNs) were used to evaluate classification algorithms, with CNNs outperforming SVM. The primary difficulty in training deep learning algorithms is the need of large-scale dataset. This thesis proposed two high-resolution face datasets to address the requirement of face images for research community to study face and skin quality. Additionally, the study of machine-generated skin patches using Generative Adversarial Networks (GANs) is conducted. Dermatologists confirmed the machine-generated images by evaluating the fake and real images. Only 38% accurately predicted the real from fake correctly. Lastly, the performance of human perception and machine algorithm is compared using the heat-map from the eye-tracking experiment and the machine learning prediction on age estimation. The finding indicates that both humans and machines predict in a similar manner.