Computer vision for skin colour invariant detection of face acne & pigmentation

• Main Author: Chaudhary, Nidhi Narsingh
• Other Authors: Alex Chichung Kot
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/167726
• Institution: Nanyang Technological University

It is important to take good care of facial skin. However, for an average consumer, it can be challenging to understand what issues our skin can face. Additionally, getting the opinion of a dermatologist for just mild skin issues can be unnecessarily expensive. With acne and pigmentation being some of the most common dermatological issues, the healthcare industry is looking for ways to automate the diagnosis and detection of such spots. However, there is a lack of research into creating a model that performs well in detecting facial spots across all skin tones. To address this issue, this study will explore methods of skin decomposition to extract haemoglobin and melanin components from facial images and propose new architectures for networks such as U-Net++ and Vision Transformers. As there is no public benchmark set regarding cosmetic spot segmentation, this study will also set a benchmark as a comparison for the proposed methods. The best model out of all methods tested will then be used in an application for spot detection. This study uses a private dataset provided by an industry partner. Using the proposed methods, it is able to achieve improvements in fairer spot segmentation. Through this exploration of multiple methods, this study aims to provide future research with a guide on plausible methods of reducing skin tone bias in spot segmentation.