Copy the page URI to the clipboard
Pswarai, Lizzy Vimbisai; Farrell, James; Khokan, Mahfuzur Rahman; Paria, Soumya; Giardina, Andrea and Adhikari, Kaustubh
(2022).
Abstract
This report will discuss the process of collecting skin tone measurements from a diverse group of participants and analyse the accuracies of machine-based readings through comparing with reference measurements. The machines usually take the readings of the skin tone level by generating the images and analysing them with various statistical techniques to read the information received. Most methods used today can be completed using machines and technology and we have taken advantage of this by obtaining image samples of the participants faces or/and parts of their arms and have the image processed by self-created algorithm to get the colour values of each image sample for every sample given. Instead of the results received being based on societies ethnicities the colours are received in RGB (red, green, and blue) reflectance values with a median number between 0 and 255 for each image sample processed. Using this information, we can observe the accuracy that is assumed for the machine to have when reading the skin tone of a range of skin tones and use the results as a basis to improve accuracy in machines when it comes to recognising those of a darker skin tone removing the Caucasian bias within them.