Machine learning for colour Palette extraction from fashion runway images
Peihua Lai, Stephen Westland
Abstract
An important aspect of colour forecasting is the process of generating colour palettes to represent collections at fashion shows. Humans have traditionally done this manually, and can do it well, but there are often too many images and it becomes an unmanageable task. In this paper, automatic machine-learning methods are developed to generate colour palettes for a fashion show based on the runway images. A set of ground-truth data to test the models was constructed based on asking each of 22 participants to select three colours to represent each of the 48 images from a particular fashion show. A close agreement was shown between these data and the colours automatically generated using a model that incorporated both supervised and unsupervised machine learning. The work could be extended to analyse millions of images from social media feeds to provide data-driven insights for colour forecasting.