Recommender system for fashion using machine learning algorithms for image recognition

Abstract

The main objective of the research is to check the potential effectiveness of automated personal stylist system, based on recommendation engine. The following research will help to test effectiveness of simplified and automated version of personal stylist. The research base was inspired by article written by Dale Markowitz [1]. The more theoretical the following technology [2] developed by Cisco. The main functionality, used in project based on the following approach. User inserts URL of inspiration picture to the application, or path to already downloaded inspiration picture, on the local machine. Outfit inspiration image is used as a representation of style most similar to the one desired by the user. Meaning the main goal of recommendation model is to return a combination of wardrobe items, which are most similar to combination of items, present on inspiration image. The algorithm tries to identify clothing items present on picture. Afterwards, for each identified clothing item, algorithm is looking for most similar clothing items, within users’ wardrobe. To improve user experience, number of recommended wardrobe items, for each identified item, is bound up to three. The following research might help to prove if users’ familiarity of his own wardrobe items will help to increase general satisfaction. Based on the idea that user already has an idea whether proposed items will be a good match in real life, based on size, fabric etc.