Virtual Try-On and Customization

Abstract

The rapid growth of online jewelry shopping has introduced several challenges, including a lack of user confidence, limited personalization, and increased return rates due to the absence of a "try-before-you-buy" experience. To address these challenges, this paper reviews recent advancements in virtual jewelry try-on and recommendation systems through the development of "VirtualGems," an AI-powered web-based platform that integrates computer vision, machine learning, and intelligent recommendation techniques to enhance the online jewelry shopping experience. The system utilizes real-time facial detection and landmark analysis using MediaPipe and OpenCV to accurately overlay jewelry images onto the user’s live camera feed, enabling an interactive virtual try-on experience. Additionally, the platform analyzes the user's facial structure and automatically classifies face types to generate personalized jewelry recommendations using a database-driven recommendation model. Unlike traditional systems that rely on heavy augmented reality frameworks such as ARCore or Unity, VirtualGems employs a lightweight and scalable architecture built with a Django-based backend and a React-based frontend, ensuring flexibility, performance, and ease of deployment. The system also stores user interaction data, try-on history, and customization records to improve recommendation accuracy and user engagement over time. This paper evaluates the architecture, performance, advantages, and limitations of AI-assisted virtual try-on systems like VirtualGems and highlights their potential to improve user confidence, personalization, and satisfaction in online jewelry shopping environments.