Virtual Try-On System Using Image Segmentation and Pose Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current online shopping platforms lack the ability for users to effectively try on clothing before making a purchase, leading to issues with sizing and style suitability, resulting in high return rates due to the absence of a reliable virtual try-on service.
Innovation Solution
A method and system that uses image processing techniques to extract clothing data from images, complete missing parts, and overlay the clothing onto a target model, allowing users to virtually try on clothing through a web-based service, incorporating segmentation algorithms, image completion methods, and pose estimation for accurate fitting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If users are provided with a virtual try-on service, then the ability to try on clothing before purchasing is improved, but the system complexity increases
Solution Approach 1:
The system segments the virtual try-on process into distinct modules: image acquisition, clothing extraction, pose estimation, clothing transformation, and image synthesis. Each module handles a specific task independently, making the complex system more manageable and easier to implement while maintaining high accuracy through specialized processing at each stage
Solution Approach 2:
The patent introduces an intermediary clothing transformation module that acts as a mediator between the extracted clothing images and the target model. This intermediary component performs pose estimation and geometric transformation to adapt the clothing to the target model's pose and body shape, simplifying the overall system architecture while achieving accurate virtual try-on results
2Measurement precision
If detailed body measurements are collected from users, then the fit accuracy is improved, but the user input requirement increases
Solution Approach 1:
The system performs self-service by automatically extracting body measurements and pose information from uploaded user images using image processing algorithms. Instead of requiring users to manually input measurements, the system processes the images autonomously to obtain the necessary data, thereby maintaining high measurement precision while significantly reducing user input requirements
Solution Approach 2:
The patent replaces the mechanical manual measurement input process with automated image processing and computer vision techniques. The system uses algorithms to detect and measure body dimensions from images, substituting the manual measurement mechanism with automated digital processing, thus improving ease of operation while maintaining or enhancing measurement precision
3Reliability
If body scanning tools are provided, then the virtual model accuracy is improved, but the tool complexity and ease of use deteriorates
Solution Approach 1:
The system creates a virtual copy or representation of the user's body and pose from uploaded images. Instead of requiring complex physical body scanning tools, the system generates a digital model by processing and analyzing the provided images, thereby achieving high virtual model accuracy while maintaining simplicity and ease of use for the end user
Data Source
AI summary
A method and system for generating image data representative of a person wearing one or more items of clothing includes using computing hardware to process data of one or more images of people wearing items of clothing to extract via segmentation processes data representative of the items of clothing; using computing hardware to process the data representative of the items of clothing to complete one or more missing parts of the data representative of the items of clothing to generate corresponding clothes data; using computing hardware to analyse an image of a target model to determine a pose of the target model in the image; and using computing hardware to metamorphose at least a portion of the clothes data to fit the image of the target model to generate metamorphosed clothes data, and to overlay the metamorphosed clothes data onto the image of the target model.