AI Skin Spectral Reconstruction for Personalized Condition Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional hyperspectral imaging and RGB approaches struggle to accurately identify skin conditions due to information loss, leading to incorrect identification and ineffective treatment, especially across diverse skin types, and existing computer-related techniques fail to capture user-specific skin needs, making it difficult for users to access effective skin care products.
Innovation Solution
AI-based systems generate and evaluate reconstructed multi-spectral images using a hyper-spectral reconstruction model, skin attribute model, skin mapping model, cosmetic attribute model, and population model to analyze user-specific skin conditions and recommend personalized skin care products.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional hyperspectral imaging is used to capture detailed spectral information, then measurement precision is improved, but device complexity increases and temporal resolution decreases
Solution Approach 1:
The patent uses a generative adversarial network (GAN) to create a virtual hyperspectral imaging system that copies and reconstructs spectral information from standard RGB images. Instead of requiring actual hyperspectral cameras, the system learns to map RGB pixel values to hyperspectral reflectance values through training on paired datasets, thereby achieving spectral analysis capabilities using conventional imaging devices.
Solution Approach 2:
The patent replaces the mechanical/optical complexity of hyperspectral imaging systems with an artificial intelligence-based computational approach. The GAN model substitutes physical spectral sensors and complex optical paths with neural network layers that process standard RGB images to generate spectral signatures, eliminating the need for specialized hyperspectral hardware.
2Measurement precision
If hyperspectral imaging is used to identify skin conditions, then measurement precision is improved, but loss of time increases due to operational complexities
Solution Approach 1:
The system performs preliminary training of the GAN model on extensive datasets of paired RGB and hyperspectral skin images before actual use. This pre-computation phase enables the model to quickly infer spectral information from RGB images during runtime without requiring time-consuming hyperspectral data acquisition or complex real-time processing, thus reducing operational time while maintaining precision.
3Device complexity
If conventional RGB imaging is used to simplify the system, then device complexity is reduced, but loss of information increases due to integration of hyperspectral radiance into RGB values
Solution Approach 1:
The patent introduces an intermediary computational model (GAN) that acts as a bridge between standard RGB images and spectral information. The GAN learns the complex mapping relationship between RGB pixel values and hyperspectral reflectance, effectively recovering spectral information that would otherwise be lost in RGB integration. This intermediary layer enables information recovery without requiring actual hyperspectral sensors.
Data Source
AI summary
Artificial intelligence-based systems and methods are described for generating and evaluating reconstructed multi-spectral images depicting skin. A digital image of a user is received at an imaging application (app) and comprises pixel data of at least a portion of a skin area of the user. A hyper-spectral (HS) reconstruction model, trained with pixel data of a plurality of digital images depicting human skin, outputs one or more reconstructed HS images, which can be used as input to one or more AI models. The imaging app generates, based on output from the one or more AI models, user-specific comparison data of the user, reconstructed HS images of the user, or mapping data of the user.