Autoencoder Makeup Removal for Face Recognition Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current face recognition systems face challenges in performance due to age, spoofing, and facial makeup, which alter and hide the original appearance, making recognition or verification tasks more difficult, and existing research on facial beauty prediction relies heavily on landmark annotation and lacks a unified criteria for attractiveness evaluation.
Innovation Solution
A system and method for analyzing facial makeup on digital images, using locality constraints on discriminative low-rank dictionary learning to detect and remove makeup, categorize makeup styles, and evaluate attractiveness by training autoencoders on attractiveness scores, allowing for fully automatic prediction without landmark annotation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If facial makeup is applied to alter appearance, then attractiveness is improved, but face recognition accuracy deteriorates
Solution Approach 1:
The patent detects facial makeup regions and uses them as guidance to focus computational resources on removing or neutralizing the makeup's interfering effects. By identifying where makeup is applied, the system can针对性地 (targeted) process those regions to restore original facial features for recognition, converting the harmful makeup into a detectable pattern that guides the restoration process
Solution Approach 2:
The patent extracts and removes makeup components from facial images by detecting makeup regions and applying removal algorithms. This extraction process separates the makeup layer from the underlying facial features, allowing the recognition system to access the original face structure without makeup interference
2Measurement precision
If traditional facial beauty evaluation methods are used, then attractiveness prediction is achieved, but heavy landmark annotation is required
Solution Approach 1:
The patent replaces manual landmark annotation with automatic facial feature detection algorithms. Instead of requiring experts to manually mark facial landmarks, the system uses computer vision techniques to automatically identify and locate facial features, substituting mechanical human annotation work with automated computational methods
Solution Approach 2:
The system performs self-annotation by automatically detecting and marking facial landmarks without external human intervention. The facial detection algorithm autonomously identifies key points on faces, enabling the beauty evaluation system to operate without requiring manual preparation of annotated training data
3Adaptability or versatility
If geometric feature-based beauty scoring is used, then facial symmetry evaluation is achieved, but non-unified attractiveness criteria are obtained
Solution Approach 1:
The patent develops a unified beauty evaluation framework that integrates multiple evaluation dimensions (symmetry, proportion, texture, color) into a single comprehensive system. This universal framework can evaluate different types of facial features using consistent criteria, making the attractiveness assessment adaptable across diverse populations and contexts while maintaining unified standards
Data Source
AI summary
A system and method are provided to detect, analyze and digitally remove makeup from an image of a face. An autoencoder-based framework is provided to extract attractiveness-aware features to perform an assessment of facial beauty.