Personalized Facial Expression Classification via Synthesized AU Datasets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning systems for facial expression classification are not accurate when trained with generic datasets across all faces, as they fail to capture individual variations in facial Action Units (AUs) and their intensity categories, leading to inefficient classification and increased computational costs.
Innovation Solution
A method is developed to create a personalized dataset by synthesizing new images with specific AU combinations and intensity categories based on a subject's facial image, which is then used to train a machine learning system for improved classification accuracy, reducing the need for a large initial dataset and enhancing computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If generic datasets across all faces are used for training, then the machine learning system can process diverse facial expressions, but the classification accuracy decreases due to failure to capture individual variations in facial Action Units and intensity categories
Solution Approach 1:
The patent applies local quality by transitioning from generic datasets to personalized datasets that capture individual variations in facial Action Units and intensity categories for each subject. The system synthesizes images specific to each individual's facial characteristics, enabling the machine learning system to accurately classify facial expressions while accounting for person-specific differences in muscle geometry and expression patterns.
2Measurement precision
If large initial datasets are used to improve classification accuracy, then the machine learning system can capture individual variations, but computational costs increase
Solution Approach 1:
The patent uses copying by synthesizing virtual images of facial expressions for each subject based on their unique facial characteristics and Action Unit patterns. Instead of requiring large quantities of real images for training, the system creates synthetic copies that replicate individual variations, thereby achieving high classification accuracy while significantly reducing computational costs associated with collecting and processing extensive real-world datasets.
3Measurement precision
If real images are collected for each subject to capture individual variations, then classification accuracy improves, but the time required for data collection increases
Solution Approach 1:
The patent applies preliminary action by pre-synthesizing images for each subject before actual classification tasks. The system creates a personalized dataset of synthetic facial expressions that captures individual variations in advance, eliminating the need for time-consuming real-time image collection during the classification process. This preliminary synthesis of training data significantly reduces the time required to prepare subject-specific datasets while maintaining high classification accuracy.
Data Source
AI summary
A method may include obtaining a facial image of a subject and identifying a number of new images to be synthesized with target AU combinations and categories of intensity. The method may also include synthesizing the number of new images using the facial image of the subject as the base image with the number of target AU combinations and categories of intensity with a number of new images that have different AU combinations than the facial image of the subject. The method may additionally include adding the number of new images to a dataset and training a machine learning system using the dataset to identify a facial expression of the subject.