Facial Expression Recognition Using Anchor Images for Occluded Faces
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep neural networks struggle to accurately detect facial expressions, especially when they are occluded or neutral, leading to inaccurate identification of a person's state.
Innovation Solution
A neural network with two parallel branches is trained using images with and without facial expressions, utilizing shared weights to extract features and classify expressions by combining or subtracting these features, improving classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional facial expression recognition systems are used, then the system is simple to operate, but the detection accuracy deteriorates when faces are occluded or have neutral expressions
Solution Approach 1:
The system divides facial expression recognition into multiple independent analysis branches: a first neural network branch processes images to detect occluded expressions, while a second neural network branch processes anchor images to detect neutral expressions. Each branch is specialized for specific expression types, improving overall detection accuracy without requiring a single complex system.
Solution Approach 2:
The system dynamically selects which neural network branch to use based on the input image characteristics. The system determines whether the input image contains an occluded expression or a neutral expression and routes it to the appropriate specialized network, allowing the system to adapt its processing approach to match the specific detection needs.
2Reliability
If a single neural network is used for all facial expressions, then the device complexity is low, but the detection reliability deteriorates for occluded and neutral expressions
Solution Approach 1:
The facial expression recognition system is segmented into specialized neural network branches, where the first branch handles occluded expressions and the second branch handles neutral expressions. This segmentation allows each branch to be optimized for its specific function, improving reliability for different expression types.
Solution Approach 2:
The system introduces an intermediate determination step that analyzes the input image to identify whether it contains an occluded expression or a neutral expression. This intermediary classification guides the selection of the appropriate neural network branch, ensuring that the most suitable processing path is chosen for each input type.
3Measurement precision
If deep neural networks are used to extract high-level features, then the feature extraction capability is improved, but the difficulty of detecting facial expressions deteriorates due to neutral and occluded states
Solution Approach 1:
The system applies different processing qualities to different expression types: the first neural network branch uses feature extraction optimized for occluded expressions, while the second branch uses feature extraction optimized for neutral expressions. Each branch extracts features with local quality tailored to its specific detection target, overcoming the limitations of generic feature extraction.
Solution Approach 2:
The system dynamically adjusts its detection approach based on the expression type detected in the input image. When an occluded expression is detected, the system activates the first neural network branch with appropriate feature extraction capabilities. When a neutral expression is detected, the system switches to the second branch, making the detection process adaptive to the specific challenges of each expression type.
Data Source
AI summary
System and techniques are described herein for processing images to detect expressions of a subject. In one illustrative example, a method of recognizing facial expressions in one or more images includes obtaining, by a computing device, a first image of a person; obtaining expression information based on the first image and an anchor image associated with the person; and determining an expression classification associated with the first image based on the expression information.