Multi-Scale Spatial Attention for Robust Facial Expression Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Facial Expression Recognition (FER) technologies face challenges in complex scenarios due to lighting changes, facial occlusions, and subtle expression changes, leading to inaccurate recognition of emotional states.
Innovation Solution
A novel FER framework (HNFER) utilizing multi-scale feature extraction and spatial attention mechanisms, including a CoordAtt attention mechanism and SAFM module, to enhance feature extraction and classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional FER methods use static and global feature extraction mode, then the system complexity is low, but the recognition accuracy in complex scenarios deteriorates
Solution Approach 1:
The patent divides the feature extraction process into multiple scales (different pooling ratios) to capture facial features at various levels of detail. This segmentation allows the model to analyze both global facial structure and local expression details simultaneously, resolving the contradiction between simple global extraction and complex detailed extraction.
Solution Approach 2:
The patent introduces a new dimension by adding spatial attention mechanisms (CoordAtt and SAFM modules) that process features across different spatial dimensions and scales. This transforms the traditional single-scale feature extraction into multi-scale feature analysis, improving recognition accuracy without linearly increasing system complexity.
2Measurement precision
If conventional FER methods use single-scale feature extraction, then the processing speed is fast, but the ability to capture subtle expression changes deteriorates
Solution Approach 1:
The patent segments the feature extraction into multiple scales using different pooling ratios (e.g., 1:4, 1:8, 1:16), allowing each scale to capture different levels of expression details. This enables the model to detect subtle expression changes while maintaining manageable processing complexity through hierarchical organization.
Solution Approach 2:
The patent applies spatial attention mechanisms that assign different weights to different spatial regions and scales of facial features. This local quality approach allows the model to focus computational resources on critical regions (e.g., mouth corners for smiling detection) while maintaining overall processing efficiency.
3Reliability
If conventional FER methods lack dynamic attention to local features, then the system simplicity is maintained, but the robustness to occlusions and lighting changes deteriorates
Solution Approach 1:
The patent implements dynamic attention mechanisms (CoordAtt and SAFM) that adaptively adjust feature weights based on input characteristics. This allows the model to dynamically focus on relevant facial regions regardless of occlusions or lighting conditions, improving robustness while the modular architecture keeps complexity manageable.
Solution Approach 2:
The spatial attention mechanisms serve multiple functions: they handle occlusions by focusing on visible regions, compensate for lighting variations through adaptive weighting, and capture expressions at multiple scales. This multi-functionality improves reliability without requiring separate specialized modules for each condition.
Data Source
AI summary
The present invention relates to the technical field of expression recognition, and in particular, to an expression recognition method and system based on multi-scale features and spatial attention. The method includes: performing feature extraction on acquired facial image data by using an HNFER neural network model to obtain an original input feature map; performing pooling and concatenation on extracted features based on a CoordAtt attention mechanism to obtain a feature map; performing deep convolution processing on the feature map to obtain an attention map, and then performing element-by-element multiplication to obtain a final feature map; and performing feature transformation and normalization on the final feature map to obtain an expression category probability and output the expression category probability. In the present invention, by integrating scale perception and spatial attention technologies, the model can recognize and classify different emotional states more accurately and maintain high performance even under complex environmental conditions.