Facial Expression Recognition via Adversarial Elimination

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Facial expression recognition in natural environments is challenging due to factors like lighting, occlusion, and small sample sizes, leading to network overfitting and classification errors, especially when using attention mechanisms that require additional resources and multiple sub-networks for feature extraction.

Innovation Solution

A method employing adversarial elimination to preprocess data, generate diverse facial expression recognition sub-networks with different weight distributions, and integrate them using a relative majority voting method to improve accuracy and reduce resource usage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a network for recognizing facial expressions based on an attention mechanism is used, then the recognition accuracy on natural expression data set is improved, but the device complexity and resource consumption increase due to additional input images and multiple attention sub-networks

Engineering Contradiction:

Improverecognition accuracyVSAvoiddevice complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts and eliminates the most influential interference factors (occlusion, lighting, posture) from the training data through adversarial elimination. By removing these harmful factors during training, the network learns more robust features that are less sensitive to such variations, thereby achieving high accuracy without requiring complex attention mechanisms or multiple sub-networks.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the training parameters by introducing adversarial examples with controlled occlusion, lighting, and posture variations. Through gradual adjustment of these parameters during training, the network adapts to handle natural environment variations, improving recognition accuracy while maintaining a simple network structure.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If a network for recognizing facial expressions based on an attention mechanism is used, then the recognition accuracy on natural expression data set is improved, but the use of energy increases due to parallel network operation

Engineering Contradiction:

Improverecognition accuracyVSAvoiduse of energy

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent merges multiple training objectives into a single unified training process. Instead of running multiple networks in parallel, the method combines adversarial training with standard supervision in one network, reducing energy consumption while maintaining or improving recognition accuracy.

Inventive Principle:

Principle #5Merging (Combining)

3Adaptability or versatility

If natural facial expression data set is used for training, then the network is closer to real scenes, but the network overfitting phenomenon becomes serious due to small samples and interference factors

Engineering Contradiction:

Improveadaptability to real scenesVSAvoidnetwork overfitting

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent applies preliminary anti-action by introducing adversarial examples with occlusion, lighting, and posture variations during training. This pre-exposes the network to potential interference factors, enabling it to learn robust features that resist overfitting and generalize better to real-world scenarios.

Inventive Principle:

Principle #9Preliminary anti-action

Solution Approach 2:

The patent converts harmful interference factors (occlusion, lighting variations, posture changes) into beneficial training signals. By deliberately introducing these factors as adversarial examples during training, the network learns to ignore them and focus on relevant facial expression features, thereby reducing overfitting and improving generalization.

Inventive Principle:

Principle #22Blessing in disguise (Convert harm into benefit)

Data Source

PatentUS11804074B2Method for recognizing facial expressions based on adversarial elimination

Publication Date: 2023.10.31 DIBI (CHONGQING) INTELLIGENT TECH RES INST CO LTD

AI summary

The present disclosure relates to a method for recognizing facial expressions based on adversarial elimination. First, a facial expression recognition network is built based on a deep convolutional neural network. On a natural facial expression data set, the facial expression recognition network is trained through a loss function to make facial expression features easier to distinguish. Then some key features of input images are actively eliminated by using an improved confrontation elimination method to generate a new data set to train new networks with different weight distributions and feature extraction capabilities, forcing the network to perform expression classification discrimination based on more features, which reduces the influence of interference factors such as occlusion on the network recognition accuracy rate, and improving the robustness of the facial expression recognition network. Finally, the final expression classification predicted results are obtained by using network integration and a relative majority voting method.