An efficient semi-dynamic ensemble pruning method for facial expression recognition

Danyang Li,Guihua Wen,Zhuhong Zhang

DOI: https://doi.org/10.1007/s11042-024-18329-2

IF: 2.577

2024-02-14

Multimedia Tools and Applications

Abstract:Dynamic ensemble pruning (DEP) can select the best set of models for an unseen sample to improve a classification system's prediction accuracy and robustness. However, little work has been done to apply DEP to the facial expression recognition (FER) system and reduce DEP's computational burden. This paper proposes a semi-dynamic ensemble pruning algorithm (SDEP) to implement the FER task. First, a subspace-based classifier sequences selection method is proposed to break DEP's neighborhood construction limitation and avoid the exhaustive similarity calculation. Moreover, a pre-prediction mechanism is introduced to calculate the relationship between an unseen sample and data subspaces. Ultimately, the calculated relationship scores and their credibility will be used as the weights for fusion. SDEP reduces the DEP's computational burden while retaining the dynamic characteristics through changing the classifier selection into the subspace-based classifier sequences combination. SDEP demonstrates superior classification accuracy and efficiency to several state-of-the-art dynamic and static EP algorithms in benchmark FER datasets Fer2013, JAFFE, CK+ using a homogeneous ensemble of 799 CNN classifiers.

computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve

The main purpose of this paper is to propose a new Semi-Dynamic Ensemble Pruning (SDEP) method to improve Facial Expression Recognition (FER) tasks. Specifically, the SDEP method aims to address the issues of excessive computational burden and potential errors introduced during the local region construction process in traditional Dynamic Ensemble Pruning (DEP) methods.

#### Main Objectives:
1. **Reduce Computational Burden**: By changing the classifier selection method, transforming classifier selection into a subspace-based classifier sequence combination, thereby reducing the computational burden of DEP.
2. **Improve Classification Accuracy**: By removing the local region construction process while maintaining dynamic characteristics, thus improving the prediction accuracy and robustness of the classification system.
3. **Effectively Distinguish Mixed Expressions**: Efficiently identify mixed expressions using fewer classifiers and improve recognition accuracy.

#### Method Innovations:
- **Removal of Neighborhood Construction Process**: SDEP surpasses typical DEP methods by enhancing performance through the removal of the neighborhood construction process.
- **Dynamic Combination of Subspace Classifier Sequences**: Dynamically selects and combines subspace classifier sequences based on the relationship between the test sample and different categories, thereby objectively selecting classifiers suitable for FER features of different intensities and similarities.

Through these improvements, SDEP demonstrates superior classification accuracy and efficiency over existing dynamic and static ensemble pruning algorithms on standard FER datasets.