Open AccessGA-APEXNET: GENETIC ALGORITHM IN APEX FRAME NETWORK FOR MICRO-EXPRESSION RECOGNITION SYSTEM
Author(s) -
Qiushi Jin,
Huang-Chao Xu,
Kun-Hong Liu,
SzeTeng Liong,
Yee Siang Gan,
Shu-Wen Su
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1544/1/012149
Subject(s) - computer science , overfitting , artificial intelligence , robustness (evolution) , feature selection , genetic algorithm , pattern recognition (psychology) , frame (networking) , feature (linguistics) , expression (computer science) , artificial neural network , algorithm , machine learning , telecommunications , biochemistry , chemistry , linguistics , philosophy , gene , programming language
This paper introduces a novel method to recognize the facial micro-expressions by directly adopting state-of-the-arts deep learning algorithms. Specifically, genetic algorithm (GA) is applied based on the principles of evolution in searching an optimal solution to generate discriminant features. Prior to that, the feature vectors of each video are first encoded using existing apex-based feature descriptors, viz, Bi-WOOF, OFF-ApexNet and STSTNet. Then, GA is utilized to enrich the features by eliminating irrelevant information that do not contribute to the expression prediction. It is acknowledged that the overfitting phenomena can be avoided in the feature selection process, as GA employs tournament selection and deterministic mutation procedures during the evolution procedure. As a result, Genetic Algorithm in Apex Frame Network (GA-ApexNet) is introduced and significant improvement of the recognition performance has been achieved. A standard experimental configuration is designed to evaluate the robustness of the proposed framework. Following the suggestion in Facial Micro-Expressions Grand Challenge (MEGC 2019), the UF1 and UAR obtained are 78.85% and 78.28%, for the composite dataset that comprises CASME II, SMIC and SAMM. We also note that this is the first work that fuses GA in micro-expression recognition system.