Tamil Emotion Classification Using Attention-CNN with Hybrid Seagull-Grasshopper Optimization

View at publisher Summarise with VIRAT
Authors: Cynthia Joseph, A. Rajeswari
Cited by: 0
References: 20

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

In this study, the Meypaatiyal dataset forms a basis for the Tamil emotion classification system, capturing authentic emotional expressions from drama and theatre situations. Previous face emotion identification research has been based on shared emotions and has generally overlooked the subtlety and richness of cultural variation in the emotional spectrum, such as the eight primary emotions described in “Tolkappiyam.” Other limitations of the traditional approaches include the failure of existing approaches to properly account for the diversity and complexity of such emotions as nagai, azhugai, ilivaral, marutkai and many others. In this study, the limitations have been overcome and a robust Local Artificial Attention-based Convolutional Neural Network–Hybrid Seagull and Grasshopper Optimization Algorithms model is proposed for Tamil emotion classification. Initially, the Meypaatiyal dataset with 948 images of facial expressions and a preprocessing process of resizing using KNN-RNN and the data are normalized using Z-score normalization for improved model performance. Techniques for augmenting data, such as flipping, scaling, translation and rotation, are employed. While Local Binary Patterns and Artificial Neural Networks are applied for optimizing the extraction of features, Hybrid Seagull and Grasshopper Optimization Algorithms are used for feature selection. These extracted features classify focus-attention-based convolutional neural networks, focusing on areas and condensing it to obtain accurate predictions in emotions. Lastly, the performance metrics used in this study revolve around F-measure, NPV, FPR, FNR, accuracy, precision, sensitivity and specificity, where the achieved accuracy is 99.23%.

Fields of Study & Subjects

Emotion and Mood Recognition Face and Expression Recognition Face recognition and analysis

Publication details

Journal / source
Journal of Circuits Systems and Computers
Pages
2650208
Publication date
15 May 2026
Language
EN