Improving the Accuracy of Human Emotion Recognition through CNN Layering Architecture

Abstract

In accordance with the proverb "a picture is worth a thousand words," facial expressions have the ability to convey a wide range of emotions depending on the circumstances. Observing facial expressions during face-to-face interactions allows us to infer the thoughts and feelings of others. Therefore, the development of facial expression recognition systems holds great significance in the field of artificial intelligence. Recognizing facial expressions involves several fundamental stages. Firstly, the images used in the system undergo pre-processing. Following this, important features are collected and extracted from the dataset images. Finally, the resulting expressions from the images are classified using prediction techniques. This study employed convolutional neural networks (CNNs), a well-established image processing method with a layered architecture, to develop an efficient model for accurately recognizing facial expressions with optimal accuracy. To thoroughly analyze the proposed model, the following datasets were utilized: Facial Expression Recognition 2013 (FER2013), Extended Cohn-Kanade (CK+), and Japanese Female Facial Expressions (JAFFE). These datasets consist of supervised data and cover seven different emotions: neutral, happy, angry, sad, fear, disgust, and surprise. The findings demonstrate that the suggested technique consistently outperforms contemporary methods across all of the aforementioned datasets, achieving notable improvements in accuracy.