Abstract
REAL TIME FACE AND EMOTION RECOGNITION USING CNN
Nazmin Begum*, Dr. Md Shoaibuddin Madni and Dr. Ismath Unnisa
ABSTRACT
One of the most exciting areas of research, facial expression recognition, has attracted the attention of many academics for many years. An emotion recognition system based on the concept of convolutional neural networks is introduced in our paper (CNN). Prior to convolution, the image of the facial expression is first normalised, and then the edges of each layer are recovered from the image. Each of the feature images has the retrieved edge information applied to it in order to preserve the edge structure information of the picture's texture. The max pooling method was employed in an effort to reduce the dimensionality of the recovered implicit features. A SoftMax classifier is then used to classify and identify the expression of the test sample image. In order to recognise face expressions, this research aims to learn and identify information representations from 2-dimensional gray-scale pictures. The learned features are provided via a constructed convolutional neural network (CNN). The developed CNN model facilitates fast learning of information from images by cascading different layers together. Because it does not contain a high number of layers and handles the overfitting problem at the same time, the developed model is computationally efficient. Using different datasets such as FER-2013, CK+, and image datasets, our suggested approach assists us in focusing on crucial aspects of human faces to detect emotion.
[Full Text Article] [Download Certificate]