Facial emotion recognition using convolutional neural network based krill head optimisation

Abstract

Emotion recognition from facial image expression is always a problem for computer vision algorithms in classifying the human facial emotions. On other hand, over fitting is considered as a fundamental problem that prevents the deep learning model to fit the observed data on training data and unseen data on the test data. In this paper, we develop a convolutional neural network (CNN) based emotion recognition from facial images. Initially the study uses proper background removal tool to remove the image background and secondly, we use CNN to classify the facial features based mainly on the extraction of facial feature vector. The CNN is trained with huge number of facial features from the input images causes over fitting. This is mainly due to noise presence, limited training data size and classifier complexity and to avoid this, the study uses krill herd optimisation (KHO) that enables the fitting of the classifier. The simulation is conducted to test the efficacy of the model with expressional vector that recognizes the five types of facial features. The database consists of 20,000 images collected from 300 persons and the study obtains an accuracy of 98.6% during training and 98.45% during testing. The comparative analysis between various other deep learning models show that the proposed CNN-KHO has higher rate of accuracy than other methods.