A Study on Detection of Emotions with The Help of Convolutional Neural Network

Facial Expression is a mean which can be employed to find out what is running in the mind of the person whom we talk. It is always an easy task for humans to derive some insights from the expressions. But it is a complicated task for machine to derive it by using Computer Algorithms. But recent developments in the field of Computer Vision and Machine Learning has enhanced the availability of resources so that it become possible to derive certain conclusions from the input we are giving to particular Machine. This paper is planning to propose an approach to find out what emotional stage the particular person is running which is termed as Facial Emotion Recognition using Convolutional Neural Network (FERC). The FERC constitutes two parts: The first part of this Convolutional Neural Network is used to remove background from an image and second part is used to extract features from facial expressions. It consist of a database of around 10,000 image. The final layer which is a perceptron which works in series with two layer Convolutional Neural Network. The Perceptron is used to adjust weights and exponent values after going through each iteration. Again a background removal is applied just before generation of Emotional Vector. This will help us to avoid from multiple problems which can occur. Using Two stage CNN model we found out accuracy near about 84.92 percentage which is a better value based on insights from 24 values. The Two Layer CNN works in series where remaining layer is used to adjust weights after each iteration. FERC follows an exactlynewapproach compared to Single stage CNN and thus enhancing the accuracy. Further it uses a novel background elimination procedure compared to technology of EV earlier existed which could avoid coping with coping with more than one troubles that may occur. FERC is employed with about more than 28K pictures by using FER2013 dataset. We can make use of FERC in many cases including predictive learning, Lie detection… etc.