International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 03 | Mar 2019 www.irjet.net p-ISSN: 2395-0072 © 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 5010 Emotion Classification of Human Face Expressions using Transfer Learning S. Poonkodi 1 , D. Alen Benard 2 , Sornapudi Aditya Vineeth Raj 3 , G. Sowmiya 4 , V.Geetha 5 1,2,3,4 Final Year B. Tech, Dept. of Information Technology, Pondicherry Engineering College, Puducherry, India 5 Associate Professor, Dept. of Information Technology, Pondicherry Engineering College, Puducherry, India ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - Emotive analytics is an interesting fusion of psychology and technology. The ability to understand a person’s emotion from face is quite difficult. Moreover, it plays a vital role in non-verbal communication since 90% of our communication is non-verbal. Frequently, the words that the person speaks does not match with how they feel. Different types of emotions are available which have influence on how an individual live and communicate with others. The decisions, actions, and the perceptions that humans have are all influenced by the emotions that people experience at any given moment. According to psychological research, there are six universally accepted facial emotions. They are happiness, sadness, disgust, anger, fear and surprise. In this paper, the emotion is classified with the given test image starting from processing of image and the emotion category is found using transfer learning in which the already trained CNN model is fed and hence the computational time is reduced [1]. Key Words: CNN, Emotion detection, facial features, local binary pattern (LBP), Pre-trained network, Transfer learning. 1. INTRODUCTION Emotion detection is one of the most important in software environment. Words may lie but face don’t. The only way to understand the emotions is by emotion detection which allows to know the mental or emotional state of the person. The emotion is detected by using transfer learning. Transfer learning is a Deep learning method where a particular trained model developed for a task is reused for some events of another task. Transfer learning [6] is a machine learning method where a model developed for a particular task is reused as the starting point for a model on a another task. It is recently extremely popular and attention grabbing in the field of deep learning as a result of this it makes the model to train advanced deep neural networks with relatively little data. With transfer learning, what has been learned in one task to boost generalization in another is tried to exploit primarily. The learning has the tendency to transfer the weights that a network has learned at task A to a replacement task B. Usually, there is a desired heap of data to train a neural network from scratch however there is no access to enough data and training takes more computational time. Transfer learning, is not a replacement methodology which is extremely specific to deep learning. There is a stern distinction between the traditional approach of building and training machine learning models, and employing a methodology following transfer learning principles. Traditional learning is totally different since it occurs purely based on specific tasks, datasets and training separate isolated models on them. No training knowledge is reserved which might be transferred from one model to another different model. In transfer learning, the knowledge can be swayed from previously trained models for training newer models and even tackle computational issues such as having less knowledge learned by the model for the newer task [6]. Deep learning systems and models are layered architectures that learn different new emotion features at various layers of the model. These layers are then finally connected to a last layer which is called fully connected layer. This efficient and layered architecture allows us to utilize a pre-trained network. This is used by breaking the fully connected layer without its final layer as a fixed feature extractor for emotion detection. The next step is to fine-tune the pre-trained model. This is a more involved technique, where not replacing the final layer for classification, but also selectively retrain some of the previous layers to add any new features while re-training. 2. LITERATURE SURVEY 2.1 Leveraging Unlabeled Data for Emotion Recognition with Enhanced Collaborative Semi- Supervised Learning Zixing Zhang [9] proposed the system that given a small set of labeled data and a large set of unlabeled data, the classifier at first trains the labeled data and recognizes the unlabeled data with confident samples selected via entropy (Semi- supervised learning). In each iteration, same number of samples per class is permitted. Then it recognizes and train the unlabeled data with the help of labeled data which is called self-learning. After a model finishes self-learning, training is done between the classifier namely SVM and RNN to learn the strength from each other and to avoid weakness. This process is called co-training (Collaborative semi- supervised learning). After co-training for each iterations, merging is done with the recognized confident data