Facial Expression Recognition Based On Active Learning

In daily life, we can not only express our feelings accurately and subtly by expression, but also get to know and analysis each other’s inner world according to their face. Because the expression is a basic way of human emotions, and plays an irreplaceable role in nonverbal communication.Entering the21st century, with the continuous development of artificial intelligence technology and computer technology, face detection, tracking and recognition technology constantly improve, facial expression recognition technology has been widely used in remote education, virtual reality, safe driving, clinical medicine, and intelligent monitoring and other fields. People have an growing demand for human-computer interaction, and they hope robots could have emotions and emotion like humans, it will change the relationship fundamentally between human and computer and provide better serve for humans. Human-machine interaction analyzes people’s facial expressions and their variations automatically through the computer to determine the people’s emotional changes and ideological activities, so human-computer interaction, intelligent nature have realized. Specialized study of facial expression recognition has become a hot spot in the field of pattern recognition and artificial intelligence. At present, the technology is still in investigate stage, further studies are needed to solve many problems.This paper has a research on the history and status quo of the facial expression recognition, and analysis and sums up the current related technologies used in facial expression recognition. Like other machine learning problems, facial expression recognition is also need certain number of labeled data which is used as the training data, which requires a lot of time and manpower on facial expressions annotation, and this is common to traditional classification algorithms. To solve the problem, a new way of thinking is put forward, in order to make up for a lack of labeled data,it is through the integrated use of labeled data and unlabeled data, joint training, so can reduce the manual workload greatly. Thus, the active learning method arises at the historic moment, which provides a new thought for facial expression recognition.In view of this, this paper’s main work is concentrated in two aspects. First, the work of this paper is mainly focus on facial expression feature extraction and classification, static facial expression image preprocessing, the facial expression feature extraction and expression classification are studied. Besides, the paper chooses the multiple faces library as the experimental data. Based on the research work of analysis about facial expression analysis and recognition at home and abroad, the paper adopts a method of automatic to extract key characteristics of facial expression, which reduces the complexity and improve the accuracy greatly. Second, this paper uses the method of active learning to recognize the facial expression, and the prototype system of expression recognition uses the method of human-computer interaction Based on the part of the labeled training set by computer to select the best sample by manual tagging and update the labeled training set, training in this iteration until the model is satisfied by users. Finally, by comparison with experiment the paper proved that active learning method achieve the goal of facial expression recognition better in reducing the artificial workload, using part of the labeled data.