Facial Expression Recognition Research

In recent years, digital equipments like digital camera, mobile phones with high pixel get more and more popularity so that everyone has one for each one. Even in daily life, high quality of image acquisition is easy to achieve. Face image with high quality makes our job easier for facial expression recognition. Lower noise, more obvious characteristics reduce the workload and difficulty of facial expression recognition. Advances in computer technology,of course, for us to study expression recognition, plays a huge role in promoting as well. Facial expression recognition gets much attention in today’s development in science and technology field. Great parts of the high-end phones are equipped with smiling face capture function. In addition, the expression as a social life, one of the most important performance characteristics of interpersonal communication, natural contains a large amount of information. It seems that our research work for facial expression recognition is significant. Expression presents the change on the person face facial structure and we know that the human face is a so complex system that there are not two exactly same face in the world. It reveals the difficulty of facial expression recognition naturally. For facial expression recognition work, this article refers to the basis of predecessors’ research results, combined with my own thinking and experiment, gets the information of the preliminary analysis. The main work concludes:a) Analyze the research background and project prospect of facial expression recognition at home and abroad. Introduced the research history, current research status in the field of facial expression recognition. Get whole knowledge of facial expression recognition in every part. Distinguish the key work for facial expression recognition.b) Based on existing expression library JAFFE facial expression image, preprocess face image, including gray processing,binarization processing, histogram equalization, image rotation,image cropping, etc. Get ready for the next step in analyzing image data, and improve the recognition speed and efficiency. Preprocessing expression image has great influence on the feature extraction and classification so that we’d better not do our identification on the original data directly. The characteristics should be preserved which can highlight the difference in every expression. At the same time eliminate irrelevant information as far as possible, such as illumination, posture and other factors.c) Introduce a greatly important feature extraction algorithm in the field of identification areas, principal component analysis (PCA). Analyze the mathematical principle of PCA in detail, prove that the principal component analysis based on K-L transform can achieve the purpose of reducing dimension in math. Get Eigenface expression and get the projection of preprocessed expression image. After this step, the processed image data can describe the main features of human faces, and data’s redundancy is small, which facilitates our classification, so that we can easily obtains higher efficiency.d) Relatively simple classification algorithm, K neighbor and nuclear K neighbor, is introduced. The core idea in Neighbor classification algorithm is to measure the distance between the training samples and test samples to achieve the purpose of classification and recognition. Algorithm is simple, and has clear classification principle. The larger interval between each class in sample collection, the better classification effect will achieve. Study its application in facial expression recognition, and analyze the experimental effect.e) Introduce an important method in the field of pattern recognition-the method based on support vector machine. Use SVM method for facial expression recognition. Introduce the basic theory of SVM, statistical learning theory and the optimal classification plane theory and generalized optimal classification plane theory; SVM is to solve binary classification problems, when meeting classification samples can’t divided into just two classes directly, put forward an improved algorithm based on SVM+K neighbor, to achieve further accurate classification.