Analysis And Research On Algorithm Of Facial Expression Recognition Based On Wavelet Transform

Facial expression recognition is extract feature from facial expression information, according to human's epistemology,understanding and mode of thinking to understand and classify, and then based on priori knowledge in the field of human emotions, allowing the computer to simulate human minds about thinking and reasoning, by analyzing the information of human face to understand human emotions. Facial expression recognition is a challenging cross-subject, involving research field of biometrics, pattern recognition, image processing, machine vision, physiology, psychology and other. It is an important component of affective computing and human-computer interaction.It is also a research hotspot in the field of pattern recognition and artificial intelligence. So facial expression recognition has high academic value and a huge potential in market applications.In recent years, the joint efforts of major research institutes,well-known universities and IT companies both at home and abroad, the technology of facial expression recognition developed rapidly. This paper is based on previous research in related fields, through refering to and analyzing large amounts of academic papers and literature of facial expression recognition at home and abroad on, efforts to study the key techniques and methods od facial expression feature extraction and recognition. Purpose of this paper is to seek algorithms with achieving relatively simple,speed of feature extraction and recognition, high recognition rate and is suitable for facial expression recognition. In this paper, as follows:1. From human emotion, this paper comprehensive overview the background and significance, the purpose and applications, research status and development trend of facial expression. Then, from the technical aspects of facial expression recognition,sorting the main techniques on face detection,pretreatment, feature extraction and classification of facial expression. Finally, introduces several major facial expression database in the current world.2. This paper proposes an efficient facial expression recognition method by combining the Two-dimensional Discrete Wavelet Transform (DWT) with PCA/Fisher Linear Discriminant Analysis(FLD). DWT has good properties of time-frequency localization and energy concentration, which make the information of decomposed subgraphs contains most of the original image information. Therefore, this article using DWT decompose the original expression image on the basis of DWT with this excellent features, then extract the wavelet coefficients as the feature. Based on academic theory of facial expression recognition, we know that valid information is mainly reflected in the changes of the eyebrows, eyes and mouth, followed are reflected in the relationship between each other. These details vary mainly produced by changes in the image edge depending on the knowledge of image processing, so getting the details information of the image is helpful to face expression recognition. The high frequency transformed by DWT is characterized image texture details. Therefore, this paper extract the low-frequency information while retaining some of the high-frequency information after wavelet decomposition. For the small sample size problem in facial expression recognition, this paper uses principal component analysis to project samples in the high dimensional space to a low dimensional space, which ensuring that the matrix of between-class scatter is non-singular, while the removal the correlation between the data. Then,using FLD to extract feature from these non-singular matrix. Finally, classification with the nearest neighbor. In order to verify the validity and robustness this algorithm, this paper did two sets of experiments, one was people-related and other was human-unrelated.The Japanese JAFFE facial expression database was used in the experiments, and database contained six facial expressions (anger, disgust, fear, happy, Sadness, surprise), experiments results show that this algorithm had a certain ability of discriminant expression, and feature extraction speed was faster nearly 12 times then without DWT...