Research On Understanding Of Facial Expressions Information In Sign Language And Its Implementation

Sign language is the language which is used by the deaf, it is a stable expression system which is composed of gestures, facial expressions and head gestures. Relative to a wide range of body language of common person, sign language have more regularity. Recognition of sign language can help hearing people to have a better understanding of the psychological state of the people using sign language and promote the harmonious development of human-computer interaction.The experiments of sign language recognition show that when rule out the facial expression that only identify the gestures, people can not understand more than 60%. Psychologist Mehrabian pointed out that in people’s daily communication, the information conveyed by facial expression can be reached 55%. Facial expressions contain a great lot of sensibility information, so it has the important research value to identify the facial expression information in sign language video. In this paper, we focus on the feature extraction and dimensionality reduction of facial expression information recognition to improve the recognition accuracy. Follows show the specific research work of this paper:1. This paper focus on studying the feature extraction of facial image sequences. Since the expressions are diversity and the changes of expression may only lead to local image texture changes, the features extracted from image sequences should be able to contain this information. Gabor is an effective appearance feature extraction method, but the Gabor filters of adjacent direction or scale are very similar, so local Gabor filters are used to extract the features. Whereas Local Binary Pattern(LBP) is a very effective texture description operatiors, the feature extraction method called Histogram Sequences of Local Gabor Binary Pattern(HSLGBP) is researched.2. Since the hign dimension of texture features, dimensionality reduction on high dimensional features is researched. The essence of principal component analysis(PCA) is to extract the main features of original features (principal components) to reduce features redundancy while maintaining the most information of the original features, PCA do not consider the distinction between different classes of data. Linear Discriminant Analysis(LDA) is to maximize the ratio of between-class scatter and within-class scatter to choose the optimal projection direction. The projection features has more distinction. This paper focus on the feature dimension reduction method of combination of PCA and LDA, finally recognize the facial expression information using HMM.