Facial Expression Recognition Based On Lifting Wavelet And FLD

Expression perceiving by computer, according to the perceiving habits of human, is becoming an important method for HCI. Human facial expression recognition is an interdisciplinary of psychology, bio-informatics, image processing, computer vision, pattern recognition and machine learning, which has been paid more and more attention from its emergence in1980s. Artificial emotion is part of artificial intelligence, and it’s an attractive area which uses artificial method and technology to imitate, extend and expand human emotion. As robots used in more and more fields, people need robots have the characteristics of intelligence, autonomy and interaction. Artificial emotion makes it possible. Facial expression recognition based on computer vision is an important part in artificial emotion research.The facial expression recognition system mainly contains three parts:face detection, feature extraction and classification. Among them, feature extraction is the most important part which directly affects the final recognition results.There are several factors that affect the expression recognition rate. One of them is external environment, such as, lighting, angle, obstructions and position of the face. Another is complex and volatile facial expression. The last one is that the control over the human expression is limited. In order to reduce the recognition time and improve the recognition rate, many experts have made a lot of innovations and pur forward many new algorithms.Some classical algorithms used in facial recognition have been imported into this field and have good results, like Principal Component Analysis (PCA), Independent Component Analysis (ICA), Fisher Linear Discriminant (FLD).The lifting wavelet has many advantages. Wavelet Transform can be implemented in the spatial domain, so it is helpful for the feature extraction of expression details. Because of the Simple structure and Low amount of computation, the lifting wavelet needs less storage space. We use the whole character made up by the LF and HF components as the face feature which contains the main expression feature of the expression image.The Fisher Linear Discriminant (FLD) used is improved to utilize the principal component analysis (PCA) as a pre-processing step aiming to reduce the dimensionality of the vector space.The main steps of the new algorithm are as follows:Firstly, the facial expression images are processed by lifting wavelet. Then the images obtained contain the main expression feature of the expression image.Secondly, the improved Fisher linear discriminant (FLD) is used to extract features form the images we got.Finally, the K-neighbor method is used for classification.We use a new algorithm based on lifting wavelet and FLD for feature extraction. Because of its local characteristic in the spatial domain, the recognition rate increased by0.4percentage points to Gabor wavelet feature extraction. It proves the lifting wavelet is more useful to improve the correct rate of facial expression recognition. Also the new algorithm can shorten the span3times. Feature extraction based on lifting wavelet and FLD can not only improve the recognition rate, but also shorten time. It comes to the conclusion that the realization process is easier and more efficient.