Research And Implementation Of Facial Expression Recognition Algorithm Based On Gabor Wavelet

For a long time the human-computer interaction is the focus of future scientificand technological development. Make the machine a good understanding of humanmental activity, will dramatically change the way human-computer interaction.Understanding human mental activity need to identify facial expression, whichcontains a wealth of emotional information. Therefore, facial expression recognitiongets more and more attention. And become the focus of research of affectivecomputing. This paper studies the key technologies of the whole recognition process,including face detection and localization, expression feature extraction, classificationand face recognition. Expression feature extraction is particularly important aspect.Research works and results are as follows:(1) Do pretreatment for facial expression image. Firstly, study the human facedetection method based on Adaboost algorithm. After the detection, use the integralprojection method to locate the human eye. Then use the eye coordinates as the datumto cut the expression regions from the original image. Do processing like sizenormalization and illumination equalization, Finally, get a uniform standardexpression image, to prepare for the follow-up work.(2)Using Gabor wavelet to extract the original expression feature. Based onthe Gabor wavelet theory, use Gabor wavelet to extract features of expression image.Gabor wavelet transform is actually a set of filters generated by scaling and rotationof the Gabor kernel function. Compare the recognition rate affected bydifferent scales and directions through the experimental. Choose the most suitableparameters to construct the multi direction and multi scale Gabor filter group toextract the texture information of expression image.(3)To reduce the dimension of expression feature. Because the feature extractedby Gabor wavelet has a very high dimension, which is not conducive tothe classification. And it has information redundancy. Therefore there is a need toreduce the dimensionality of Gabor features to obtain the feature which is themost beneficial to classification. This paper studies two kinds of algorithms: the principal component analysis (PCA) and local linear embedded (LLE).Experiments show that using LLE to reduce nonlinear data can get a better result. Souse LLE algorithm to reduce the dimension of feature. In order to make all kindsof facial expression feature has better differentiation. This paper presents a methodthat combine Gabor Wavelet with LLE and FLD algorithm to extract expressionfeature. Finally, compared this algorithm with Gabor wavelet combined withPCA and FLD algorithm, proving the feasibility and effectiveness of the proposedalgorithm.(4) To classify and recognize the features of expression. First, the support vectormachine (SVM) is used to classify and identify the expression, and get a betterrecognition result. In order to improve the recognition more, fuse the SVM anddistance classifier to classify expressions. Fusion method based on fuzzy integralmethod. By identifying the fusion rate comparison before and after, to verify thefusion method can increase the recognition rate. All algorithms in this paperare carried out in the JAFFE expression library. The experiment was divided into theexpression recognition related to people and not related to people.