Facial Expression Recognition Based On Local Features

With the rapid development of science and technology,more and more friendly human-computer interaction experience is highly expected.Facial expression,as part of the biometrics,displays the intricate and subtle inner world of people and achieves a more humane mode of human-computer interaction.In addition,facial expression recognition technology has been widely used in intelligent safety monitoring,medical surveillance and criminal investigation and other fields.Therefore,how to efficiently extract expression features and accurately identify them is an urgent problem to be solved in this field.In this paper,we mainly study the method of face expression recognition based on different features fusion and feature preprocessing to improve the recognition rate of face expression algorithm.The main works of this paper are as follows:1.The difference between the facial image and the corresponding neutral image is the key of facial expression recognition.The projection error between each sub-block of the expression image and the corresponding neutral image is considered.The larger the projection error is,the richer the facial expression information is.Therefore,this paper proposes a method using the combination of LBP features and projection errors to achieve the effect of feature stretching and feature differentiation enhancement,and improves the expression recognition rate.On this basis,this paper also compares the performance of facial expression recognition with features extracted by standard LBP algorithm and uniform LBP algorithm.The experimental results show that the uniform LBP feature can reduce the dimension and improve the recognition rate.2.Based on the rich information included in multi-scale features,this paper improves the existed single-scale HOG features,and the HOG feature fusion of two-scale and three-scale has been proposed.It is showed that the fusion of multi-scale HOG feature is superior to the single one.3.Due to the disadvantage that a single feature can not accurately represent all the information of an image,this paper proposes a facial expression recognition method based on weighted feature fusion,which merges the extracted uniform LBP features and HOG features.At the same time,the feature extraction is not performed directly for the whole image of the facial expression,but adopts the regions of the eyes and mouth which change more obviously with different expression.The feature dimension is reduced.Under the allowable complexity of the algorithm,experiments on JAFFE and CK + are performed.Experimental results show that the proposed method is superior to the one obtained by using a single feature.