Research On Facial Expression Recognition Based On Multi-scale LBP

Facial expression,which reflects the complex emotional,conveys emotional information.Facial expression recognition improves quality of life and life style,which widely used in security,artificial intelligence,emotional robots and other fields,and gradually become the focus of today's research.Aiming at the face recognition in video surveillance is susceptible to noise,illumination effect,the recognition rate is low.A facial expression recognition algorithm based on multiscale LBP features is proposed.It is multi scale histogram statistical method of ACILBP(Around Center Instable Local Binary Pattern)features and image features important regional Gabor with weights.The main research contents and innovations are as follows:First,to detection face in images,we use the method that Adaboost face detection algorithm based on HAAR features.Geometric normalization,histogram equalization,gray normalization for the detected face.On the basis of this,the paper defines two kinds of facial expressions: weighted Gabor features of the important region and ACILBP features of multi scale histogram statistical features.Weighted Gabor features of the important region: First,extract the Gabor features of three important regions of the eyes,nose and mouth are extracted,then extract the Gabor feature extraction of the whole face image.Finally,two different weights are assigned to each feature.And string them as weighted Gabor features of the important region.ACILBP features first extracted LBP features,then extract NLBP features.Finally,according to the two characteristics of the corresponding position value and absolute value of the corresponding position to determine the final ACILBP characteristic value.In the extraction of ACILBP features.A multi-scale histogram statistics method is proposed.The method is divided into two scales of facial expression image.At each scale,the ACILBP feature histogram is divided into blocks.This method can take into account the local texture information and the whole texture information.Second,connect the extracted two features,and input them to support vector machines and extreme learning machines for training and testing,Finally obtains the facial expression classification and the recognition resultFinally,the experiment selects the CK database and the JAFFE database.In order to verify the validity of the algorithm,compared with a variety of facial expression recognition methods in two databases,such as Orthogonal Combination Of Local Binary Patterns,Symmetric Local Graph Structure,Noise-resistant Local Binary Patterns,Completed Robust Local Binary Pattern,Local Mesh Patterns,Joint Local Binary Patterns,Center-symmetric Local Binary Pattern,Multi-scale Block Local Binary Pattern,Extended Local Binary Pattern.The experimental results show that this algorithm has higher recognition rate and robustness.