Font Size: a A A

Facial Expression Recognition Algorithm Based On Local Features

Posted on:2022-06-13Degree:MasterType:Thesis
Country:ChinaCandidate:D Y GaoFull Text:PDF
GTID:2518306557971349Subject:Electronics and Communications Engineering
Abstract/Summary:PDF Full Text Request
Since the 21 st century,facial expression recognition technology has been developing continuously and has achieved great success in many fields,such as intelligent voice robot in the field of game leisure,medical monitoring system in the field of medical treatment,automobile auxiliary system in the field of transportation,etc.Facial expression recognition technology has greatly improved people's quality of life.Although the current facial expression recognition technology has made great achievements,it still faces many challenges,such as the complexity of facial expression,the diversity of environmental factors,foreign body shielding and so on.Therefore,how to design a feature extraction method with high recognition rate,fast recognition speed and high robustness is still a challenging problem.This paper mainly studies facial expression recognition based on different feature extraction methods.The specific research contents are as follows:(1)Analysis of LBP features and extension,and summarizes the advantages and shortcomings of various LBP features,puts forward an improved LBP feature?In view of ignoring the relationship between each pixel in the neighborhood when calculating the traditional LBP feature,this feature proposed to use the average value of the difference between the neighborhood pixel and the center point as the threshold value to increase the relationship between the neighborhood pixels? At the same time,according to the FACS theory,the importance of eyes and mouth to facial expressions was highlighted.The image was divided into blocks first,and then the concept of weight was introduced to add certain weight to the blocks belonging to eyes and mouth.The two improvements were combined with the advantages of basic LBP features to form the block weighted equivalent threshold circular LBP feature.In order to verify the effectiveness of this improvement,the two improvement points were compared with the basic LBP characteristics.Experimental results show that the proposed improvement points can effectively improve the LBP feature recognition rate?(2)Considering that HOG feature does not have scale invariance and HOG feature of different scales can express the same object differently,it is proposed that multi-scale HOG feature can be used to extract facial expression image features and retain more image information.At the same time,considering the risk that multi-scale HOG feature has too high feature dimension,PCA algorithm is introduced to reduce feature dimension while retaining the main feature information as much as possible.Combined with HOG feature and PCA algorithm of multiple scales,the recognition effect of two multi-scale fusion methods was explored.The experimental results on CK+ show that the multi-scale HOG feature obtained by the two fusion methods is faster and has higher recognition rate than any single scale HOG feature.(3)A single feature is extracted from only a certain point of view.For example,LBP feature lacks the image gradient information of HOG feature,so it cannot fully express all the information of the image by relying only on a single feature of the expression image.Based on features fusion,this paper proposes a multi-feature fusion algorithm for facial expression feature extraction.The improved LBP feature is fused with multi-scale HOG feature,and the PCA algorithm is used to reduce the dimension of the fusion feature.The fusion feature combines the image information of the two features and contains more abundant image texture information.The control experiment on JAFFE library shows that the performance of the fusion feature is better than that of the single feature.
Keywords/Search Tags:facial expression recognition, local binary pattern, histogram of oriented gradient, principal component analysis, feature fusion
PDF Full Text Request
Related items