Research On Facial Expression Classification Algorithms

In recent years,with the development of science and technology,more and more attention has been paid to pattern recognition and machine vision in which facial expression recognition is one of the most important research directions.For example,it has broad application in safety driving monitoring,remote interactive teaching,game entertainment and other fields.However,in view of the disadvantages of slow recognition and low accuracy,which cannot meet the requirements of fast recognition and high accuracy.It has great significance to improve recognition rate and processing time for facial expression recognition.In this dissertation,process of facial expression recognition is analyzed.After much deliberation,we select improved Local Binary Patterns(LBP)for features extraction and improved Extreme Learning Machine(ELM)for classifying.Following work is completed:Firstly,an expression collection device is designed.The advantage of the device is that it can move flexibly in order to reduce the influence of illumination on image recognition.At the same time,it can communicate with external devices such as computers in real time,making full use of network resources.It has certain application prospects and commercial value in the long run.Secondly,the expression images obtained are transformed from time domain to logarithm-Laplace domain,in which images preprocessing are carried out and the defects such as uneven illumination and shadows are corrected to further optimize the facial images.Next,in order to improve the speed of image feature extraction,double local binary patterns(DLBP)are used to divide the image into several parts;as a result,the processing speed can be increased by multiple orders,which can meet the needs of practical application.Also to enhance the effectiveness of the extracted features,Taylor expansion(TE)is applied to image processing.Double local binary patterns-Taylor expansion(DLBP-TE)algorithm is combined to enhance the effect of neighborhood pixels on the central pixels.On the basis of these steps,more effective feature information is extracted.Finally,the effectiveness of the proposed method is tested by comparing experimental data.Thirdly,bat algorithm(BA)is applied to improve ELM algorithm and the new algorithm is named BA-ELM.In the new algorithm,BA is used to optimize the pre-allocation of hidden layer nodes,improve the random selection of hidden layer parameters and other shortcomings in order to improve the classification of image features.Finally,we test the effectiveness of the algorithm through experiments.Finally,the improved DLBP-TE feature extraction algorithm and BA-ELM classification algorithm are experimentally performed by images which come from collecting device and facial expression database.Experiments show that improved DLBP-TE feature extractionalgorithm improves the processing speed and the validity of features.Similarly,the improved BA-ELM classification algorithm has better real-time performance.