Research On Face Detection And Facial Expression Recognition Under Severe Illumination Conditions

Facial expression is a biological feature inherently reflecting human emotional changes.Seven basic facial expressions based on the classification of human face activity unit coding system have the same emotional meanings among different individuals.This classification method makes facial expression classification possible with scientific criteria and quantitative analysis.With the development of computer vision technology,the classification of facial expressions can be automated by machine recognition and neural network technology,so that computer can imitate human eyes to realize the correct recognition of facial expressions.In the research of expression recognition,three basic processes have been gradually formed: face detection,feature processing and expression classification.Among them,the purpose of face detection is to solve the problem of face location and to determine specific regions for feature processing.Feature processing mainly includes feature extraction,dimension reduction and classification,in which feature extraction can use different kernel functions to extract suitable feature types according to the needs of classifier design.Facial expression classification is the key step to realize the correct recognition of facial expressions.In the research process,there are many problems that affect the accuracy,such as illumination,posture,scale and so on.In this paper,image preprocessing,feature extraction and expression classification are studied.The main research work is as follows:1.The problem of face detection under harsh illumination.Photographing or videotaping in the daily environment often can not avoid the influence of illumination factors.Especially in the night,light and occlusion environment,it is very easy to lead to the situation that the face can not be effectively detected,which will affect the efficiency and even cause errors in security inspection,monitoring and night shooting.In order to reduce or even avoid the adverse effects caused by illumination factors,this study conducted a study of illumination normalization for face detection under harsh illumination conditions.This research mainly improves the histogram equalization method.The basic idea is that the equalization algorithm combines with threshold segmentation algorithm to process the image hierarchically.After image mosaic and filtering,it can get a good visualization effect,and then realize face detection smoothly.2.Facial expression recognition based on machine learning is an important research direction of face detection.Based on face detection,this study uses machine learning method to do the following research in image processing,feature extraction and feature classification stages:In the image processing stage,in the sample segmentation,this study proposes a method to segment irregular face images based on Ensemble of Regression Tress(ERT)algorithm to determine the feature points of the core region of human face;In the stage of feature extraction and classification,LBP convolution kernels with rotation invariance and illumination invariance are used to process face features at different scales,and feature values of different scales are extracted from the same region of different faces on the basis of the same expression,which provides training data for Adaboost adaptive enhancement algorithm.Texture features extracted by LBP have strong expression attributes,which exclude the influence of color,brightness and rotation angle from the data level.At the same time,because circle LBP can adjust the radius and number of sampling points,it can provide a lot of training data for Adaboost.