The Research Of Face Recognition Under Illumination Variations

With the increasing development of the fields of anti-terror alert,military security,finance and e-commerce,identity verification and face recognition have played a significant role in more and more areas.Recognition accuracy and efficiency are not only two key problems of face recognition technology,but also mutually transformational.From the perspective of scientific research,face recognition accuracy is the core problem of face recognition,the level of recognition accuracy is often associated with application scenarios,the object of recognition,illumination,angle,background,facial expression,etc.On the one hand,after decades of development,the face recognition algorithm has been greatly improved,especially with the continuous advancement of machine learning and deep learning's face recognition algorithm,which brings new vitality and vigor to face recognition technology.But on the other hand,the expanding facial database will not only cause trouble for precise classification,but also give rise to heavier computational load for the machine learning algorithm,while the traditional algorithm for image quality still plays an important role in improving the recognition accuracy.This paper mainly focuses on the following aspects of the research and the main research contents and methods of the feature extraction algorithm based on the illumination and the recognition algorithm.First of all,This paper mainly focused on how to make a better illumination compensation for the uneven illumination.For the existing problems in the relative enhancement algorithm of face images under the influence of illumination,this paper proposes an improved adaptive Gamma correction algorithm based on multi threshold value OTSU(Otsu algorithm).This algorithm divides the facial illumination image with the OTSU segmentation algorithm,and stored the threshold values of different segments with binary method.It reduces the effect of illumination on face recognition accuracy.Then the image of the best quality is obtained by using the iterative algorithm based on Gamma correction algorithm.Secondly,in order to solve the problem of high false positive rate of human face,an adaptive face preprocessing method based on illumination type estimation is proposed.In addition,an improved algorithm is proposed for the local two value model of Local Binary Pattern(LBP)and Local Ternary Pattern(LTP).Firstly,the method classifies the captured facial images with OTSU multi threshold values segmentation and accumulative histogram correlation algorithm.And then it determines whether to carry out adaptive Gamma correction according to the image types.Thirdly,this method conducts DOG(Difference of Gaussian)filtering and equalization operations on the entered images of all types and finally outputs the processed facial images.Finally,the real-time face recognition system based on illumination compensation is designed and realized.First,it preprocesses the collected face images with the preprocessing method proposed in this paper.Based on the premise of real-time system,it estimates facial segments with the skin color detection algorithm,and then uses the human eyes calibration to accurately locate facial regions.Second,it conducts feature extraction on insensitive illumination with the improved LTP operator,and finally carries out classication and recognition with the nearest neighbor classification algorithm based on Euclidean distance.Related testing experiments verify that the system has certain practical value.Experiments show that all the proposed methods have good robustness in improving illumination,image quality and the recognition rate.