Font Size: a A A

Dynamic Face Recognition System

Posted on:2001-09-11Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y G LiuFull Text:PDF
GTID:1118360185995617Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
This thesis discusses the Motion Face Recognition. The previous research mainly focuses on motionless face recognition In order to satisfy requirements of application, the motion face recognition become more necessary Combining with some practical applications, we put forward new ideas about the motion face recognition. The contributions of thesis are as followsIn order to assure the real time property, it should extract most face images from the motion face image sequence and keep some moment face images This procedure is named as the image segmentation It should speed up the recognition process and keep the inner relation information of motion face image sequence In this paper, according to the trait of Hidden Markov Models, we proposed two kinds of segmentation algorithms, which are based on the time serial and MPEG encodingThe face location is another problem of motion face recognition Based on classification theory in Topology, we put forward the face location based on Hausdorff distance, which has the better speed than other algorithms, and it can guarantee the recognition rate of successive Hidden Markov Model recognition procedureWe know that there is the motion relation in the motion face image sequence It is necessary to utilize the inner relation to improve the recognition rate in motion face recognition system We defined three kinds of related coefficients, which can describe the inner relation in the face image sequence and can modify the initial probability distribution to improve the recognition rateThe Hidden Markov Model is used as a motion face classifier. Based on the face classification, we can define statuses and status property images of Hidden Markov Models According to the relation between objects in the observation sequence and status property images, we can calculate the initial probability distribution Then Train and recognize the motion faces In additional, some disturbance effects should also be considered. We gave a modified recognition algorithm under the rotation condition, which can alleviate some disturbance...
Keywords/Search Tags:Motion Face Recognition, Image Segmentation, Face Location, Correlation Process, Hidden Markov Models
PDF Full Text Request
Related items