Research Of Face Recognition Method Based On Fusion

Machine understanding of human language has become a key research topic sincethe 1970s. In our daily life, faces clearly communicate abundant and exquisite feelingsand psychological information, therefore, facial perception computing, especially facerecognition has gradually turned into one of the hot research subjects in resent years.The another important reason is that face recognition is of enormous and potentialvalue in the field of economy, security, social security, crime, military affairs and so on,especially in identity validation or recognition environment. Face identificationtechnology not only avoids users being involved too much but also combines theadvantages of contactless data acquisition and convenient hiding without any damageto the users, so it is called the most promising method of verifying the useridentification in the 21st century. The research of face recognition is helpful for otherobject recognition, thus, face recognition is both one of the important tasks of the basicresearch and a subject with important theoretical research value.Based on the basic theory and key techniques in face detection and recognition,the paper lays emphasis on face detection in color images, image enhancement, featureextraction, multiple classifiers for face recognition, ear recognition, profile facerecognition, fusion scheme on the basis of face and ear recognition, multi-agent facerecognizing structure model(MAFRSM) for 3D face recognition etc.The contributions in this paper are described below.(1) Aiming at the major difficulties of face detection algorithms, the paperpresents a new method of face image enhancement. Experimental results show themethod may convert brighter or dimmer face images into clearer images, and it canobviously improve the accuracy of face detection and recognition.(2) The paper proposes an algorithm based on the color of skin and binocularinformation for color images in complicated background. First of all, this algorithmroughly detects face-like regions for input images according to the color of skin, rejectsthe areas that the faces do not exist in, gets possible face-like regions at the same time,so, it can shorten detection time. Experiments prove this algorithm has betterrobustness and can be efficiently used in the conditions of multi-faces, different facesizes, various expressions and complex background. It has stronger adaptability forfaces with certain rotation and side slip angles, accordingly, it can effectively reduceerror detection rate and achieve better detection effects.(3) In this paper, a recognition method for multiple classifiers is proposed, whichcombines an improved eigenface method with support vector machine(SVM). Thecombining classifiers can make use of high recognition rate for SVM and high speedfor distance classification. The distance classifier may classify the input images andgive the final results when the rejecting rule is satisfied. Otherwise, these images aredelivered to SVM for further classification. Experiment data show that the fusion ofmultiple classifiers for face recognition has higher efficiency, accuracy of recognitionand lower rate of error recognition.(4) Both ear recognition and profile face recognition are new subjects in the fieldof biometric recognition. Few achievements have been made in this area at home andabroad. The paper initially explores and attempts ear and profile face recognition, andpresents a novel method for ear recognition, which gets the feature extraction based ongeometric technique. Experimental results demonstrate that the approach has higherperformance for clear ear images without illumination variety. The present methodsbased on geometric features of human face locate profile face features withoutconsidering their neighborhood information. The accuracy rate for profile facerecognition is lower when using the actual minimum Euclidian distance orsegmentation curvature method of profile face contour line. In view of the deficiencies,we detect main edge contour for profile face images by adopting continuous wavelettransform, extract the principal components using principal component analysis(PCA),and recognize profile face through support vector machine(SVM).(5) Non-invasive characteristic of ear and profile face recognition contrary toother biometric recognition, unique ear features and ubiety about face and ear of 3Dhuman head ensure the feasibility for fusing face and ear recognition. The paperproposes a new approach in decision fusion, the method uses less data than otherfusion and has a faster recognition rate. The fusion based on face and ear recognition isa new idea to improve recognition rate of identity recognition and believable degree, inaddition, it is a meaningful attempt to explore a novel method of biometric recognition.Eyes are parallel to the recognition process of facial patterns, but actual computerarchitecture is serial. At present, multi-biometrics authentication systems have not auniform frame construction. The paper describes the process of 3D face recognitionusing recent multi-agent system theory in artificial intelligence for the first time,proposes MAFRSM for 3D face recognition which combines concurrency researchresults and the specific characteristics of 3D face recognition. The task of 3D facerecognition in various aspects can be accomplished by the structure model, and amajority-decision algorithm based on multi-agent is presented.