The Research Of Face Recognition Under Limited Conditions

The biometrics play an important role in identity recognition technology due to its uniqueness.Because of its convenience,security,extensibility,uniqueness and other advantages,face recognition technology stands out among various biometric technologies which therefore attracts lots of attention from both academic and industry.In recent years,study and application of face recognition technology have made some great achievements.Now it is widely used in various situations where identity verification is indispensable.But there is no doubt that face recognition technology has its defects as well.The change of environment,resulting from illumination,pose,facial expressions and other related factors,is the biggest one.All of these changes have a great influence on the face recognition accuracy,making it only possible under the specific scenes.In order to deal with more complex scenarios,we have conducted a great deal of research on face recognition under limited conditions,especially the following two aspects.One is the input face image resolution which is so low that the face image is of low quality,due to the long distance,poor light,or small memory device.The other is that the input face resolution is too low or the target person avoids being photographed by evading the surveillance camera deliberately.Then little effective facial feature information can be extracted.Aiming at the problem of face recognition with low resolution,this paper proposes an image super resolution method for the two generation ID face photo based on random sampling and local constraints.The motivation comes from the fact that the identity card is used on many occasions where verifying identity is necessary.But the resolution of the face photo obtained from the ID card is very low.If it is used to identify directly,the correct rate will be very low.To improve the reliability of face recognition by raising the effective face information,we use the technology of super resolution to improve the resolution of the input face image and raise the effective face information.If the input face image is unable to be reconstructed by super resolution due to the too low resolution or the surveillance device cannot get the target face image,it is usually called the face recognition under limited resources.With the help of simulated portrait,we can get the input face sketch.But sketch and photo have great differences in the expression of texture information and features.If we use sketch to recognize,we cannot achieve the purpose of retrieval.Therefore,this paper propose a face synthesis method based on back projection of GAN.It solves the problem of heterogeneous image recognition by algorithm which transform the sketch and photo into the same feature space.Finally,we carried out experiments on four public face database,and the result shows the effectiveness of our method.Compared with the existing methods,ours is the best.Based on the above research,we set up "intelligent surveillance face recognition system".The software uses Microsoft Visual Studio 2013 as the development tool,with Qt5.4 as the development framework of software interface and Open CV as the Visual library for image processing.The software has three main functions: real-time identification,human-card verification and sketch-photo synthesis.They realize the face recognition under the normal environment and under limited resources respectively.