The Analysis On Heterogeneous Face Image By Using Super-resolution Reconstruction

Heterogenerous face image recogniton and analysis is a biometric recognition technology,which analyses the target's facial features in order to achieve the target identification according to the different quality images obtained by different methods and different sources.Sketch face image is a kind of common heterogeneous face image and palys an important role in criminal investigation,bank security,public security,law enforcement and other fields due to the limitations and particularity of the police to obtain information of the suspect.Because of the structural difference between the sketch and the face photo,traditional face recognition method can not be used directly.In order to overcome this drawback,face sketch recognition strategy based on local feature can effectively reduce the difficulties during recognition and analysis.However,this method is still unable to meet the actual observation needs of the police during law enforcement and can not do further analysis or comparison of the sketch in the police face database.To overcome the above shortcomings,this thesis carried out further work based on sketch face recognition and analysis via style transformation.The main work and innovations in this thesis are summarized as follows:1.A method for sketch face recognition based on super-resolution reconstruction is studied.First,eigenface algorithm is used to synthesize a pseudo-photo according to the input sketch.Then super-resolution reconstruction via sparse representation is executed on pseudo-photo.At last,Principal Component Analysis is used to recognize the pseudo-photo which is formed before the reconstruction and after.Because of the introduction of priori knowledge,super-resolution reconstruction by learnig can describe the image details better and enhance the image resolution significantly,hence improve the visual effect of the image greatly.Applying super-resolution reconstruction to sketch face recognition can not only improve the quality of image but also effectively increase the recogniton rate.The experimental results show that after super-resolution reconstruction the pseudo-photo can describe the facial details better,like eyes.What's more,because of the introduction of priori knowledge,the sketch face recognition rate is improved after reconstruction.2.There are two weaknesses in super-resolution based on global dictionary learning,one is the high time-consuming of dictionary training,while the other is the insufficient representation of different morphological structures of the image via global dictionary.To overcome the two shortcomings,an adaptive multi dictionary learning(AMDL)method is studied for super-resolution to reconstruct the pseudo-photo.Compared with global dictionary learning,this method introduce the clustering of the training samples according to the structure of the image,hence the time-consuming of dictionary training for each group is shoter than global dictionary training.In the process of image reconstruction,it is necessary to find the optimal dictionary for each image patch for the sparse decomposition and reconstruction,so this method can express the intrinsic structure information of image better.Two-dimensional marginal fisher analysis and learning to Hash are adopted to achieve face recognition and retrieval of the reconstructed pseudo-photo respectively.The experimental results show that super-resolution based on adaptive multi dictionary learning can effectively enhance the image quality and detail information of the pseudo-photo,hence can achieve ideal recognition rate and retrieval rate.Hash encoding and iterative quantization ensure the quick search of approximate face images efficiently.