Research And Application Base On Multi-scale Images Algorithm In Low-resolution Face Recognition

Since the 1960 s,face recognition has first appeared in the field of digital image processing.It has been a half-century of development.Recognition research has also changed from a single simulation environment to adapting to a variety of complex environments.It also ranges from simple frontal images with sufficient illumination and no change to a variety of facial expression changes,older age spans,and multiple occluded faces.Although it has been developed to date,many existing face recognition algorithms can achieve excellent recognition results in a specific constrained environment,but if applied to real life,strip the set specific environment and face the low resolution in real life Face recognition performance is often not ideal.The research object of this paper is the non-traditional face image in this situation.There are mainly two existing methods for solving low-resolution face recognition.The first is super-resolution enhancement,which aims to match the high-resolution dimensions by enhancing low-resolution face images.The second is to simply and directly extract features from low-resolution face images to obtain effective facial information.There are many mature high-resolution feature extraction methods,so based on this foundation,this paper uses the methods that have achieved good results to analyze and improve,and combine the multi-scale space theory to apply them to low-resolution face images.The main work of this paper is as follows:(1)Research on low-resolution face recognition algorithm based on fusion of Gaussian pyramid featuresIn order to further overcome the difficulties of recognition and verification of lowresolution face images,this paper combines the multi-scale space theory and combines a variety of corresponding feature extraction methods with Gaussian image pyramids,and proposes a fusion part based on low-resolution face images.And global information facial feature extraction method.This method first builds a Gaussian pyramid by decomposing the face image to achieve the purpose of constructing a multi-scale space;secondly,different operators are used to obtain the characteristic spectrum of each layer image in the tower,and the local binary mode operator is used at the bottom layer,and the middle The layer uses a set of Gabor filtering,the top layer uses HOG to extract edge information to obtain the HOG feature spectrum,and the three-layer feature spectrum fusion forms a feature fusion pyramid;finally the obtained feature vector is decided and fused through the classifier to complete the entire low-resolution face image Identification process.(2)Research on low-resolution face recognition algorithm based on multi-scale featurelevel fusionBased on the existing research,in order to further improve the performance of lowresolution face recognition,a low-resolution face recognition algorithm based on multi-scale feature-level fusion is proposed.First,two feature extraction methods are used in parallel for low-quality face images,namely,multi-scale binary mode to obtain global information and GHOG with both local information and edge texture information;then,the obtained face feature map is passed through two-dimensional The inverse transformation of empirical mode decomposition obtains a feature-level fusion image with complementary facial features.Finally,put the obtained image feature spectrum into multiple classifiers for training to obtain the best recognition effect,and use the multi-scale voting method to complete the decision-level fusion to obtain the final classification result.In this paper,a variety of simulation experiments are designed in AR,CMU-PIE,YALE,ORL,FERET and other face databases,as well as comparison experiments with other targeted algorithms.Experiments have proved that the algorithm proposed in this paper is more accurate than the recognition accuracy of low-resolution face data.