Face Feature Extraction Algorithm Based On Multi-pattern And Dictionary Learning

Feature extraction is a key step in face recognition.Among the current mainstream feature extraction algorithms,some are based on single images and some are based on multiple images.In this paper,detailed introduction is given to feature extraction algorithms based on single images,and a study was also carried out on feature extraction algorithms based on multiple images.In order to solve the problem bad performance while using a feature matching method,studies the basic theories of face recognition approaches based on single images and multiple images,through analysis of the existing theories and algorithms,the main content of this paper is as follows.PCA with invariant normalization feature extraction matching algorithm.When plane transposition is conducted on training images,this algorithm can accurately represent the inherent information structures of human face.With this algorithm,images can be normalized and extracted alternately.By obtaining the minimum mean square error between a normalized image and a reconstructed image,the optimal feature space is created,enabling image normalization and image representation to promote each other,so as to get a high recognition rate for face images.Multi-direction and multi-level double intersect pattern feature extraction algorithm.The solution uses the first-order derivative of Gaussian operators to reduce the influences of illumination on recognition.The multi-direction and multi-level dual-intersection mode is a new solution used to represent face descriptors.It mainly uses the inherent structural information in human faces and obtains feature information at the overall level and detailed levels respectively,overall level and detailed level promote each other.Because of this,the solution has a strong discrimination ability,as well as robustness for changes in illuminations and expressions.Dictionary learning and joint feature based on image set algorithm.Both the test set and the training set contain images from different attitudes,illuminations,and facial expressions.And meanwhile,image sets contain many low-resolution images.By learning feature projection matrices and structured dictionaries simultaneously,this method extracts more discriminating information,and retains details information of image reconstruction,which produces higher accuracy compared with other dictionary learning algorithms.