Research Of Face Recognition Based On Doctionary Learning

With the extensive application of face recognition technology,face recognition has become one of the most important research fields of pattern recognition and computer vision.In recent years,the methods based on dictionary learning algorithm has become one of the most popular researches,and have been successfully applied in the real life.By learning the most essential features of the training samples,the test samples can be better represented by the training samples.However,due to variations of the facial expression,posture,and illumination,the performance of the face recognition method in practical application has been limited.At the same time,over-high dimensional face image leads to great computational cost,which harms its wide application.The appearance of noise affects the learning of the feature,so it is not good enough to merely use dictionary under the complex noisy environment.This thesis puts forward a method that connects dictionary learning and sparse expression through the study of statistical independence and sparse.It not only avoids the problem of noise in the process of learning the dictionary,but also improves the recognition accuracy.At present,some algorithms ignore their generalization ability.Therefore,most of experiments in this thesis focus on blind source separation and recognition tasks,which validate effectiveness of the proposed method,in the presence of noise with both signal data and face image data.The highlights of the two proposed methods are described as follows:1.This thesis presents a method of extracting independent component analysis from sparse mixture.In this method,the objective function is solved by alternating iteration method,which uses the independent constraint condition of negative entropy.2.This thesis presents a method,which combined with the independent dictionary,to improve the regularization term.Meanwhile,the sparsity constraint can improve the robustness of the method in face recognition,and reduce the expression of interference and noise.At the same time,the gradient algorithm is used to optimize the objective function.The optimization algorithm for the derivation of the objective function,greatly reduces the computational complexity of the model.In addition,in order to verify the accuracy and effectiveness of the model,experiments are carried out in the field of signal processing and face recognition.