Research On Face Recognition Algorithm Based On Dictionary Learning

With the advent of the information age and the rapid development of technologies such as artificial intelligence and big data,people’s lifestyles have undergone tremendous changes accordingly.Face recognition technology is one of the most important technologies that affect people’s daily life.This technology has brought people into an era of "face-brushing".However,because face images are susceptible to different lighting conditions,different facial expressions,and some occlusions from foreign objects in the daily collection process,there is a certain degree of difficulty in the classification of face images.The face recognition algorithm based on sparse representation and dictionary learning can effectively solve the above problems and has good robustness.However,there is still room for improvement in dictionary learning algorithms applied to face recognition.Current dictionary learning algorithms treat dictionary atoms equally during the learning process,but in fact,different dictionary atoms have different contributions to classification,so different dictionary atoms need to be given different weight constraints.And when completing the collection of face sample images,we are often disturbed by the environment,which affects the quantity and quality of the collected face sample images,making the problem of face recognition a small sample size problem.In this master thesis,through in-depth research on related algorithms based on dictionary learning and sparse representation theory,two improved algorithms are proposed to solve the above problems,and the following research results have been mainly achieved:First,a face recognition algorithm based on mutual information and dictionary learning is proposed.The algorithm introduces mutual information coefficients in the dictionary learning process,sets different weight constraints according to the mutual information coefficients between different atomic vectors and label vectors,and then the objective function of the improved dictionary learning algorithm is designed,and when solving the objective function,the method of atomic loop iteration is used to solve the problem.This algorithm strengthens the connection between the dictionary and the sample label,improves the discriminativeness of the dictionary,and improves the recognition rate of the face.Secondly,a fusion dictionary learning algorithm based on sample diversity is proposed.In order to enhance the diversity of samples,the algorithm uses singular value decomposition to construct virtual samples,and performs dictionary learning on the virtual samples and the original samples at the same time to obtain the original dictionary and the virtual dictionary.In order to make the obtained dictionary complementary,weighted fusion is performed on the class index vector obtained from the original dictionary and the class index vector obtained from the virtual dictionary,and finally the classification result of the test sample image is obtained.This algorithm effectively alleviates the small sample size problem in the face recognition algorithm and improves the face recognition rate.Finally,this master thesis compares the proposed algorithm on a public face database,and compares the experimental results to verify that the proposed algorithm has a certain improvement.