Face Representation And Classification Based On Reconstruction Representation

In recent years,people pay more and more attention to information security.Due to the strong stability and uniqueness,biological characteristics has become a popular research field of identity recognition.Face recognition as an important branch of biometric features recognition,has been widely concerned.In the past few decades,many classical algorithms have been proposed,such as principal component analysis and fisher discriminant analysis.In addition,with the rapid development of compression perception,sparse representation has been widely used in pattern recognition because of its excellent recognition result.When the training sample images are ideal,sparse representation algorithm can achieve outstanding results under the illumination variations,occlusions and noises.However,in the actual environment,image acquisition is often affected by various factors(such as illumination,pose,occlusion,etc.).How to reduce the interference factor on the algorithm is the main focus of current research.Therefore,we have a deep research on face images representation and classification based on reconstruction representation.1.Face Recognition Based on Non-negative Sparse Low-rank Representation ClassificationThis paper proposes a non-negative sparse low-rank representation classification method(NSLRRC)for robust face recognition.NSLRRC seeks a sparse,low-rank and non-negative matrix over all training samples.Sparse constraint makes representation vector discriminative,while low-rank model can reveal the global structures of data.Meanwhile,non-negative representation vectors guarantee that the coefficients are meaningful and better reflect the dependence among the data points.NSLRRC can approximate the test sample and classify it to correct class based on class-specific reconstruction residuals.Experimental results on ORL face database,AR face database and Extended YaleB face database show the robustness and effectiveness of NSLRRC in face recognition.2.Robust Face Recognition Algorithm Based on Dictionary LearningFace images are often affected by noises and occlusions.Sparse representation-based classifier is proved to be effective for such situations.However,considering the actual state,when significant changes are appeared in the same class of samples,the sparse representation result is severely reduced.Focusing on the above problems,we propose a method to obtain the new dictionary containing all the valid information of the training samples by iterative solution method,and then obtain the class of the test sample.The effectiveness of the proposed method is verified on the GT face database,the AR face database and the CMU PIE face database.The experimental results show that the proposed method can improve the recognition rate effectively.3.Face Recognition Based on Segmentation RepresentationIn the face recognition,the traditional representation-based classification methods do not give a processing mechanism for samples with noises or occlusions.Sparse representation requires similar samples under the same linear subspace,taking into account the differences between the distributions of similar samples,this paper presents a classification method based on segmentation representation.This method divides the training samples into two groups on average.It requires that the training samples are similar in each group and the test sample is classified according to the "distance" between the test sample and the training sample set,which reduces the impact of sample differences as much as possible,and effectively takes into account all the information contained in the training samples.Experiments on AR face database and CMU PIE face database verify the effectiveness and practicability of the algorithm.4.Linear Regression Classification Based on Sample Correlation ConstraintsEach person has unique face features,but similar faces exist in the real world.Linear regression classification method only utilizes each sample to represent the test samples linearly,without considering the similarity between reconstructed samples.In general we want the similarity between constructed sample in the test sample class and reconstructed samples in other classes as small as possible.This paper proposes an improved linear regression classification model for face recognition.Based on the linear regression classification,this method adds the constraint on the reconstructed sample correlation,then we calculate the new representation coefficients,and classify test samples.A large number of experimental results on the AR face database,GT face database,CMU PIE face database and Extended YaleB face database show that the improved algorithm proposed in this paper has certain effectiveness and robustness.