Fast Face Recognition Research Via Sparse Coding

Face recognition always is an important research topic in Biometric IdentificationTechnology. And many researchers from computer vision and pattern recognition havedone a lot of research. Face recognition will be applied more extensively in thisinformation age.This thesis mainly focuses on the extraction of face feature and feature recognition.The paper firstly introduces the exiting face recognition methods, then talks about twoalgorithms—sparse coding and extreme learning machine (ELM). The key point isdeveloping a new kind of face feature extraction method via sparse coding and ELM.Lastly these features are sent to classifier. Different databases and classifiers are usedfor comparison and checking performance. The main work is following:1) Introduce background and existing method on face recognition, Including PCA,LDA, SC.2) The basic theory and realization of sparse coding and ELM are introduced.3) Because of the disadvantages of the existing methods, such as relying on practicaldata and speed limited by sparse coding, This paper comes up with a new method:learn the basis function from natural images, finish the face feature extractionthrough sparse coding and ELM regression, choose the ELM classifier to finishrecognizing.4) ORL and PIE database are chosen to do experiment. A state of art performance isgot. Adjustment of ratio between the training number and testing number is donefor checking its influence. The experiment chooses three kinds of classifiers—one layer network using the softmax activation function, ELM, multi-classesSVM. Analysis of the feasibility of this paper’s method also is processed in thischapter.Lastly, this paper is summarized. And the aspects for improvement are proposed.