Models,Quantitative Index System And Applications Of Sparse Representation Based Classification

Sparse representation is a sparse coding technique based on over-complete dictionary.In the current era of big data,sparse representation of high-dimensional data has been vigorously developed because it has a good mathematical foundation and does not require learning and training.Sparse representation becomes one of the research hotspots in the fields of image processing and computer vision.Sparse representation-based classification theory and application have also been widely studied.Sparse representation based classification(SRC)proposed by Wright et al.is a typical positive projection sparse representation model,which is used for face recognition and has achieved good results.After that,SRC has been widely applied to other fields,and many improved SRC models have been proposed.With the development of sparse representation theory,SRC model is widely used in many practical application fields.How to evaluate various SRC models,there is no quantitative index system for comprehensive measurement.In mathematics,the emphasis is on the establishment of the model and the design of the algorithm.However,in practical applications,it is difficult to understand how to choose the appropriate classification model,so it is necessary to construct a quantitative index system.Based on this,the thesis classifies SRC models according to the projection ways.Moreover,an inverse projection-based collaborative space SRC model is proposed.Furthermore,a quantitative index system from feature representation learning to classification is constructed and used in face recognition and tumor recognition.The main work is summarized as follows:(1)SRC models are classified from the perspective of projection ways.Specifically,SRC models are divided into positive projection-based SRC models and inverse projection-based SRC models.The different representation space,prior information and classification criteria,the characteristics are introduced,respectively.Several classical SRC models are as examples.(2)An inverse projection-based collaborative space SRC model is proposed.On the basis of making full use of the information contained in unlabeled samples,the training samples belong to the same category are added as a collaborative representation.The proposed model is optimized by alternating direction method of multipliers(ADMM),and the convergence analysis is carried out.Finally,the validity of the model is verified on two-class and multi-class public databases.(3)A quantitative index system is constructed to measure and select SRC models.SRC model is mainly divided into two stages: the representation stage and the classification stage.A relatively complete quantitative index system is defined to comprehensively measure SRC models from these two aspects.It quantifies the performance of SRC,and provides an objective reference for how to choose the appropriate SRC model,so that it can be intuitively selected according to requirements in practical applications.(4)SRC models and the quantitative index system are applied in face recognition and tumor recognition.Experiments are conducted on six public databases to verify the performance of SRC model,and demonstrate the feasibility of the quantitative index system to measure the performance of SRC models.