Image Classification Based On Sparse Representation Dictionary Learning

The emergence of a large number of images,video and other information,making the image classification has become an essential technology.For more and more image information,how to accurately and efficiently identify the needed content has become a hot topic of concern.Sparse representation theory has been widely used in the field of image classification in recent years.This method uses redundant dictionary to decompose the signal,and enhances the robustness of the model.This dissertation mainly focuses on the research progress of sparse representation theory and sparse representation dictionary learning.Related work of other researchers is introduced,based on those research results,further investigation are also carried out to overcome the shortcomings and deficiencies.Specific research work can be summarized as follows:The current dictionary learning model uses the coordinate descent method to solve the Lasso problem with norm constraints,which increases the complexity of the classification task.In order to overcome this shortcoming,this dissertation designs a kind of dictionary pair learning model,which combines learning structured analytic dictionary and comprehensive dictionary.In the training process the model is simplified to the standard least squares problem,in the classification stage the analytic dictionary is used to solve the sparse coefficients efficiently,so as to solve the problem of classification efficiency.At the same time,the Fisher discriminant criterion is used to encode the coefficients,and then obtain a dictionary pair with the ability of discrimination.In fine-grained image classification,categories usually have a large number of shared visual features,so it is more challenging to classify them.How to extract the features of the sufficient resolution and obtain the structure of the features becomes very important.In order to solve the problem that traditional methods are not suitable for fine-grained image classification,this dissertation proposes a kind of dictionary learning model based on the dictionary pair,which combines learning shared and category-specific dictionaries.The shared dictionary is used to study the shared characteristics of various types,and the category-specific dictionary is used to learn the characteristics of each class,which makes the dictionary model more discriminative.Finally,the dictionary pair model is applied to image classification.The experimental results show that the dictionary pair can not only improve the classification accuracy,but also greatly reduce the computational complexity.Then the improved dictionary pair model is applied to the fine-grained image classification,better classification results are obtained by the combination of the shared and category-specific dictionaries.