Research On Image Classification Based On Low-rank And Sparse Representation

With the development of information technology,image has become the most important way to obtain information,which is intuitive,vivid and clear way of information transmission.Facing with the huge amount of image data,we need to retrieve and classify them.However,the workload is so huge that it is impractical to be handled manually as in the past.Therefore,it is urgent to be solved that how to use the computer to help us to classify images automatically.Image representation is a basic issue in the field of image processing.Sparse representation is one of the most widely used representation methods for image,which has achieved fruitful research results in image processing and shows great vitality.Based on the above mentioned,we study the theory of image sparse representation,low-rank representation,dictionary learning and their applications in image classification.The purpose of this paper is to improve the accuracy of image classification by seeking an image sparse representation model with discriminative ability.Our main work includes the following aspects:1.We propose a novel sparse image classification method based on sparse representation and low-rank supervision.The classification model is implemented with sparse representation and codes the image discriminatively.A supervised matrix by low-rank representation is incorporated into the sparse coding model.The self-supervision constraint of image samples is obtained by using the correlation among samples to make the features of coding more significant and enhance its discrimination.At the same time,numerical optimization algorithm is also developed to solve the objective function in this model.The proposed algorithm is verified and analyzed through comparative experiments.2.We propose an image classification model based on low-rank shared dictionary learning.According to the image sparse representation theory,it has many advantages to represent image with dictionary.This method learns the relative dictionary for different categories in dataset.On this basis,a collaborative dictionary learning model including both of shared and feature dictionary is established,in which common features embedded in data are eliminated to improve the discrimination between different categories.The low-rank constraint is applied to the dictionary learning model,which makes the atoms in the feature dictionary to present more individual characteristics of different categories and improve the classification accuracy.