Research On Texture Classification Based On Deep Dictionary Learning

Texture is a visual feature that reflects homogeneous phenomena in an image,and it reflects the properties of surface structure organization arrangement with slow or periodic changes on the surface of an object.Therefore,texture features are the basis of computer vision tasks and have been widely used in face recognition,target detection,medical image analysis,and other fields.Most of the current problems for texture classification are solved with the help of deep learning methods,such as neural networks,self-encoders,convolutional neural networks,etc.These methods often rely on a large number of training samples to achieve effective classification,which is often not satisfactory for texture classification problems with few samples.To address the above problems,the following work is done to solve the texture recognition classification problem with few samples:(1)A deep dictionary learning texture classification algorithm based on dictionary reconstruction is proposed.In this paper,the multilayer dictionary learning method is applied to the texture classification problem,and the fusion of features at different levels is achieved by reconstructing the dictionary to improve the classification performance.On the other hand,in the case of fewer samples,the features of a single modality often cannot comprehensively characterize the properties of an object,so this paper achieves the classification performance improvement by fusing the features of different modalities;(2)a deep dictionary learning texture classification algorithm based on adaptive dictionary generation is proposed.In order to reduce the misclassification of texture classification for samples with the same material,the method first performs coarse classification,i.e.,material classification,for each sample by a deep dictionary learning method.Then,for each sample,an adaptive dictionary is generated based on the material labels and other information,and then the adaptive dictionary is applied to the dictionary learning model for fine classification,i.e.,texture classification.In the subclassification stage,a dictionary with specificity is generated for each sample based on its material label,i.e.,each sample has its own independent dictionary matrix,thus greatly reducing the interference between other samples of different materials and further improving the performance of texture classification.The experimental results show that the two proposed methods have higher classification accuracy compared to other methods for the texture classification problem with fewer samples,which proves the effectiveness of the proposed methods...