Image Classification Algorithm Based On CSIFT Characteristics Research

With the rapid development of Internet technology and digital image processing technology,it has become more and more convenient to access,transport and switch image.Digital images fulfill the Internet and our ordinary life.Image classification and management with manual method is time-consuming and laborious,and is unpractical.Therefore,the method for image classification based on semantic information has become a focus research topics.Image classification techniques has broad range of applications,such as: image content retrieval,tourism navigation,medical image applications.The main difficulty in image classification is to develop an efficient method for image feature extraction and matching.In traditional image classification algorithm,the color images are transformed into gray images,and then,image classification performs with the local feature extraction algorithm SIFT.The color information is discarded.In this paper,an improved algorithm based on the characteristics of pyramid tree CSIFT algorithm(vocabulary-guided pyramid match kernel,VGPM)for image classification is proposed.Firstly,CSIFT algorithm(Colored Scale Invariant Feature Transform)is used to extract the color features of the image;secondly,an image feature descriptor is established,and according to the VGPM principles,the VGPM tree pyramid is built from the feature vectors;thirdly,the spatial relations between local features are introduced with VGPM method;lastly,a linear SVM classifier is used to perform the classification.The experimental results show the algorithm has good classification performance.To solve the sparseness of features extracted by CSIFT algorithm,we propose a sparse coding space-based Pyramid CSIFT feature matching classification algorithm.Firstly,the CSIFT local features are sparse coded,and remodeled through sparse matrix,and secondly,build the pyramids matching linear space kernel,through space pyramid features,the coefficient reconstruction coefficients are mapped to different scales of high-dimensional sparse vector to generate an image of the same area representation,Finally,linear SVM algorithm is used for classification.The experimental result shows that,compare to the traditional non-linear kernel classifiers the proposed image classification algorithm has higher accuracy and efficiency.