Research On Multi-feature Fusion Scene Classification Method

With the rapid development of multimedia technology,a large number of digital images have been produced.The use of artificial classification of these images is a tedious and time-consuming process.In view of this phenomenon,image classification has become a hot research topic.At present,there are many image classification algorithms proposed by researchers.Because the information is very rich contained in the image,and the spatial distribution of each other is complex.Aiming at the deficiency of single feature description,this paper proposes an image scene classification algorithm based on multiple features fusion.This paper first analyzes the research background of image scene classification,the research status and related applications at home and abroad,then introduces the models and techniques used in this paper,including probabilistic latent semantic analysis(PLSA),Locality-constrained Linear Coding(LLC)and support vector machine(SVM)working principle,then add the multiple features fusion method in the model and algorithms,classification experiments and the results are analyzed and compared.The main research contents are as follows:On the one hand,a multi-feature fusion image scene classification method based on PLSA is proposed.The method extracted the LBP features and SIFT features of the image,The combination of the two can be used to describe the image,which can provide more abundant information.Two types of features are used to quantify by the bag of visual word model,and the corresponding feature of the word bag is generated,and then the words of the two characteristics are directly connected to get the word bag representation of the image.Then,the word bag representation of the image is combined with the PLSA model to study the distribution of the visual theme and the distribution of the underlying theme.Finally,the potential distribution of the image is combined with the SVM classifier for classification and discrimination.The first part,the optimization parameters of PLSA model,and then the image scene classification algorithm were compared with the single feature and other image scene classification algorithm.The experimental results show that this algorithm improves the classification accuracy and verify the feasibility of the PLSA model framework based on multiple features fusion.On the other hand,a multi-feature fusion image scene classification method based on Locality-constrained Linear Coding is proposed.Firstly,this method extracted GIST feature,SIFT feature and PHOG feature,Then the SIFT feature is encoded using the LLC method to obtain the sparse vector representation of the SIFT feature.Then the SIFT feature sparse vector representation,GIST feature and PHOG feature fusion in series,and get the image feature representation.Then the feature vector representation of the image is combined with multi-class linear SVM classifier to get the final classification result.The experimental results show that the proposed method has better classification accuracy compared with the single feature linear coding method.Compared with other classification framework,the proposed method has better classification performance.The validity and stability of the feature fusion classification model are verified.