Construction Of Visual Dictionary Based On Bag-of-Words

With the rapid development of the Internet,a large number of digital images arise in our lives.In the face of huge amounts of image resources,how to accurately and efficiently use image classification,retrieval and mark to manage pictures,which has become the current hot topics in the study of intelligent information processing.Bag-of-words(BOW)model is one of the most widely used in the classification algorithm,and has excellent performance in the image classification,so it has been widely studied and applied.In the BOW model,visual dictionary is the basis of dictionary model.Constructing more descriptive visual dictionary not only improve accuracy of image classification,but also reduce the computational complexity of the algorithm.An important research on BOW model is how to create and improve vocabulary to represent images effectively and improve performance of BOW.This paper studies on the methods of optimizing and improving visual vocabulary,which aims at creating distinguish visual vocabulary.And the optimization of vocabulary is implemented in image classification,to improve the accuracy of classification.Around the above content,the content of this paper is mainly reflected in the following two aspects:Firstly,on the basis of the traditional phonetic model and the existing Shannon entropy,the problems of the two are analyzed,and then the problems are improved in order to screen out the visual dictionary with high recognition.First,the training set is clustered by category.Then,the Shannon entropy of each visual word in each class is calculated by using the improved Shannon entropy,and the visual word with Shannon's entropy is deleted,that is,the visual word with low degree of recognition in each class;Finally will be optimized after each class of visual dictionary together to get the optimized dictionary.Experimental results show that the new method can choose a more representative word,improve the classification results.Secondly,the paper analyzes the problems existing in the process of word selection in traditional mutual information,improves the traditional mutual information,and applies it to visual word selection to screen out high visual words with high recognition.This method makes a comprehensive analysis of the influencing factors and shortcomings of the word selection of mutual information,which is more conducive to the selection of representative visual words.In order to solve the problems in the construction of the visual dictionary,this paper proposes two methods:optimized Shannon entropy and mutual information.In this paper,the essence of two kinds of visual dictionary method is selecting the representative visual words to generate each class of visual dictionary.In this paper a visual dictionary with recognition is applied to image classification,which improves the accuracy of image classification and the performance of bag-of-words,and improve the speed of algorithm.