Improvement Of Bag-of-visual Words Model And Its Application In Image Classification

Image classification technology is one of the most important and challenging research topics in computer vision,which has been widely used in many fields,such as image retrieval,video retrieval,medical application et al.In recent years,many scholars have made a deep research on image classification technology,and the Bag-Of-Visual words(BOV)model is one of the most successful and widely used image classification models.However,there are still some shortcomings in the traditional BOV model,this paper will improve it from the following aspects:1.Concerning the problem that the scale of visual dictionary is too large and the discrimination ability of visual dictionary is poor in the BOV model,a Weighted-Maximal Relevance-Minimal-Semantic similarity(W-WR-WS)criterion was proposed to optimize visual dictionary.Firstly,the Scale Invariant Feature Transform(SIFT)features of images were extracted,and the K-Means algorithm was used to generate a original visual dictionary.Secondly,the correlation between visual words and image categories and semantic similarity among visual words were calculated,and a weighted parameter was introduced to measure the importance of the correlation and the semantic similarity in image classification.Finally,based on the weighing result,the visual words which correlation with image categories was weak and semantic similarity with among visual words was high were removed,which achieved the purpose of optimizing the visual dictionary.The experimental results show that using the optimized visual dictionary to image classification can improve the performance of image classification.2.In order to solve the problems that the lack of the spatial distribution information of the local features and the poor semantic property of image classification in the BOV model,an image classification method based on Probability Latent Semantic Analysis(PLSA)and visual phrases was proposed.Firstly,the visual dictionary was optimized by using the W-MR-MS criterion,and the visual phrases were established on the basis of the optimized visual dictionary which increased the spatial distribution information of local image features.Then,a new semantic visual dictionary that combined with visual phrases and visual words in the optimized visual dictionary was constructed.Finally,PLSA model was used to dig out more semantic latent themes based on the semantic visual dictionary.The experimental results show that the combination of visual phrases and PLSA modelcan improve the performance of image classification.