Research Of Image Classification Based On Improved Bag-of-words

As the development of the informational society,the internet makes information growth at a speed of indexation.A large number of unclassified digital images is a great challenge for people to classify.While traditional manual annotation and classification can't adapt the demands not only on time complexity but also on efficiency.Nowadays,the method of bag-of-words achieves great achievements in the field of automatic image classification,while classical methods still exist some shortages at feature extraction stage and image representation stage.The focus of this article is to improve the classical bag-of-words algorithms.The article puts forward a modified image classification algorithm based on the bag-of-words model.The main work is as follows:At first,in order to eliminate the shortcomings of the traditional method that extracts unstable numbers of SIFT features,this algorithm cuts the image into subblocks,and then exact SIFT features from the subblock images to improve the stability of SIFT feature;Single feature can't represent the overall information of the image,this paper fuses SIFT feature with HOG feature at feature extraction stage.And then,at the stage of image representation,this paper proposes a new allocation scheme related to distance order to obtain more accurate image representation.Facing the problem that classical method ignores the spatial information,this paper combines the allocation scheme related to distance order and the SPM theory forming a new algorithm based on the bag-of-words method.This paper uses the database Caltech-101 and the database 15 Scenes to verify the proposed method,and compared with the classical bag-of-words method.The experiment results show that the classification accuracy of the proposed method is higher than that of classical classification accuracy in the database mentioned above.