Research On Image Classification Based On Bag-of-words Model

Image classification is a classical and important topic in computer vision,and it can be applied to the fields of image and video retrieval,digital library management and medical image processing."bag-of-words" feature is commonly used for image feature representation.In recent years,it has been widely concerned by researchers.Firstly,the model extracts local features and describes them;Secondly,generates visual dictionaries by clustering the image descriptors;Finally,the occurrence frequency of all the visual words in the whole image is counted,and then the image content is expressed.This article studies the bag-of-words model from the following two aspects:1.background information will disturb the classification performance of image,use image segmentation to remove background information,extract target area,and buildvisual bag-of-words model for target area;2.The category information of the known training set is a key information in image classification.A new image classification method based on the known training sets category information is proposed to build the visual bag-of-words models.The main specific work is as follows:1?Aiming at background information interfering with image classification performance,this paper try to removet background information and extract target area by image segmentation.Firstly,the background information is removed by image segmentation,and the object area is extracted.Then,extracting bag-of-words model for object area.Finally,used SVM to classify images.The experimental results on PASCAL VOC2006 and PASCAL VOC2010 data sets show that the image classification method which extracts the feature of the target area bag-of-words has good classification performance.2?Aiming at the existing image classification work without consider the optimization of visual dictionary using the class information of the known training sets,and proposed an image classification method based on TF-CDF bag-of-words model.Firstly,the SIFT feature is extracted from the image and processed by clustering the SIFT feature to generate visual dictionary.Then,TF-CDF algorithm is selecting feature words,and the selected words are used to describe image information.Finally,this paper use sLDA model to classify the images.The experimental results on the LabelMe and UIUC-Sport data sets show that the image classification method based on the TF-CDF bag-of-words model has better classification performance.In this paper,two classification methods based on bag-of-words model are proposed.The first one method of image classification is extracting the bag of words in the target area,and the second method of image classification based on TF-CDF bag-of-words model.Experiments on four real datasets show that the accuracy of the proposed method is much better than that of the existing methods,and the proposed method achieves better results.