Image Classification Based On Improved Bag Of Words Model

With the development of network and science and technology,image data stored digitally increase rapidly,so it has become one of the research emphasis to classify different categories of images accurately by the use of image classification in short time.Image classification is a fundamental research in the field of computer vision.In recent years,image classification based on bag of words model has achieved great breakthrough,but there are still several drawbacks such as the scarcity of information in single feature,large quantization error and lack of representation upon image features in image classification tasks.To solve these problems,bag of words model is ameliorated on the basis of the original model and the classification accuracy is also enhanced.The specific research and innovation are shown below:To address the impact on image classification accuracy of the scarcity of information in single feature,image classification method based on multiple features fusion is proposed.Fusing local features and encoding directly on the fusion features before image classification.The results show that this method can well overcome the shortage of scarcity of information in single feature.The method improves the classification accuracy of bag of words to a certain extent.To address the problem of large quantization error in the model,hierarchical feature learning method combined with bag of words model is proposed.A multilayer structure for feature learning is built in which the features of images are used for dictionary learning and sparse coding.Then connect the image features of each layer for image classification.The method of hierarchical feature learning reduces errors of the encoding process and improves the discriminability of finally image feature.The experimental results show that this method can improve the classification accuracy dramatically.Hierarchical feature normalization method is proposed to make the best of the space information in image feature normalization.Features in each sub area are normalized respectively and weights are introduced for normalized features.Then normalized features are connected together by weights as the final feature of image.The effectiveness of the proposed method is verified by experiments.