Research On Image Classification In Social Media

With the rapid development of the Internet, our network is now filled with a lot of social media data, such as images, videos, blogs and so on. Meanwhile, a large number of popular social networking sites, as well as pictures uploading and videos sharing sites begin to emerge. How to organize and manage these vast amounts of data efficiently so that the users can obtain the necessary resources easily, quickly and accurately, especially the image data, is becoming more and more important. Under this background, classification and content identification of the image has become a hot research topic in the field of computer vision, getting more and more attention of scholars. The traditional image classification which based on the model of bag of words image representation is on the one hand unable to take full advantage of the rich contextual information, and on the other hand is a direct access to all the local features of an image. So the local features of the main area or object in the image can not be accurately expressed since the accuracy is affected by the irrelevant area. In our paper we present an accurate image classification method based on social image to solve the problem described above.In this paper, we first introduce some currently popular image feature representation, but focus on the relevant methods used in our paper. We present a new algorithm of the image classification based on major area extracted of the image (ROI) and the saliency of the feature points. It can not only effectively represent the image, but also greatly reduce the number of feature points of image local feature extraction. Based on this method, this paper also presents a new image classification method based on social images. We first detect local features of an image within the area determined by the ROI algorithm. Then, we integrate the pyramid principle into the traditional bag of words model (BOW) to improve the accuracy of image representation and classification. Experiments have shown that compared with the traditional image classification based on the bag of words model, the accuracy of this method has been greatly improved. Moreover, according to the efficient image classification method introduced in this article, this paper presents an image tag recommendation method based on social media image.