| Recent years, with the rapidly development of information technology and imaging technology, the number of digital images is growing at an alarming rate, the demand of information to the people is gradually shift from text to image. To the users, finding the interested image from a large number of images in the library is very difficult. Therefore, image recommendation has become a hot issue in the multimedia field. This paper focuses on how to establish effective learning model to solve the problem of personalized image recommendation.The existing image recommendation methods mainly have two categories:the recommendation method based on image text and the recommendation method based on image content. Taking advantages of the above two categories, this paper puts forward a personalized image recommendation method based on semantic analysis, deduce the user implicit search intention through personalized analyzing the historical images from users, and recommends images which from the large image library to conform the users requirement. In order to analyze the historical images of users, we make full use of the convenience and magnanimity of social image website. First, we built a database called SPRI, which includes training dataset and testing dataset. Training dataset contains40users, each user contains several scenes,2000images, and testing dataset contains240panoramas. Second, multiple instance learning is used to spread the image tag to the region, image region label is combined to complete the image classification. Third, we develop a predictive framework based on the latent SVM model to retrieve the most relevant images from the dataset at an individual user level, which model relate to global-level features that influence it in a globally optimal way.In order to demonstrate the effectiveness of the proposed algorithm, we make a large number of experiments based on SPRI database. Experiments show that the algorithm has good effect on image recommendation. |
PDF Full Text Request