Content-based Web Image Reranking

With the continuous growth of online digital image, video and other digital resources, image/video searching has been a very active and challenging research area. Compared with the past, the ranking algorithm should not only deal with huge amount of data, but also satisfy users’needs and improve their search experience. However, the existing search engines are mainly based on text information, which cannot satisfy the needs of users. Therefore, the technology of how to realize the web image reranking algorithm based on content using the text-based information became an urgent request.According to the above problems, this paper researched two web image reranking algorithms based on content:First, active reranking algorithm base on improved Ranking SVM; in the light of the characteristics of reranking objects, this algorithm introduced the improved Ranking SVM to take the initial ranking result as one of the image’s features, and then combined with the active learning to make up for the defects of supervised learning algorithm requiring manually labeled greatly. In the last, we did horizontal and vertical comparisons, which show the algorithm is beneficial for improving the initial ranking result.Second, image reranking algorithm based on visual pattern; it is a self-reranking method, which does not rely on the external knowledge, but just excavate the information of image itself. This paper caught the principal of the BOW feature, starting from the principal component analysis method, gained the visual pattern information from the BOW feature, and then used the distance concept to realize the thought. Learned from the experimental results, this algorithm is simple, with small amount of calculation and good applicability, but greatly improved the initial search ranking performance.