Date of Publication :7th November 2016
Abstract: For the improvement of the performance of a text-based image search, Image reranking is a efficientmethod. There are two reasons for which the reranking algorithms are limited and they are: One is that the data that is connectedwith images is not coordinatedwith the actual visual content and the second reason is that the reextracted visual features do not exactlydescribe the consequentialsimilarities between images. The relativeof retrieved images to explorequeries has been more correctlydescribed by user clicks, in recent years. However, the lack of click is the data critical problem for click-based methods, since users have clicked a small number of web images.Consequently, the solution to this problem is byguess image clicks. A multimodal hypergraph learning based sparse coding method is proposed for image click prediction, and be relevantclick data that has been obtained to the reranking of images. To build a group of manifolds, a hypergraph is adopted. A hyperedge exist in a hypergraph is the edge that connects a set of vertices, and conservethe constructed sparse codes. The weights of dissimilarmodalities and the sparse codes are obtained by an irregularoptimization procedure. Finally, to describe the predicted click as a click or no click, a voting strategy is used from the images that was matchingto the sparse codes. Image reranking algorithms are used to progressthe performance of graph-based the use of click prediction is exposedby an supplementaryimage reranking experiments on real world data that is useful.
Reference :