A neural re-ranking method for searching ancient Arabic documents on the Web

Web users want a quick and accurate access to images. The method currently used by search engines is the analysis of text surrounding an image which usually causes errors. Since there is a huge gap between the content of the image and the textual description associated. Hence, realizing a search engine for images in the web considering their contents became therefore mandatory. In this paper, we propose a method for collecting images of old Arabic documents from the Web. This work focuses mainly on content based image retrieval by texture feature using a neural network for classification and trying to integrate the user in the search loop. The system begins with the formulation of a query text, which is expanded and sent to a conventional search engine. Then, the obtained results are filtered by a neural network and finally displayed to the user for agreement. The experiments with various query texts shown good performances and hundreds of old Arabic documents were collected.