A Context-based Image Retrieval Method for Persian Databases

Descriptive summarization of documents in databases results in better indexing and management of information. Images in documents usually contain valuable information and retrieving them provides tools for document summarization. In context-based image retrieval systems, descriptive tags for the images are extracted from auxiliary information sources available to them. The search engine uses these tags to retrieve images. Here we suggest an automated image tagging method that exclusively relies on information mined from the document’s text associated with the image. Because of complications in the Persian language, lack of resources, and studies on this language, it has received little attention in the literature. The suggested method is built based on Persian documents. Two groups of tags are suggested. Specific tags are extracted from the caption of the image and the nearby text. General tags are obtained from the keywords of the document. Suggested methods are evaluated on the test data from the Iran scientific information database (GANJ), the largest database of Persian scientific documents. The F-measure of our suggested method is 43% for the specific tags. As for general tags, 89% are descriptive and the false positive rate is 0.002. Although suggested method has been tested on scientific documents it can be generalized for any type of Persian.