Image Annotation System Using Visual and Textual Features

We present an automated Image Annotation system called I-Tag which uses both visual and textual information of the images and recommends relevant tags for them. The automatic generation of metadata would allow image searches and content-based image retrieval (CBIR) to be more effective. We use state of the art tools on text based retrieval and image content based retrieval to retrieve similar images. We use an open source image retrieval engine called FIRE [8], to perform content based image search. We use the visual features as well as the title of the image to recommend tags for the image with high relevance. Experiments are conducted on approximately 5000 images downloaded from FLICKR to compare the accuracy of the recommended tags. Our experiments show that the mean average precision of the recommended tags is 69%.