A framework for high level semantic annotation using trusted object annotated dataset

Abstract

Dramatic expansion and eminence of the multimedia data from the last decades, culminates to a trouble in managing, accessing and annotating the data. The high level semantic annotation (HLS) of resources in general and multimedia resources in particular, is a resilient job. The Progression in automatic annotation mechanisms have not been able to comprehend with adequately accurate results. To outfit multimedia (e.g. image/video) retrieval capabilities, digital libraries have hung on manual annotation of images. Providing a track to enact high level semantic annotation automatically would be more worthwhile, efficient and scalable with magnifying image collections. This paper intent to equip the high level semantic annotation for images, and consequently, contributes to 1) calculating semantic intensity (SI) of each object in the image depicting the dominancy factor, (2) image similarity on the bases on metadata tag with the images, and (3) clustering approach based on the image similarity to tag set of images with a high level semantic description with their calculated similarity values. The experiment on a portion of randomly selected images from LabelMe database manifests stimulating outcomes.