Specific Topics Oriented Image Annotation and Retrieval

The development of the Internet and the popularization of portable cameras, bring big challenges to the organization, retrieval and management of images. This paper chooses multiple topics (e.g. haze, waterlogging, flowers and maple leaves) that people pay more attention to in daily life as research objects, collects each topic's images with text descriptions from the Internet and extracts text and visual features respectively. At first, this paper processes the text information, which mainly involves the keyword extraction and the semantic extension of the keywords. Secondly, this paper proposes the improved algorithm which weights the image blocks adaptively with partial color information entropy, and extracts LBP texture and Gray-Level Co-occurrence Matrix. Finally, based on the image annotation, this paper proposes an image organization and retrieval framework of combination of image and text information, to improve user experience through a finer presentation.