Description:VisuAl and SemanTic image search
Abstract:VAST (VisuAl and SemanTic image search) system is a large-scale content-based web image retrieval system, which could help users to highly efficiently find their required image information from Internet. Everybody can use VAST to help them find what they need from Internet. The main research issues involved in VAST system are mass image information management and content-based image retrieval technologies. VAST automatically combines the keyword and visual features for web image retrieval, whose processing are divided into off-line processing and on-line processing. 1. Data Gathering: The crawlers regularly traverse the web to collect network images, surrounding text, URL and etc. into a raw database. 2. Data Analyzing: The collected images and related data are analyzed to extract useful information, which includes low-level image content features (e.g. color, feature, shape, layout), text semantics (e.g. title, category, surrounding text), and other image attributes (e.g. height, width, data type, author), stored into visual feature database and semantic database. 3. Indexing: Indices are created for visual feature database and semantic database, including the indexing and clustering operation for the databases. The online processing is as follows: the User Interface module allows users to graphically pose traditional text-based queries (e.g. Query by Keywords). After taking the query specified at the User Interface, the Retrieval module uses the distance metric to evaluate the query with the Visual Feature Database and the Semantic Database, and returns the images that are best matches to the input query.
The application was ported within the framework of the DEGISCO project.