Automatic Image Annotation Based on Region Feature - Multimedia and Signal Processing - page 137

Automatic Image Annotation
Based on Region Feature

Ke Chen 1,2 , Jinxiang Li 2 , and Liang Ye 2

1 JiangSu Province Support Software Engineering R&D Center for Modern Information
Technology Application in Enterprise, Suzhou, 215100, China
[e-mail]
2 Suzhou Vocational University, Suzhou, 215104, China
{ljx,yel}@jssvc.edu.cn

Abstract. An automatic image annotation approach is proposed to bridge the
semantic gap issues in image retrieval. It begins with building connection
between semantic concept and image region by image segmentation, and then
extracting visual features from image region in order to find the correlativity
between semantic concept and image region in the annotated image while
annotation, calculating the similarities between different image regions to
annotate the unannotated target images, using the given correlativity as a priori
knowledge. Experiments conducted on a 2000 image dataset demonstrate the
effectiveness and efficiency of the proposed approach for image annotation.

Keywords: Semantic Concept, Image Region, Visual Features, Priori
Knowledge, image retrieval.

1 Introduction

The study on management technology of multimedia data has always been an
important direction in the whole database study. Comparing with classic database, the
multimedia data are different with characteristics like large quantity, unstandard
format and so on. How to bring the multimedia data into the handlerable range of
database management system is the most important problem need to be solved for
adapting the database management for new using demands. The image data used as
main part of multimedia data, has mostly been stored and managed as general data
type. What actually been stored is just a document name, the system can reach
expected effects such as showing and reading images by calling this document while
using. However, storing and showing image data effectively are way far from enough,
besides, it should be flexible enough for handling various kinds of contents included,
such as retrieval the interested image content from a large mass image database,
getting images which fit for some criterion of semantic constraints and visual feature.
i.e. a image data should be generally stored and managed as a dataset instead of an
independent data document. Image contents need to be organized systematically and
managed by database management technology, by this way users can take advantages
in using, managing and manipulating image data.