Xiangyang LI, China | ISBN: | Submitted: Jun 24, 1998

Content and Semantics Based Image Retrieval for the Preservation, Restoration and Pattern Creation of DunHuang Fresco

The DunHuang Art Cave, located in Gansu province of China, is an unparalleled grotto art museum in the world. It spanned ten dynasties of over one thousand years(304~1368). Well preserved to this date in the caves are more than 555 grottoes, over 2000 painted statues and 5000 square metries of wall paintings. With the greate advances in multimedia techniques, image processing and optical storage, we try to develop image management system to store, manage, and retrieve fresco for the preservation, restoration and pattern creation of Dunhuang art.

Currently, the content based image indexing and retrieval is a promising alternative to relying on text retrieval. Dunhuang fresco is different from the traditional image, the fresco is created artifically by the painter during the different dynasty, and the fresco image contains some semantic information about art creation and Buddhism culture. So we combine the visual image content with the sematic information to retrieve the DunHuang art information.

The main idea involved image indexing and retrieval in this paper is as follows:

1) Based on color histogram to retrieve fresco. The pigmentation rule is a key component of Dunhuang fresco. We use the perceptional color space (HSV) and cluster the color space to establish the color histogram for each piece of fresco. To enhance the system performance, we combine the "coarse granularity" histogram with "fine granularity" histogram together to retrieve image .The method is also very important to the fresco color restoration.

2) Based on shape to retrieve pattern element. Dunhuang fresco contains all sorts of shape, which reflect the pattern creation style during the different dynasty, it forms the Dunhuang fresco characteristic. We will offer the pattern shape representation and establish the shape indexing .We also offer the interactive and automatic tool to extract the contour of pattern shape.

3) Based the layout of the image object to retrieve fresco. The layout of pattern element forms the fresco style, which depends on the specific Dynasty. This paper will present the time-space relation representation of the image object layout, and the relevant retrieval method is presented.

4) Based on the semantic information to retrieve fresco. The different color, shape, layout of pattern element stands for the specific Buddhism culture. They contaion the different context information in the fresco. This method is very useful for Dunhuang expert to retrieve and study Dunhuang freco. In this paper, we offer the formalization of the semantic information and try to implement the map between the low level image information----such as color histogram, shape, pattern layout inforamtion---and the high leval abstract semantic information. The relevant retrieval method, such as the spatial reasoning and analogy resoning, is introduced.

5) In the end, we will present the reseach result of our project.

*) This project is supported by China Natural Science Foundation Key Project.

[more information]

[other authors]
Dongming LU,Yunhe PAN

[keywords]
Image Database, Content Based Image Retrieval, Semantic Based Image Retrieval