Content-based Antique Coin Image Retrieval

With the development of computer and internet technology, digital museum has become to a new mode to manage the natural and cultural heritages by many countries in the world. Digital museum can store and manage all kinds of information of the natural and cultural heritages in digitalized format, and offers digitalized exhibition, education and research to user through the computer network. To construct modern digital museum, we need many new technologies such as 3-D laser scanning and measurement, multi-media retrieval based-on the contents, Human-Computer Interaction in the 3-D virtual scene, construct Large-Scale Multimedia Databases, Digital Watermarking etc.At the past two decades, under the development of the multi-media and internet technology, How to find out the proper images in large-scale multi-media database at a efficient method has become a tough work we confront. Under this background, CBIR appears and gets a dramatic development. Now the CBIR system has been used in many domains such as public security, Remote Sensing, medicine, digital library, construction, Computer Aided Design, geography information system etc. Antique is an important kind of things that exhibited by almost any museum. Among various antiques, old coin especially been regarded to be significant by many experts and amateurs because its high values in research and collection. So in this paper, we want to research the technology of CBIR through the antique coin image. And such a research also has its application background in the construction of digital museum.In this paper, we first introduce and analyze some successful CBIR system, after this we give a briefly introduction of our Content-based Antique Coin Image Retrieval system. Then we show how to construct the antique coin image database to test the capabilities and performances of the system. The main target of this paper is to extract features from the antique coin images. After analyzing some particular characters of the antique coin, we show three feature descriptors we had designed for the system. The first feature calls histogram of centroid distance (HCD) which is a contour based shape feature descriptor. The HCD is used as a coarse retrieval and can discriminate two antique coins that have obviously different outer contours. The following two features call modified moment invariants (MMI) and angular radial transformation respectively. Both MMI and ART are shape descriptors based on region and can use as fine retrieval. After this we also indicate the match strategy for measuring similarity at the case of multi-features. We introduce the structure how to measure two antique coin images based on the three feature descriptors we had shown. At last, we evaluate the performances of our ACIR system and analyze the results. All the experiments data can demonstrate that the features we extract from the image can work well in our retrieval system.We hope our research and practice can contribute in the improvement of the CBIR technology and the construction of the digital museum.