Image Database Indexing and Retrieval System with Parallel Computing Engine

The explosive growth of the World Wide Web has made a vast amount of information available to us especially the multimedia data such as images and graphics. The benefit that users can derive from such a resource is limited by their ability to find specific information contained in the data especially multimedia data. New generation search engine with the technology of CBIR (Content-Based Image Retrieval) is response to the need. However real-time image retrieval from a multimedia database often requires extremely high computational power and even the supercomputer requires significant amount of time for searching a medium size image database. One of the possible solutions is to use low cost parallel computer. The major objective of this project, therefore, is to develop a web-based Image Database Indexing and Retrieval System (IDIRS) using CBIR technology. Advanced CBIR algorithms as well as the traditional text-based indexing technique will be used to enhance the searching accuracy. In addition, the system will be implemented on a Beowulf Class Supercomputer with specially designed parallel algorithms for reducing the searching time.

CityWall: A 3x3 Display Wall with Use of Abacus

Video wall system will become increasingly important for applications MeshTV, Movie player, real time rendering system. It provides a next - generation large - scale display format. As a result , a visualization hardware and software should be built on this large-scale display system. Because the size of display wall is almost >10 time of typical desktop displays. The computation power must be enhauced in order to support the large amount of calculation with acceptable frame rate of the display wall . However for one powerful computer that can be used to reach this visualization requirement is enormously expensive . Therefore a system of parallel computing is considered in order to have good frame rate (30 fps) with relative inexpensive.

Image retrieval using wavelet-fractal based similarity measurement schemes

The image database, which consists of several thousand images, is very common nowadays. However, we cannot access or make use of them unless we have some dedicated methods for indexing and retrieval of these images. Content-based image retrieval is the crucial factor for digital image archiving, but is still in the preliminary stages of development. Content-based image retrieval must use an automatic indexing technique to generate a representation that allows it to preserve the semantics of the image and then to support different sorts of queries. Such features as color, texture, appearance, shape and pattern are commonly used for retrieval.

The major objective of this project is to investigate new robust feature extraction algorithms based on wavelet-fractal technology for content-based image retrieval, which is a basic requirement of MPEG-7. In practice, most of the images are stored in a compressed file format, such as JPEG. Performing the image retrieval process in this compressed domain will significantly speed up the whole searching processing. On the other hand, JPEG2000 is going to adopt the wavelet as its core element; thus we will develop new fractal-based similarity measures in the wavelet domain. Two approaches will be investigated. The main goal of these two approaches is to exploit the self-similarity of the image in the wavelet domain. The first approach is based on the fractal dimension estimation. The fractal dimension of an image is a number that characterizes the structure of the image. The second approach is based on the fractal parameters, such as the Affine Transformation Set, which is generated by the Partitioned Iterated Function Systems (PIFS). The results of this project could be used in broad areas, such as MPEG-7, digital photo albums, digital libraries and multimedia databases.

Computer Systems Lab, Department of Electronic Engineering, City University of Hong Kong, 2006