With the grow of the Internet and rapid advances made in storage devices, non-standard data such as images and videos have grown significantly in size. Rich information is hidden in these data collections that are potentially useful in a wide range of applications and research. The process of discovering valuable information from image data termed image mining finds its sources in data mining, content-based image retrieval, image understanding and computer vision. Image description constitutes the initial basis for the image mining. The ability to develop an efficient and effective image description model has been a more and more interesting and challenging topic of research. This thesis focuses on new key technologies for building an effective image description tool, namely transformation-based image description scheme. This scheme combines wavelet transform to construct accurate data modeling, XML technology to implement a general storage system for any set of images, and ontology approach to produce an effective data mining tool in a unified framework. The aim is to achieve an image description tool that allows a high level of flexibility in describing images.