Implementing Similar Image Retrieval based on a Semantic Information Retrieval System

X. Chen and Y. Kiyoki (Japan)


Similar image retrieval, Semantic information retrieval, vector space model, Fourier Transform


Image retrieval for multimedia databases is becoming very important as the increase of digital image resources and similar image retrieval is recognized as one of the most important issues in the image database field. In similar image retrieval, it is difficult to support individual variation among users because even when a same image is given as a query by different users, the required retrieval results may be different. In this paper, we propose a new similar image retrieval method. The most important feature of the method is that we use users' indications for similarities among images to deal with individual variation among users. In our method, a vector space is dynamically created for each user's own indications of similar images. The vector retrieval space is used to compute similarities dependent on individual user's intentions in the query. Images in the database are mapped dynamically onto the vector space according to the user's intentions. The components of the image vectors are different on the vector space. Similar images have larger component values on an axis than their component values on other axes. The retrieval processing is performed by selecting a subspace constructed by the axes that mostly reflect the user's intention. On the subspace, the similar images required by the user's query are obtained. As the experimental study, we utilize this method to implement an image retrieval system which is used for searching same images with different resolutions. Experimental results are shown for clarifying the effectiveness of this method.

Important Links:

Go Back