An Image Retrieval System Using Impressional Words and the Evaluation of the System

摘要

This paper describes the image retrieval system whose input search keys are impressional words, such as "refreshing", "bold" and "pretty". The databases for search are a flower image database and a brush paint image database. The input impressional words are translated into color patterns which correspond to the impressional words. An impressional words to color patterns table is used for this translation. Dominant colors are also extracted from each of the images of the databases. The similarity matching is made between these color patterns on the L*a*b* color space. And the system outputs the images in the similarity order. The evaluation of the system is made based on the recall-precision measurement that is frequently used for the evaluation of text retrieval systems. The recall-precision measurement is well established in the field of text information retrieval. For the evaluation of the image retrieval systems, a relevancy database was constructed. A relevancy database consists of a database of 200 images from flower image database and brush paint image database, 10 queries ex- pressed by impressional words and the relevancy judgements. The result of the evaluation shows a couple of points. The first point is that this evaluation is the first one that uses the relevancy database, and, as a result, the evaluation is accurate. The second point is that fOr each impressional words, the degrees of the influence of hue, saturation and intensity of colors were made clear when retrieving the images using impressional words. This was made by analyzing the relevancy database. The third point is that the effectiveness of retrieval was greatly improved by using the results of analysis of the relevancy database. As a result, the relevancy database is very necessary for the image retrieval system.