A CONSTRUCTION OF A NEW IMAGE DATABASE SYSTEM WHICH REALIZES FULLY AUTOMATED IMAGE KEYWORD EXTRACTION

Recently, a flexible image database retrieval system where image keywo rds can be captured automatically is strongly required, in order to ma nage a practical number of image data successfully. However, image rec ognition/understanding technology level is not generally sufficient en ough to achieve this requirement. In order to overcome this problem, a new type of image database framework is proposed in this paper. In th e proposed system, image keywords are extracted in fully-automated fas hion by the flexible and generalized image recognition system. Image k eywords employed in this system are a collection of recognized objects in the image, where achieved recognition levels are allowed to be int ermediate or imperfect. The concept of ''recognition thesaurus'' has b een introduced to manage these various abstraction level of keywords s uccessfully. As an embodiment of this concept, an experimental image d atabase with various types of sports scenes has been implemented and v arious retrieval evaluations have been performed. Experimental results reveal the effectiveness of the proposed method.