Fuzzy object retrieval by using histogram of fuzzy Allen relations

Relative position of object description are widely used in event understanding and computer vision tasks especially in object recognition. Use of low level features cannot give satisfactory results when high level concepts is not easily expressible in low level contents. Mostly researchers are concentrating on spatio- temporal relationship between objects or regions of an object in images. Object retrieval which is taken into account the relative position of objects in images become important. In such a case classical Allen relations are used. Searched object can take various shapes and scale according to shooting. Fuzzy methods have the ability to compensate the imprecise informations and vagueness. In this paper fuzzy histograms of Allen relations are used for object retrieval. Fuzzy histograms of Allen relations are the quantitative representation of relative object position. For this purpose Matsakis's [9] algorithm for fuzzification of line segments is refined. This representation is affine invariant. Query is made by example and only corresponding relative relation between objects is considered. Results are analyzed by a well known receiver operating characteristic curve (ROC)method.