A Data Model for Querying Wavelet Features in Image Databases

Multimedia databases deal with storage and retrieval of complex descriptors of image contents, called features. Traditional techniques consider features as “black boxes,” often represented as vectors for which the only operation defined is distance computation. This “modus describendi” resulted in the widespread utilization of similarity queries, which rest solely on the computation of distances between feature vectors. The capacity to express queries more complex than simple similarity, however, rests on the possibility of describing and manipulating the structure of the features. In order to achieve this, features should be described as complex data types inside the database, and opportune operators for manipulating these data types should be defined. In this paper we try to demonstrate the efficacy and feasibility of such a program by modeling a widely used image feature: the wavelet transform. We represent features using a complex data type derived from arrays, and propose a basic algebra from which operators can be derived to manipulate and query wavelet features.