Features for Art Painting Classification based on Vector Quantization of MPEG-7 Descriptors

Abstract

An approach for extracting higher-level visual features for art painting classification based on MPEG-7 descriptors is presented in this paper. The MPEG-7 descriptors give a good presentation of different types of visual features, but are complex structures. This prevents their direct use into standard classification algorithms and thus requires specific processing. Our approach consists of the following steps: (1) the images are tiled into non-overlapping rectangles to capture more detailed information; (2) the tiles of the images are clustered for each MPEG-7 descriptor; (3) vector quantization is used to assign a unique value to each tile, which corresponds to the number of the cluster where the tile belongs to, in order to reduce the dimensionality of the data. Finally, the significance of the attributes and the importance of the underlying MPEG-7 descriptors for class prediction in this domain are analyzed