Shape Descriptor Construction And Its Application In Trademark Image Retrieval

Trademark image retrieval has played a very important role in the trademarkregistration, trademark protection and business development. So it has significantresearch value. One of the most important problem of trademark image retrieval is to usewhich descriptor to describe the content of trademark image. The trademark images areusually artificial image, and its composition mainly includes a variety of shapes andtheir combinations. Consequently, it is very meaningful to find a proper descriptor toefficiently describe the features of the shape of the trademark image.This paper proposes an effective triangular spatial relationship to solve the shortcomingof describing the feature between two adjacent contour points, and use it and thehistogram of the centroid distances to build a new contour-based shape descriptorETSR-HCD for trademark image retrieval. The advantage of this combination is that itplays the advantage of HCD effectively representing the feature of each contour point,and the effective spatial information is integrated into the contour-based shape descriptor.A large number of contrast experiments on a standard image set have proven theeffectiveness of this descriptor.This paper proposes that we cluster the contour points of shape and use cluster centeras a new class feature point to instead of the traditional features point to describe theregion-based shape feature, and it is effective solution to overcome the defects oftraditional feature point. The features of cluster is as the feature value of the new featurepoint. For the new feature point, we present an effective triangular spatial relationshipcombining with features of object to describe the new feature points and its spatialfeatures, then use it and Zernike moments to build a new region-based shape descriptorFP-ETSRFO-ZM for trademark image retrieval. The advantage of this combination is thatit plays the advantage of ZM describing the global feature of shape, and region-basedshape descriptor can make up the deficiencies that ZM can not effectively distinguishsmall deformations. A large number of experiments on a standard image set have proventhe effectiveness of this descriptor.Many literatures show that the performance of the combination of descriptors is betterthan the single descriptor in shape descriptors. So, we will combine the contour-basedshape descriptor and the region-based shape descriptor that this paper have presented to build a new shape description ETSR-HCD+FP-ETSRFO-ZM for trademark image retrieval. Itmakes contour-based shape descriptor with the region-based shape descriptor complement each other.A large number of experiments on a standard image set have proven the effectiveness of thisdescriptor.