Hierarchical Point-set Description Of Object Edge And Its Application In Trademark Image Retrieval

Trademark is the product of economic society,as an important intellectual property,it plays a significant role in the market expansion of enterprises and even the economic development of countries.Trademark retrieval is an important link in the process of trademark registration.As registered trademarks have been increasing every year,traditional class-based and text-based methods can not meet the demands of retrieval accuracy and real-time ability in trademark image retrieval.It is of great significance to study new trademark image retrieval method with high retrieval accuracy and time efficiency.Proceeds from content-based image retrieval,aims at shape features of trademark images,this paper proposes a trademark image retrieval method based on hierarchical point-set description of object edge.The proposed method treats the edge of the object as an unordered point-set,a hierarchical description model is built by iteratively partitioning the edge of the object into progressively smaller parts along different directions.At each level of the hierarchy structure,the geometrical features of the object edge are characterized by two measurements,partition ratio and dispersion degree.Combining them,a hierarchical description of the object shape can then be constructed.A circular feature matching strategy measures the dissimilarity between two different shapes,which is specific to the hierarchical shape descriptor.The merits of the proposed method can be summarized as follows:(1)Both contour shape and region shape can be effectively described by this method,thus it has the ability for general trademark image retrieval.(2)Based on the proposed hierarchical description framework,besides the proposed two measures,partition ratio and dispersion degree,many other measures can be included for meeting various accuracy requirements on shape recognition,so the proposed method has extendibility.(3)The proposed hierarchical description scheme makes the available descriptors characterize the shape from coarse to fine,so the proposed descriptor is multi-scale.(4)The proposed shape descriptor itself is independent of translation and scaling transformation,the shifting feature matching scheme effectively guarantees the matching invariant to the rotation of shapes,Thus the proposed retrieval method is of good robustness.(5)Instead of using all the pixel points of the object,the proposed method only takes the edge points of the object into account,and the feature extracted for shape matching is global feature,for this reason,it has a relative low computational complexity.Two standard test sets including MPEG-7 CE-2 region shape database and MPEG-7 CE-1 contour shape database are used to evaluate the performance of the proposed method.The experimental results indicate that the proposed method outperforms the state-of-the-art approaches in terms of a comprehensive consideration on the retrieval rates,retrieval efficiency and general application ability.