A Method of Trademark Search Based on the Automatically Assigned Codes of Vienna Classification and Figure Shapes

Abstract

Demand for the search of similar trademarks is increasing. In the existing methods, the similarity is measured based on the similarity of appearance with the shape. However, to judge the similarity of appearance, trademark examiners also take into account the classification of graphical elements in trademarks. To realize this function, we focus on the Vienna Classification which is the classification of graphical elements in trademarks. If two trademarks have the same Vienna Classification code, they tend to have similar contents. However, the codes are currently assigned by experts, which prevent us from using the codes in search methods. We solve this problem by automatic assignment of Vienna Classification codes implemented using the Deep Learning. The proposed method takes into account both the automatically assigned codes and shape similarity for search. We have succeeded to improve a three percentage point accuracy as compared with the method without using the Vienna Classification.