Design Patent Retrieval Based on Shape Semantics

In order to improve the performance of design patents retrieval,design patent image retrieval system based on shape semantics is proposed to exploit characteristics of shape organization of design patents to bring the challenging "semantic gap".Design patent images are classified according to the shapes,and discriminative information between different patent shape categories is utilized to extract shape semantic features.Then the similarities of semantic features between query image and images in database are calculated and sorted to obtain the most similar design patents.A shape semantics based relevance feedback is proposed to improve retrieval results by adjusting feature distances based on the initial retrieval results and category information of feedback images.Experiments on real design patent images demonstrate the effectiveness of proposed method.The retrieval system meets the practical design patent retrieval demand which should reach high precision on high recall.