Fast historic document retrieval by extracting document image summary

Historic documents such as Chinese calligraphy and old newspapers usually were handwritten or printed in poor quality so that an automatic optical character recognition procedure for scanned document images is difficult to apply. Thus efficient pattern matching techniques are required in order to do content-based information retrieval based on user´s queries. In this paper, a fast pattern clustering and image matching procedure is proposed to do image/pattern search in a historic document image based on user´s query images. The image summary extracted from the document image is constructed so that a set of distinct image clusters are formed. A couple of distance measures that calculate distance between image patterns are also proposed to evaluate their cluster similarities. By precise pattern matching and hierarchical image clustering, our experimental results show that an online query image can produce accurate and faster results than traditional approaches for a broad range of historic document images.