Retrieval of free form hand-written draft image using geometric hashing

One interesting and practical problem is the storage and retrieval of handwritten documents. It will be very useful to store the documents and retrieve them later. In this paper, we propose an approach to retrieve freeform hand-drawn drafts stored in the format of image without using optical character recognition (OCR). This paper tries to rely on the structural information of the drawings to perform the retrieval. In the system, feature points are extracted and the match carried out between two feature point sets. To achieve basic affine transformation invariance, we resort to a geometric hashing technique to perform the representation and similarity calculation. For the hashing scheme, the "connection vector" is used instead of the X-Y coordinate to index the hash table so as to make the matching less sensitive to the variance of writing. A simplification of geometric hashing and component-based matching are also adopted to speed up the retrieval.