Word Spotting as a Service for Handwritten Documents

In this paper, a segmentation-free and training-free word spotting method is proposed that allows mobile device users to search instances of a query word in handwritten document collections. The method is based on a document-oriented keypoint and feature extraction pipeline together with a fast feature matching method that permits the pipeline to be effectively employed in the cloud and to be introduced as a service in modern mobile device units. Evaluation results on two segmentation-free historical handwritten datasets show the efficiency of the proposed method in terms of matching accuracy along with its fast retrieval time.