Research On Stroke Distance Based Handwriting Document Retrieval Algorithm

As one of the primary styles in Multi-Model Interaction (MMI), pen-based user interface (Pen UI) allows users to achieve a natural and efficient interaction by means of freely sketching, gestures or other interactive ways and has becoming the focus of human-machine interaction research. People's demand for pen-based user interface and researches make it widely used, including portable and interactive Tablet PC, PDA and other mobile devices, as well as the whiteboard and electronic notebooks in e-learning or smart office environment, etc. How to inquire, retrieve and position accurately in handwritten documents produced by the pen-based interaction system has becoming the focus of current pen-based interaction technology. It will further promote the application and popularity of pen-based interaction.This paper is focused on the methods of retrieving handwriting words in handwriting digital documents which were produced some kinds of intelligent human-machine interaction handwriting editing system. Through the retrieval of handwriting documents and the recognition of handwriting characters are similar in some extent, there are also important differences. The mainly difference of these two tasks is that the retrieval needs to be executed in an open set, while the character recognition can be considered as searching and matching operation executed on a pre-established sample collection. Obviously, the former task has to face more complicated situation. In order to effectively solve this problem, this paper proposed a handwriting document retrieval algorithm based on stroke distance which makes full use of time and space information.First of all, we preprocessed the handwriting words of handwriting document. Then, we computed stroke distance with DTW (Dynamic Time Warping) algorithm. On this basis, we determine the stroke correspondence by means of the least priority to neighbor algorithm and local optimal algorithm. Finally, we used the DTW distance of strokes to compute the similarity of handwritten words. Thus, we proposed handwriting document retrieval algorithm based on stroke distance.This method does not require training data as a priori knowledge and has high fault tolerance of characters written by different people. Preliminary experimental results show that the method is effective. The algorithm achieves better results in testing of the database of HIT-OR3C: compared with the whole distance algorithm and energy elastic mesh algorithm, this algorithm promotes 5%~20% in the performance of precision and 3%~5% in the performance of recall in the retrieving of the 100 most frequent words in an randomly selected online documentation. Compared with exiting methods of handwriting retrieval, the systematic method proposed in this paper has an obvious advantage in retrieval efficiency, can tolerate a higher degree of writing arbitrariness, and the precision and recall rate is also higher. This method basically meets the practical needs.