Keywords spotting in Russian handwritten documents based on strokes segmentation

The keywords spotting task in handwritten documents is as follows: a user enters text that needs to be searched for in a corpus of handwritten documents. This task can significantly simplify work with archived data. We propose a two-stage algorithm to solve this problem. The first stage involves classifying the strokes, which are the main elements of handwriting. To do this, a measure of similarity based on a Fourier descriptor for elements of the stroke representation is proposed. The second level of the algorithm involves matching the query with the text. An algorithm based on optimal string alignment distance is used for this purpose. To demonstrate the results and adjust the parameters of algorithm we use images of works completed during ”Total Dictation” exam.