Offline Handwriting Recognition Pipeline Testing Tool

Recognition of handwritten text from historical documents is a very difficult problem that has not been fully solved. Many approaches have been used to recognize text and many new machine-learning-based approaches are currently being developed. To develop new methods of handwriting recognition, many ancillary steps that are not directly related to the new method must be put in place. These additional steps include loading images, cleaning documents, measuring performance, and displaying outputs. These extra actions take time that could instead be focused on the new method. This paper describes an application that was designed and created to provide all the supplementary steps needed to develop new approaches to offline handwriting recognition. It explains a pipeline-based approach that fits the sequence of steps most often used in the recognition process. The resulting application provides the framework that allows researchers to focus on their primary area of experimentation in the handwriting recognition process.

I. Introduction

Some consider optical character recognition (OCR) a problem that has been solved. For typed or printed documents, existing methods of OCR work relatively well and result in digitized text. Even recognition of handwriting works well when the words are recorded as they are being written. This type of handwriting recognition is called online handwriting recognition and has all the information of the strokes to assist in determining the intended characters [1].