Context-bounded Refinement Filter Algorithm: Improving Recognizer Accuracy of Handwriting Drawing

Early detection of cognitive impairment can prevent or delay the progress of cognitive dysfunction. In the field of neurology, the Clock Drawing Test (CDT) is one of the most popular instruments for detecting cognitive impairment. This paper presents the development of the ClockReader system, a computerized Clock Drawing Test. The main function of the system is to handle handwriting recognition errors by itself. Since the ClockReader is a screening tool for dementia, it is not desirable to ask users to fix their input errors with the originally intended drawing of either numbers or characters. Therefore, we propose a simple machine learning technique, context-bounded refinement filter algorithm. With trial experiments, we prove that this simple algorithm improves the recognizer accuracy of handwriting drawings up to 88%.