A clustering approach to corner point analysis in hand drawn images

Drawing tasks are used widely within neuropsychology for the assessment and monitoring of a variety of conditions. Automated assessment of these tasks improves their repeatability and accuracy in use alongside a reduction in trained therapist resource loading. Using images from test responses collected in an assessment of the condition of visuo-spatial neglect, this paper presents a method utilising a standard corner detection routine (SUSAN) to establish the presence and location of corner points within an image of a drawn geometric shape. It is demonstrated that by applying a series of standard clustering systems to these located corners, the accuracy in terms of number and spatial position of corner points found assessed against actual corner position is improved.