Trial: Smart Handwriting Analysis and Recognition Platform (SHARP)

Description

Deliverables achieved:1) We have developed a new online handwriting assessment platform that allows us to collect and upload multiple students' handwriting data at the same time using tablets. Our web-based system also provides the ability to assess and analyse the Chinese handwriting speed.2) Our online handwriting analysis platform can generate a speed assessment report to help occupational therapists assess a student's handwriting ability. It can analyse the Chinese handwriting speed at the primary education as well as determining the legibility of the handwriting. In particular, our SHARP system is able to detect if the student has errors such as additional stroke, missing stroke, concatenated stroke, reversed stroke, wrong strong sequence, etc. The analysis on the student's handwriting performance will be shown both numerically and graphically on the SHARP webpage.3) Deep learning technology using long short-term memory (LSTM) under the Recurrent Neural Network (RNN) architecture has been applied to analyse the legibility of the students' handwritings. The system has been trained on thousands of collected handwriting characters and the parameters have been fine tune with the aim to boost up the performance.4) Assessment data of normative samples and disabled group have been collected using our SHARP system. We have collected handwriting data from 608 students from 6 primary schools, with 507 typically developing students, 85 students with special education needs (SEN) such as Autism Spectrum Disorder (ASD), Attention Deficit Hyperactivity Disorder (ADHD), Specific Learning Difficulty (SpLD), etc., and 16 non-Chinese speaking students. Handwritings of 200 characters from 10 students were checked manually for handwriting errors and the gold standard was provided by a panel of teachers, therapists and researchers. Statistical analysis has been carried out to compare the manual checking results and the results obtained by our SHARP system, and the outcome confirmed that our SHARP system has substantial agreement with human judgment in identifying most of the handwriting errors.