Towards Understanding and Improving Handwriting with AI.

What makes a handwriting good? If the aesthetic judgment of handwriting follows implicit rules, can those rules be recovered by observing good and bad examples? To answer these questions, we apply explainability techniques to the classification of good and bad handwriting. We show that it is indeed possible to recover these inherent rules. We develop an AI system that uses a modified version of LIME Image Explainer and generates images containing suggestions for improvement. We use single-character and word-level datasets labelled with binary labels generated via accepted rules for handwriting classification. We discuss the possible improvements to the current system as well as where this research could be applied, such as user-specific auto-suggestions.