Handwriting Recognition Challenges faced by an AI system

Giving you a a glimpse on the challenges with an example.

Handwriting primarily depends on factors like the writing style and the speed of writing. Let us explore these two factors with regards to the challenges an AI system will face while recognising a sample of my handwriting considering the words ‘expert 8’.  The assumption is that an AI system based on neural network has been trained on a small to medium sized dataset of handwriting styles.

I have a style of writing which falls well within the acceptable standards for human recognition. However, an AI system will need additional training to extract and then correctly identify the personalised features of the word, for example, the mixture of cursive and block writing while transitioning from letter ‘e’ to ‘r’. The curves and loops of each letter like ‘e’ would be written tilted upwards l  tend to not join two of the converging curves on the numeral 8.The AI system will need to output the closest match even if an important element is an anomaly compared to the closest match. For contextualising the recognised characters, the spaces used between the letters ‘p’ and ‘e’ would make the AI decipher ‘expert’ as two different words ‘exp’ and ‘ert’.

The next factor is speed of writing. It will be a challenge for the AI to recognise something written very speedily, for the perimeter of the character, angles, height, width and depth, will be larger than usual. Additionally, with varying speed the AI is confounded with the complexity around character’s bold-ness.

In summary, identification of character boundary, recognition of the personal style of writing(angles, curves, loops) and diverse contextualisation are few of the many of the challenges the AI system would face