Dear Team, Myscript
I am writing this mail in response to the focus some of the issues we have felt while using the handwriting recognition in the Demo Sample App which needs urgent care. There are some letters which are not rightly labelled on drawing certain patterns. We have provided two examples of the same as a sample.
When 'x' (lowercase) is written in a single stroke then it is never classified as x.
When we wrote 'B' (uppercase) in a single stroke in a free hand tone with a little elongated vertical line which often occurred when we had written freely. Then it was mostly classified as 'p'.
We have a similar analysis of further English alphabets inclusive of uppercase and lowercase. We would like to know if these misclassifications could be corrected at your end then we will send you the complete analysis in a structured document. A demo video recording the same issue is being attached in this mail for your reference. Hoping to hear from you soon.
thank you for the update.
To be honnest, I am not able to read "Except" in the above screenshot, but rather "Export", "Enport", "Enpert"...
You shall therefore not expect our solution to be better than a human eye. For such use-case, even proposing several candidates is not likely to be satifying.
Not much can be done for such use-case, except the writer writes "normally".
Dear Mr. Olivier,
Thanks for your response.
We tried analyzing the recognition results of lexical following your reply over our prevenient query but still the results are not satisfactory enough to put this feature into the production. We evaluated via writing "Except" 100 times but got the correct only in 45 instances. We have attached one such screenshot for your reference of wrong labelled patterns of the specimen word and can send you more if required. Kindly check into this and help us to fix this.
currently, our technology is not designed for character by character recognition, but to recognize words with a lexicon attached.
We are therefore not surprised by the results you have.
At present, we do not plan to improve isolated character recognition.