CASE 9
Is it possible to convert handwritten English into digital text?
Cost reduction via a system to recognize handwritten characters.
Case overview
We received an inquiry from a company that grades English exams as to whether it would be possible to recognize handwritten English sentences. By converting handwriting into digital text, scoring costs could be greatly reduced.
Background / Challenges
- In certain English exams, question formats and and pass/fail judgment methods were changed and updated, so the cost of scoring increased.
- In order to reduce scoring cost, we began research and development into producing a handwritten character recognition system using AI technology.
- Developed a system that uses AI to convert handwritten character image data into digital text and classify it.
Results / Outlook
AI classifies handwritten character image data into three patterns. As deep learning progresses, the ratio that AI is able to distinguish improves, leading to an increase in the recognition rate.
Our solution
- Improvement of recognition rate by AI
-
-
- 1Classify the image data of hand writing into 3 patterns by AI. ① A%：percentage of character recognition by AI ② B%：percentage of character recognition by human ③ C%：percentage of the blank data
- 2Operate “①character recognition by AI”
- 3Operate “②character recognion by human”
- Improvement of recognition rate by AI
Handwritten characters are classified into three patterns by flow chart. AI character recognition is compared to character recognition done by humans. By correcting the AI-recognized data and re-teaching it to the AI, the AI is further trained and its recognition rate improves.
-
- Image of AI handwritten character recognition system
-
AI handwritten character recognition system 
Image of the revised data of fales recognition AI recognizes handwritten characters read by OCR. Misrecognized data is corrected, further training the AI.