SmartOCR Handwriting Recognition Application: From Concept to Breakthrough in the TOP 10 Sao Khue Awards 2019

Saving time, reducing personnel costs, improving work productivity... These are what GMO-Z.com RUNSYSTEM aims to bring to businesses through the research, development, and implementation of SmartOCR – a handwriting recognition application powered by Artificial Intelligence (AI).

SmartOCR is software for recognizing handwriting and extracting data from image-based characters into Vietnamese and Japanese text. Built on AI technology, SmartOCR’s Optical Character Recognition (OCR) enables the conversion of image-based documents (scanner outputs, photos, image-based PDF files, etc.) into editable documents (text files, Word files, etc.).

At present, there is no similar product available on the Vietnamese market. Recognizing the strong demand from traditional industries that rely heavily on paper documents—such as banking, insurance, and public administration—along with the increasing need for identity verification using personal papers, the company decided to research and develop SmartOCR using AI technology.

A remarkable and proud achievement for our company came on April 21, 2019, when SmartOCR was honored in the Top 10 Outstanding IT Products and Services at the Sao Khue Awards 2019. Out of 94 IT products and services recognized, SmartOCR earned a Top 10 position based on evaluation criteria such as market scale, technology, superiority, social impact, applicability, and pioneering spirit in the Fourth Industrial Revolution. It stood proudly alongside familiar names such as Viettel, FPT, and BIDV that have won awards for many years. 

Achieving success with SmartOCR was by no means easy. It required tremendous effort and countless hours of dedication from the entire project team. Building a regular product is already challenging, but developing one based on AI technology was even more demanding.

Just as humans need to learn, artificial intelligence must be trained with data. There are many methods for building AI projects, each with its pros and cons. With limited training data, we had to resort to more traditional methods involving rules and algorithms, though less effective. At the beginning, our team had exactly zero training data to work with.

To collect enough data, the team tried everything—searching Google, purchasing datasets, generating synthetic data for training. However, these efforts were insufficient, so we asked our fellow Runners in the company to help by writing samples in Japanese and Vietnamese. Even then, the number of contributors and labeled samples was modest compared to the vast character sets needed. Therefore, we adopted a method that required less data and compensated for the shortage by generating additional variations using image processing algorithms that simulated different conditions—ultimately achieving the high accuracy we have today.

The team had a real “heart-stopping” moment when the trial deadline for a client approached, yet the accuracy was still below expectations. Everyone was on edge, working late into the night to deliver the best results possible. Fortunately, when the Japanese client brought real test samples, the results far exceeded those of our training data, and we all breathed a sigh of relief.

There was pressure, of course, but also countless joys and memories. Despite the team’s small size at the beginning, the bond was strong. Mr. Đào Bảo Linh (AI Team) shared: “In the early days, the team only had two members: me and Mr. Nguyễn Thành Đô. At one point, because of project needs and plans to expand AI staff in the Ho Chi Minh City branch, I was assigned to work there, carrying with me the hopes from Mr. Minh (Deputy Director) for a stronger AI team in the future. After returning, I learned that while I was away, my teammate often said he felt very lonely, missing me, and wishing for my return. Even now, we still joke about it from time to time.”

Today, SmartOCR is steadily improving and proving its quality, receiving enthusiastic recognition from both the media and customers. We wish SmartOCR and the team even greater success and growth in the future.