Japanese OCR Engine "YomiToku" Develops Dedicated Custom OCR Model Specialized in Receipt Reading through Technical Collaboration with WED Co., Ltd.
WED Co., Ltd. and MLism Co., Ltd. have jointly developed a custom OCR model optimized for real-world receipt data collected by WED's shopping app "ONE," through a technical partnership involving MLism's Japanese OCR engine "YomiToku." This initiative also contributes to improving the accuracy of the general-purpose YomiToku model.
📋 Article Processing Timeline
- 📰 Published: May 8, 2026 at 00:00
- 🔍 Collected: May 7, 2026 at 15:32
- 🤖 AI Analyzed: May 7, 2026 at 17:36 (2h 4m after Collected)
WED Co., Ltd. (Headquarters: Shibuya-ku, Tokyo; Representative Director: Toshiki Arai; hereinafter "WED") has jointly developed a dedicated custom OCR model optimized for real-world receipt data collected by the shopping app "ONE," which allows users to earn money, through a technical collaboration with MLism Co., Ltd. (Headquarters: Kashiwa City, Chiba; Representative Director and President: Kotaro Kinoshita; hereinafter "MLism"), which develops and provides the Japanese OCR engine "YomiToku." This initiative has also contributed to improving the accuracy of the general-purpose YomiToku model. Details of this initiative are being released today as a case study.
Background and Challenges
In the development of highly accurate OCR models, the biggest challenge is securing training data that reflects real-world environments. Real-world documents often contain numerous noises, distortions, and special fonts that cannot be fully anticipated by general-purpose datasets.
WED's "ONE" service accumulates millions of receipt images daily. These images contain a mixture of printed characters and decorative fonts, providing a rich source of real-world data that requires advanced OCR processing.
[Details of the Initiative]
MLism and WED have formed a partnership by mutually providing data and technology, and have carried out the following two joint initiatives.
1. Development of a Dedicated Custom Model for WED
Based on WED's receipt data, YomiToku was tuned, and a custom OCR model optimized for the WED environment was jointly developed. During the tuning process, the diversity of data, exceeding expectations due to folds and special fonts, reaffirmed the difficulty and usefulness unique to real-world data.
2. Model Achievements
The WED-dedicated custom model developed through this initiative has achieved improved accuracy in receipt reading processing on ONE compared to conventional methods. This model was realized thanks to the diversity and scale of millions of data accumulated daily in ONE, and it demonstrates stable reading performance even under conditions where general-purpose OCR struggled (such as folds, blurs, and decorative fonts), further enhancing service quality.
These results are also being utilized for training the general-purpose YomiToku model. Through this initiative, character recognition accuracy in specific fields has achieved over 90%.
Company Overview
WED Co., Ltd.
Representative: Representative Director Toshiki Arai
Established: 2016
Location: Shibuya-ku, Tokyo
Business Activities: Provision of data utilization-type marketing services centered on the receipt business
Corporate Vision: "Towards a world where every spend brings excitement"
URL: https://wed.company
MLism Co., Ltd.
Representative: Representative Director and President Kotaro Kinoshita
Established: December 2024
Location: Kashiwanoha Campus 148-2, Shop & Office Building 6F, 178-4 Wakashiba, Kashiwa City, Chiba
Business Activities: Research and development of algorithms and sales/provision of licenses
Corporate Website: https://www.mlism.com/
Background and Challenges
In the development of highly accurate OCR models, the biggest challenge is securing training data that reflects real-world environments. Real-world documents often contain numerous noises, distortions, and special fonts that cannot be fully anticipated by general-purpose datasets.
WED's "ONE" service accumulates millions of receipt images daily. These images contain a mixture of printed characters and decorative fonts, providing a rich source of real-world data that requires advanced OCR processing.
[Details of the Initiative]
MLism and WED have formed a partnership by mutually providing data and technology, and have carried out the following two joint initiatives.
1. Development of a Dedicated Custom Model for WED
Based on WED's receipt data, YomiToku was tuned, and a custom OCR model optimized for the WED environment was jointly developed. During the tuning process, the diversity of data, exceeding expectations due to folds and special fonts, reaffirmed the difficulty and usefulness unique to real-world data.
2. Model Achievements
The WED-dedicated custom model developed through this initiative has achieved improved accuracy in receipt reading processing on ONE compared to conventional methods. This model was realized thanks to the diversity and scale of millions of data accumulated daily in ONE, and it demonstrates stable reading performance even under conditions where general-purpose OCR struggled (such as folds, blurs, and decorative fonts), further enhancing service quality.
These results are also being utilized for training the general-purpose YomiToku model. Through this initiative, character recognition accuracy in specific fields has achieved over 90%.
Company Overview
WED Co., Ltd.
Representative: Representative Director Toshiki Arai
Established: 2016
Location: Shibuya-ku, Tokyo
Business Activities: Provision of data utilization-type marketing services centered on the receipt business
Corporate Vision: "Towards a world where every spend brings excitement"
URL: https://wed.company
MLism Co., Ltd.
Representative: Representative Director and President Kotaro Kinoshita
Established: December 2024
Location: Kashiwanoha Campus 148-2, Shop & Office Building 6F, 178-4 Wakashiba, Kashiwa City, Chiba
Business Activities: Research and development of algorithms and sales/provision of licenses
Corporate Website: https://www.mlism.com/