Skip to content
Home » First Accounting Co., Ltd. First Accounting presents research on scene text recognition using a diffusion model at MIRU2023

First Accounting Co., Ltd. First Accounting presents research on scene text recognition using a diffusion model at MIRU2023

First Accounting Co., Ltd.
First Accounting presents research on scene text recognition using diffusion model at MIRU2023
……
At MIRU2023 held at Act City Hamamatsu from July 25th to July 28th, 2023, Dr. Masato Fujitake, co-founder and chief research scientist (CRS) of our company, spoke about “Scenes using diffusion models.” I gave a general paper presentation on “Text Recognition”.
[Imaged61842-85-1177bdf9da218ac39497-0.jpg&s3=61842-85-69a1f6738619ace6d166dc19c1f8cf21-2000x1500.jpg
Chief Research Scientist Masato Fujitake, Ph.D.
The Image Recognition and Understanding Symposium “MIRU” is Japan’s largest conference on image recognition and understanding technology. It is a place for discussion and exchange among researchers, engineers, and students who will lead the next generation, regardless of university or industry, and a place for presenting and discussing the latest research from basics to applications. ■Presentation overview – The first character recognition method using a diffusion model – This presentation was also adopted at IEEE ICIP2023, an international conference on image processing technology, and realized scene text recognition using a diffusion model, which is attracting attention around the world. This is what I did. To recognize character information (scene text) from scene images such as photographs, it is necessary to recognize character information in various fonts, colors, and shapes as text. A large number of methods have been proposed in the past, but the one that First Accounting has proposed is a new method never seen before: scene text recognition using a diffusion model, which is one of the generative models. Diffusion model (noise removal diffusion probability model) is an AI technology that infers the original image from an image with added noise and learns to approximate the original image. This technology is also used in image generation AI, which has become popular in recent years, and is attracting a lot of attention. The diffusion model is very famous as a technology for generating images, but our research team considered a new approach to apply it to character recognition. Although it is a simple method compared to conventional methods with a complex structure, it achieved higher accuracy than existing methods in numerous benchmarks, demonstrating the effectiveness of the new character recognition and attracting attention not only to MIRU2023 but also to other domestic academic conferences. We are also deepening discussions on new research directions. ■About First Accounting First Accounting is a company that provides services for corporations to utilize the power of AI to streamline and automate accounting operations. In order to improve character recognition technology, he researches not only the computer vision field but also the latest technologies such as generative AI and LLM, and presents papers at various academic conferences. We develop services based on these research results and provide them to many large corporate customers and accounting vendors. Our purpose is to “instill confidence and courage by removing constraints.” We will use AI technology to remove various constraints on our customers’ operations and do our best to help them focus on more valuable operations. Company name: First Accounting Co., Ltd. (TSE Growth: 5588) Address: 3rd floor, VORT Hamamatsucho I, 1-6-15 Hamamatsucho, Minato-ku, Tokyo Established: June 2016 Representative: Keitaro Mori, President and CEO URL: https:// www.fastaccounting.jp/
■Inquiries regarding this matter First Accounting Co., Ltd. Contact: Public Relations E-mail: press@fastaccounting.co.jpAll product names and company names are trademarks or registered trademarks of their respective owners.
More details about this release:
https://prtimes.jp/main/html/rd/p/000000085.000061842.html



%d