* November 30, 2022
We have developed a deep learning OCR engine (release on December 5, 2022, see "Topics" on the left).
Error recognition rate: 1/3 (high-quality images) to 1/6 (low-quality images). Available from C/C++/C# etc. No GPU required. 32-bit/64-bit operation. Multi-threaded operation possible. Recognition speed of 600 characters per second even on notebook PCs.
Hybrid operation with conventional OCR library possible.
Utilizes user registration patterns and inherits past assets.
* November 30, 2022
Website has been updated.
* October 1, 2022
Developed an application that analyzes elementary school math word problems, converts them into intermediate representations in Prolog language, and provides solutions within the scope of the elementary school curriculum.
Runs on 64-bit. Not using GPU/Python, but capable of multi-threaded execution. Available from C/C++/C#.
Can handle word problems with two or more unknowns, and supports solution presentation at the elementary school level.
* April 1, 2022
Started development of a character recognition library utilizing deep learning (CNN).
Based on existing character recognition libraries with asset inheritance as a premise.
* 2021
Released a function for recognizing arrowheads in drawings using deep learning.
Dramatic improvement in recognition accuracy of arrows (arrowheads and their connection relationships with polylines) compared to traditional template matching.
Training with 200 patterns ?~ 360 degrees rotation ?~ scaling.
* 2019
Released a function for recognizing symbols in inherited tax land value maps.
* 2017
Completed the development of an English version chatbot using AZ-Prolog (for integration into robots).
* ^^^ update ^^^
*October 19, 2016
We have renewed our homepage.
*October 2016
We have started development of a Japanese chatbot (for implementation in a robot) using AZ-Prolog.
* September 2016
We have developed a library that applies deep learning (stacked denoising AutoEncoder + recursive feedback circuit) to the classification of text contents (names, corporate names, zip codes, addresses, building names, phone numbers, etc.)
* April 1, 2016
We have developed a library that recognizes on-screen text strings.
* September 2015
We have completed a conversation processing program project for robots.
* April 2014
We have delivered the second model of our Scan-cut projector to ASICs.
* October 2013
We have started a project to develop conversation processing programs for robots (scheduled to be completed in 2015).
* January 2013
We have delivered a Scan-cut for the Mimlabo projector.
* December 2012
We have developed and delivered a program for formatting documents captured with a digital camera.
* November 2012
We have developed and delivered a program that automatically separates multiple documents on a scanner, determines their orientation, and corrects tilts.
* 2011
We have upgraded the inheritance tax route value diagram recognition system.
* 2011
We have developed a browser for Tokyo 1/2500 digital blank maps.
* 2010
We have upgraded the inheritance tax route value diagram recognition system.
* 2010
We have developed and delivered an integrated application for raster-vector conversion and character recognition.
* 2010
We have developed and delivered a library for recognizing title blocks on drawings.
* 2009
We have developed and delivered a program that achieves high-speed processing by parallelizing the JPEG2000 encoder.
* May 2009
We have delivered the first model of our Scan-cut projector to ASICS corp.
* 2009
We have upgraded the inheritance tax route value diagram recognition system.
* 2008
We have developed and delivered an inheritance tax route value diagram recognition system.
* 2008
We have obtained the first license for our raster editing engine.
* 2008
We have developed and delivered a program for ultra-fast color dithering using blue noise.
* vvv before 2008 vvv
Development of a newspaper recognition application exclusively for the five major newspapers (Asahi, Yomiuri, Mainichi, Nikkei, Sankei)
Development of a comic panel automatic separation application
Provision of a semi-automatic font conversion application for comic speech bubbles
Provision of a raster vector conversion library for ATM drawings
Provision of a character recognition library for drawings
Development of a character recognition application for drawings
Development of a part layout application exclusively for shoe manufacturing devices
Development of a vector conversion application for line drawings
Development of a route value string recognition application for inheritance tax route value maps
Development of an application for logical inspection of electronic maps from the Geospatial Information Authority of Japan (checking logical consistency, intersection, duplication, and missing map elements)
Development of an application for digital conversion of maps from the Geospatial Information Authority of Japan
Development of an equipment catalog recognition application
Development of a contour line recognition application
Winner of the 1st Appearance Inspection Algorithm Contest 2001 (organized by the Precision Engineering Society)
Character Recognition Library
In 2022, it became compatible with deep learning character recognition.
To have character recognition functionality for servers and web applications
To automatically recognize special documents (maps, drawings, inheritance tax route value maps, cadastral maps and land registration maps)
To perform high-precision and high-speed character recognition in poor lighting conditions at distribution centers and factories
To recognize documents with special fonts like cursive fonts and CAD fonts
|
Deep Learning AI Integration for Existing Systems
I want to add deep learning AI integration to existing system functionality.
Hybrid processing of natural language processing and image processing (NLP-CV hybrid).
Business card recognition.
Document image recognition (recognition of architectural drawings, civil engineering drawings, line drawings of maps, etc. with mixed text and graphic elements such as core lines, contours, dotted lines, arrow lines, pull-out lines).
Similar drawing search (patent drawing search system).
AI service provision support in standalone mode.
Deep learning AI integration using C++/C# in addition to Python.
Support for both 32-bit and 64-bit applications.
Support for applications that do not use GPUs.
Example 1: Fast and moderately accurate conventional OCR library + re-recognition of only low-confidence characters using deep learning --> High-accuracy OCR while maintaining speed.
Example 2: Automatic categorization of recognized result strings using deep learning integrated natural language processing AI (corporate name, personal name, address, postal code, telephone/FAX number, email address, URL, date, etc.).
Coming soon: Outputting Prolog language programs with ChatGPT. Integration with Prolog language system.
Under development: Fast conventional RV conversion library + determination of broken lines, characters in contact with lines, interruptions in line drawings, arrowheads, and types of intersections using deep learning + expanding RV conversion functionality while maintaining speed.
|
Raster-to-Vector Conversion Library
Capable of recognizing dashed lines, arrows, dots, and circles with lower cost, higher speed, and lower error compared to other products.
Specialized in text processing as an OCR manufacturer.
|
Raster Editing Library
A library for constructing monochrome binary raster editing applications, the world's first of its kind.
Capable of going back and forth between raster editing and paint editing.
Want to reshape, move, duplicate, and delete line drawings (roads, buildings, lead-out lines, etc.) without affecting other shapes, while keeping the raster data intact.
|
Automatic Nesting Library
Want to arrange a wide variety of parts on irregular materials with high yield.
We have started providing nesting processing programs specialized for shoe design support systems as libraries that can be used in apparel and mold.
|
Shoe Design Support System
Want to arrange patterns of shoes and gloves on materials with high yield.
Want to arrange parts on materials with directional constraints.
Want to manually arrange evenly spaced objects on the workspace while projecting the projector screen.
Want to arrange various sizes of mold parts with high yield using holes, etc.
|
