TopSoft is a Software Development Co-op with offices in the USA and Ukraine.
We began development of camera-based OCR technology in 1998.
At that time, our primary business was to provide software development services to European consumer electronics companies.
To help further our business, we opened a new development center in Ukraine in 2004 and we employed the top graduates from Ukraine's leading engineering university.
These new software engineers partnered with their more experienced American counter-parts to create software development teams that had a very high level of productivity and cost-effectiveness.
When not employed on particular projects, TopSoft's engineers worked on an internal project called "TopOCR".
This was one of the first OCR applications that supported digital cameras.
It was released in 2005 as a shareware application and has been downloaded over a millions times!
TopOCR (1998 - Today)
TopOCR is today the fastest known OCR engine! It is able to process more than 10 pages per second on a fast CPU!
It is based on an algorithm that converts an auto-scaled character bitmap into a 2D edge map.
This transformation provides a data reduction of 16 to 1!
We developed a feature detector based on lines and "loops" as well as their positions derived from this edge map.
From there a trainable feature detector tree could be created which could quickly arrive at a character prediction based on branching from a small number of character features.
This is the OCR engine we use for scanning with TWAIN flatbed scanners and for reading multi-page PDF files.
The TopOCR OCR engine running on a state-of-the-art 5 ghz Intel CPU is able to read PDF image files at the rate of 600+ pages per minute on a SINGLE CPU!
TopOCR is not just an OCR engine, but also an application that uses three specialized OCR engines and has advanced document image dewarping features and provides an Accessible OCR User Interface.
Nemonika Mobile Browser with Speech Recognition and Synthesis (2007 - 2009)
Nemonika included a sophisticated Voice Recognition and Navigation system powered by an optimized version of the PocketSphinx voice recognition engine.
Audio input and output was through a high quality bluetooth headset.
This project was intended to run on a single core 600 MHZ ARM CPU Samsung smartphone.
Unfortunately, the project was discontinued when Microsoft stopped development of Windows Mobile 6.0.
The Nemonika Browser also included a built-in file browser/viewer with image auto-rotation, super-fast DCT jpg scaling and OCR with Text To Speech output.
TopSearch (2013 - 2017)
TopSearch was a research project aimed at creating an Accessible Search Client with a simple keyboard interface and text to speech.
Guten Browser (2017 - Today)
Guten Browser is an ultrafast, light-weight web browser designed specifically for use with Project Gutenberg which offers over 57,000 free eBooks.
It provides a simple tablet friendly interface focused on users with low vision who want to read on-line ebooks from Project Gutenberg.
SeeHear Object Recognition System (2017 - Today)
SeeHear is a single-button webcam based object identification system that can recognize objects from any webcam with automatic Text To Speech output!
Our next version of SeeHear is nearly finished! It provides 20X faster neural network calculation (and reduced memory storage) with BINARY XNOR neural networks implemented in AVX assembly language running in multiple threads. This will be the basis of a future release that will provide real-time video analytics on low end Windows 10 hardware using the CPU only.
Snap Reader OCR (2012 - 2016)
Snap Reader is a Windows 10 Universal Windows Platform (UWP) Accessible OCR application based on TAO OCR.
It is meant to run on a low cost Windows Tablet PC with a built-in 5.0 MP rear-facing webcam.
SnapReader uses the tablet's internal gyros for 2D optical image stabilization to produce exceptionally sharp images even from a handheld tablet.
Average reading time on a low-end Tablet for a 5.0 MP image is about 5.0 seconds.