Neurosoft has been conducting research in the field of image processing since the very moment of establishing the company. This research has resulted in modern technologies connected with Optical Character Recognition (OCR) and unique methods of electronic publication of documents.
Optical Character Recognition
Neurosoft has prepared a series of laboratory systems for printed character recognition and handwriting recognition, as well as commercial software to a variety of specific applications. The OCR module called PRP, sold in a specialized AutoLektor device designed for blind people to read books and newspapers, is an example of such software.
In terms of Optical Character Recognition technology Neurosoft uses pattern matching, omnifont and its own neural network technology called spatiognitron (Spatiotemporal and hierarchical artificial neural networks based on the Time Delay Neural Network concept).
The experience gained while working on the OCR has been used to develop algorithms which recognize the content of the number plates of vehicles (Automatic Number Plate Recognition). The ANPR algorithm consists of two basic parts:
- detection of areas containing number plates,
- recognition of the content of the number plates.
Neurosoft has developed an effective algorithm to detect number plates which is to a great extent independent of confounding factors such as e.g. lighting conditions, contrast, angle of torsion.
An innovative recognition algorithm combined with the use of a set of neural network dedicated to different countries and the use of the syntactic analysis enables the recognition of not only the content of the number plates but also the country of their origin. The use of the neural network of the spatiognitron type eliminates the need of pre-segmentation of the image into individual characters which significantly improves the recognizability of highly distorted images.
The Make and Model Recognition system enables automatic description of the vehicle concerning its type (passenger car, van, truck, bus), make (e.g. Fiat, Volvo, Scania) and model (e.g. Ford Focus, Mercedes Atego, Volkswagen Transporter). The subject to the analysis is the image of the front of the vehicle, that is the outline of the bonnet, lights, grille, bumper etc. The image is converted in a complex, nonlinear process of parameterisation into a descriptor, which is compared with the reference database containing patterns of several thousand different vehicles. In the current database there are 1043 different vehicle models and the number of the vehicles collected exceeds 2500 (taking into consideration the model versions).
The ANPR and MMR algorithms are used in a wide range of products. Depending on the application, they run in real time (the sequence of images from a camera) or in the single image recognition mode.
Electronic publication of documents
In 2001 we conducted the analysis of the solutions for electronic publication of documents available in the market. None of the available formats provided the functionality which was necessary for Neurosoft to realize their projects. As a result, we decided to develop our own technology which would enable publication of electronic documents via the Internet.
Neurosoft prepared a special multi-layered format of the document’s image record (Neurosoft BIP™) and a number of procedures enabling electronic publication of documents. NeuroBook, as it is called, is a logical continuation of the development of the eBook technology. The NeuroBook technology combines the advantages of eBook and WebBook and provides a range of additional features.