Perceptech's focus is on computer vision research, consultancy, and prototype development. We are active in high-speed sorting of security papers, and in the area of high profile CCTV. Perceptech has extensive experience with inspection vision applications, where reliability and real time processing is mandatory.

With twenty years of leading computer vision research, and with large experience in applying this research to challenging industrial problems, Perceptech aims to bring the frontier of computer vision science to effective industrial prototypes.

Brain2Scan Banknote Soiling Lab-tool

Perceptech is proud to announce its lab-tool for banknote soiling fitness determination, the Brain2Scan. The tool is able to determine the soiling level for banknotes, derived from a learning set of fit and unfit (soiled) notes. The tool uses a neural network like approach to learn the visual features distinguishing the unfit set from the fit banknotes. As such, Brain2Scan mimics the judgment of your quality control team.

Brain2Scan can easily learn new currencies, automatically performs denomination and facing, and is very easy to operate. The Brain2Scan fitness determination operates in the visual wavelength range, exploiting the full RGB color information from a high resolution, double sides scanner with sheet feeder with a capacity of approximately 20 banknotes. Processing speed is up to 15 banknotes per minute. Brain2Scan is pre-trained for euro banknotes. It can be easily trained for other currencies by the user.

The Brain2Scan consists of a portable scanner and software, which can be easily installed on a windows laptop. Brain2Scan makes use of a high quality and widely available consumer scanner (Canon P-150), thereby allowing Brain2Scan to be very competitive in price. Brain2Scan is considered to be the ideal tool for central banks to establish banknote soil fitness based on human perception, either on location or in the lab.

Visual Banknote Fitness Sorting for Soiling

Perceptech together with the University of Amsterdam has developed machine learning software for visual sorting of banknotes based on soiling in support of central bank's recirculation frameworks. The prototype software has demonstrated itself to closely follow the ECB guidelines for false fit rates, while performing excellently when regarding the false unfit rate (unnecessary shredder rate). The software uses a machine learning method for banknote soiling determination. The method has been validated on a set of over 8,000 banknotes from the Eurosystem, while being learned on only 300 banknotes per denomination. Ongoing is the construction of a prototype on operational banknote sorting machines.

See the report: "Learning Banknote Fitness for Sorting".