Address reading / OCR

Image acquisition

One of the most critical parts of the address recognition chain is image acquisition. SOLYSTIC has designed a high-resolution acquisition system called COPERNIC™, which captures the images of letters, flats or parcels in multi-grey levels (MGL) and in colour. For each of them, several binary images are also generated. All these images (binary, MGL and coloured) are then all sent to the address recognition unit. At the same time, a MGL and a colour image are sent to the image server so that, if necessary, they can be transmitted to the video coding consoles.

Letters, flats, parcels

Handwritten or typed addresses

Horizontal, vertical, upside down or skewed addresses' position
White, coloured or noisy backgrounds

The OCR operates at the same speed as the fastest sorting machines. It locates the address on an object image, then the address block is processed through SOLYSTIC algorithms. As soon as the information required for sorting has been decoded, it is transmitted to the machine in real-time.
The OCR system detects and recognises the following informations:

• position and orientation of the address, 
• recognition of characters and words,
• overall interpretation of the address, location and recognition of any information found on the postal item (ie: barcodes, logos, stamps, etc.)

The OCR can be equipped with a barcode reader (BCR) which is capable of handling up to 50 letters/second. It can also locate a barcode printed by the customer (WABCR), wherever it is on the envelope.

Multi-resolution - Multi-binary coding - Multi-grey levels
Multiple OCR recognition engines - Multiple OCR combinations
High computing capacity
Possibility of integrating solutions at low hardware cost
Algorithmic innovation (new solutions for handwritten cursif characters)
Interaction between OCR and VCS

The combination of results obtained by the different OCRs increases the reading rate, whilst lowering the risk of error. If in doubt, the system turns to VCS to confirm or complete its decision. The advantage of this interaction lies in the use of data shared by the two systems providing an optimized recognition process. The rate of coding is then 30% higher than with standard processing and requires far less operator skill.