MICR CMC7 Recognition Engine
History: MICR CMC7 font was developed in France in late 1950s, for automatic cheque processing. It's currently used in Europe, Mexico, Brazil and in other countries world-wide.
Standard: digits' font print size is 2.85mm, while symbols' size is 2.70mm. Each character is made up by vertical lines divided by five narrow spaces and two larger spaces.
Capability: our MICR CMC7 recognition engine is size tolerant and it's able to read codelines with small size variation. The recognized character set is made up by the following digits and symbols:
According to the picture order, the last symbols are called: S-1 (Internal), S-2 (Terminator), S-3 (Amount) , S-4 (Not Used), S-5 (Routing) and, after recognition, they are respectively reported like "/" "=" "!" ">" "^" in ASCII characters.
Accuracy: with MICR CMC7 recognition engine, the accuracy reaches 100% in case of good quality scanned images; if the quality is low or medium, error rate will usually be less than 0.5%. For each recognized character a confidence level is reported, to let you to appreciate how much accurate has been the recognition process.
Technology: our engine is based on three distinct classifiers using different methods: artificial neural network, features extraction, statistycal matrix matching. An expert system, using fuzzy rules, reconciles all the outputs obtaining the best match.
Speed: speed is linked to cpu speed. Using Pentium processors, time performance are usually included between 50 and 150 CPS.
Input: monochrome images (like memory DIB handles, Device Independent Bitmap) at resolution of 200 or more DPI are accepted in input. Better results can be obtained using grayscale images with our dynamic thresholding SDK.
Output: output is made by recognized characters ASCII codes, by confidence level and by coordinates of each area containing characters.
Platform: all of the Windows 95/98/NT/Millennium/2000/Xp platforms are supported.
Packaging: the engine is packaged in DLL format. The total size is 500 KB. No extra file is required.
Here is a sample of CMC7 codeline image, automatically readable with 100% accuracy:
