Image analysis – Pattern recognition – Classification
Patent
1996-09-10
2000-09-26
Au, Amelia
Image analysis
Pattern recognition
Classification
382309, 706 52, G06K 962
Patent
active
061252085
DESCRIPTION:
BRIEF SUMMARY
BACKGROUND OF THE INVENTION
The invention relates to a device for recognising characters in a document with at least two writing recognition units, an evaluation device for evaluating the results of the writing recognition units with the aid of a database of rules for evaluating the sample recognised by the writing recognition units. The invention also relates to a process for the recognition of characters which makes use of the device.
The recognition of character samples or written symbols is an ever important task in the modern business world. Data processing installations with character recognition units are used, particularly in the banking area and in other finance for reading remittance orders, cheque forms and other documents without the help of manual input.
There are in particular two character or writing recognition methods known in the Federal Republic which are on the market. The Computer Gesellschaft Konstanz mbH (CGK) process is marketed under the type reference "CSL 2610". AEG Elektrocom GmbH Erkennungssysteme has developed the "Polyform" character recognition software. This software is employed in the AEG readers and in equipment from other companies.
The two methods are based on different writing recognition algorithms and they can therefore be described as "orthogonal" to one another. This difference in the algorithm employed leads to the result that the two methods give different results for the same text. This variation in results is particularly noticeable in the case of handwritten text since the characters in such texts are often written very quickly and not clearly.
In order to improve the reliability of the writing recognition, tests have been carried out with a so-called "multi-voting" method where the results of reading of the same characters from several writing recognition units are supplied to an arbitration unit. The arbitration unit, as currently used, compares the results and frequently selects the overall result from the writing recognition process in accordance with the following rules: result is trivial, i.e. the total result will be selected from any desired writing recognition unit. the overall result is "reject" (not recognisable). one of which has a reliability value greater than a fixed threshold value (e.g. 50%), the most reliable value is selected as the overall result. result is "reject" (not recognisable).
The "multi-voting" method only slightly improves the reliability of the writing recognition process compared with the conventional method. However, it requires the use of at least two writing recognition units which each use a different recognition algorithm. This is considerably more costly in terms of licence fees and the cost-benefit relationship is dubious.
SUMMARY OF THE INVENTION
The object of the invention is therefore to develop an improved writing recognition method which derives more benefit from the use of several writing recognition units.
This object is achieved in accordance with the invention through a calculating means for calculating a confidence level for each of the samples recognised by the writing recognition units and a sample vector memory component for storing the confidence level, where the credibilities are accessible to the evaluating means in the operating condition. With the aid of the calculated credibilities, the principles of "fuzzy logic" or the teachings of uncertain quantities are employed for evaluating the results of the plurality of writing recognition units. Simulations have shown that the reliability of the writing recognition method in accordance with the teaching of the invention is improved by 57% compared with the known conventional "multi-voting" method.
In a further embodiment of the invention the device contains a degree of fulfilment table for storing the results of the rules and the degree to which they are met. The degree of fulfilment is used in order to select the result of the rule which most satisfactorily represents the result of the writing recognition process. An output means for searching the degre
REFERENCES:
patent: 5189728 (1993-02-01), Yamakawa et al.
patent: 5257323 (1993-10-01), Melen et al.
patent: 5455872 (1995-10-01), Bradley
patent: 5544279 (1996-08-01), Li et al.
Proc. Tencon '93 IEEE Region 10 Conf. on Computer, Communication, Control and Power Engineering, vol. 2 of 5, Oct. 19-21, 1993, Beijing.
Using Fuzy Logic, Jun Yan et al. pp. 54-55, 1994.
C. Hong et al, "Hierarchical Control Structure for Multisensor Object Recongition System", IEEE Tencon'93 pp. 811-814.
J.M. Keller et al, "Information fusion via fuzzy logic in automatic target recognition", Proc. of the SPIE--The International Soc. for Optical Engineering., 20th AIPR Workshop 1992, vol. 1623 pp. 203-208.
Maier Udo
Ruppert Werner
Ahmed Samir
Au Amelia
International Business Machines Corp.
LandOfFree
Writing recognition unit does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Writing recognition unit, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Writing recognition unit will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2107139