Cost-effective computer-based recognition of ink jet coding finally a reality. (Application Area).
Many companies have been forced to implement expensive, customized OCR solutions specifically designed to accommodate inconsistent readings. Additionally, as print conditions change, these companies were compelled to pay their suppliers to modify or enhance the original OCR systems so that they might once again produce accurate reading results.
Ink Jet Obstacles
Certain conditions must be identified and accommodated in order for an OCR system to provide reliable recognition of the kinds of characters that are typically produced by industrial ink jet printers. Industrial OCR systems work best when the characters they are presented with are of a known font and size, exhibit good contrast with respect to their backgrounds, and are consistently printed so that they are upright, along the same line, well formed (non-distorted), and have a fixed amount of spacing between them. Ink jet printed characters are, as a class, relatively difficult to read. Because font and contrast are often the only two printing attributes that companies are willing and able to maintain reasonable control over, OCR systems are required to achieve accurate character recognition, and to read character strings in less than ideal conditions.
A reader combining illumination that produces sufficient contrast between characters and their backgrounds will generally provide acceptable levels of performance on ink jet printed characters. To achieve the best possible performance, the information to be read should be printed in a standard OCR font. Specifically designed to be read by machines, variations of OCR fonts have been given additional features to maximize the differences between them and all other characters that are very similar in appearance when printed in a non-OCR font.
Extensive online testing and analysis has revealed that to achieve robust recognition of ink jet printed characters, an OCR system must, as a minimum, address and offer solutions to common problems. Below are the eight major reasons why the automatic reading of ink jet printing on products and packaging has proven to be so problematic.
1) The widths of characters may vary.
2) The positions of the top or bottom rows of characters may vary, causing changes in character heights.
3) General variations in dot placement may alter the overall shapes of characters.
4) The spacing between characters may vary.
5) Characters may print at an angle.
6) Ink may smear or spatter, causing characters to become connected or degraded.
7) When characters print close together and lean to one side, their leading/trailing edges may overlap the trailing/leading edges of adjacent characters.
8) In multi-line marking applications that feature tight line spacing, the dots from the bottom row of one line may move down and those from the top row of the line beneath it move up to essentially eliminate vertical separation between lines.
The Problem At Hand
Examining how one company in the pharmaceutical packaging industry, MGS Machine Corporation, was able to successfully address the challenges presented by ink jet printing, we can see that cost-effective, computer-based recognition of ink jet printed characters is finally a reality.
The MGS project illustrates how common ink jet printing related problems were cost-effectively addressed by implementing Concepts in Computing's (CIC) low-cost industrial OCR library/engine. This project involved reading the date codes, lot codes, and replacement part numbers on ink jet printed, three line format labels selected by a major Midwestern manufacturer of pharmaceutical products for use on a new packaging line.
While the labels used were relatively "clean" and exhibited good contrast, over the course of long runs, the dots that constitute each 7 x 5 format dot matrix font character were observed to display the kind of variation in placement that is characteristic of industrial ink jet printing. This random movement of the dots resulted in characters with altered shapes and varying sizes. In addition, it sometimes caused the loss of vertical separation between adjacent lines. As the speed of the manufacturing line was increased, the characters began printing at a slant such that the right edge of one character would overlap the left edge of the one beside it to varying degrees. Another effect that became prevalent as the line speed increased was the bottom rows of dots developing a wavy or rolling form. This translated into characters having varying aspect ratios. This also resulted in information that would sometimes print at a 3-degree upward rotation.
These variations in printing caused unacceptable low levels of reading accuracy. Eighty-five percent, initially, was the highest level that could be achieved. When the additional effects described above became factors, even after retraining of characters was done to improve the training data, reading accuracy continued to decline. At that point, the decision was made to use another font that the print head manufacturer deemed to be more "OCR friendly."
However, after concluding that varying the font was far too likely to result in similar character confusion and incorrect recognition, MGS decided to continue to use the original 7 x 5 font, but modify the ink jet printer's software.
Having determined that an OCR system was necessary in order to provide the requisite level of performance on ink jet printed characters, MGS installed EconoCR version 5 from Concepts In Computing (CIC). Featured in the DVT Corporation reader for this particular project, CIC's EconoCR software offered the accurate and cost-effective solution MGS was looking for.
During initial testing, one instance of each character in the 7 x 5 font was taught. After a printing run of approximately 1,100 labels, the acceptance rate was 100%. A second run of another 2,500 randomly selected labels yielded an acceptance rate of 99.4%, with no rejections due to failed or misread characters. In fact, all failed reads were the result of positioning errors or some other non-OCR cause.
As a result, MGS Corporation's desired level of character reading performance was finally achieved and maintained.
Concepts In Computing, 114 Prairie Hill Rd., South Beloit, IL 61080.
Write In 160--Or Reply Online
MGS Machine Corporation, 9900 85th Ave. North, Maple Grove, MN 55369.
Write In 161--Or Reply Online
DVT Corporation, 1670 Oakbrook Dr. NW, Norcross, GA 30093.
Write In 162--Or Reply Online
|Printer friendly Cite/link Email Feedback|
|Date:||Jan 1, 2003|
|Previous Article:||The electronic common technical document: transition to a new paradigm in regulatory submissions. (Application Area).|
|Next Article:||Sartorius North America, Inc. (15th Annual Profiles).|