INSPECTION
Label accuracy stops unpleasant surprises
by PAN DEMETRAKAKES, Editor-In-Chief
One of the most basic functions of a package is to say what’s inside. It’s important to make sure your packages are telling the truth.
Mislabeled consumer products can have consequences ranging from annoyance to injury or worse, in the case of allergens. As speeds increase and automation removes human eyes from the packaging line, accurate labeling can be more of a challenge than ever. Mislabeling is one of the most common reasons for FDA food recalls.
But automation can hold the key to insuring that the right label, film or carton gets matched up with the right product. Machine vision systems can constitute a vital component in this aspect of quality control. Matching the proper amount of functionality with the application means that end users can get needed protection for the lowest possible price.
Verification applications vary widely in their degree of complexity. In some cases, labels or packaging differ greatly from one stock-keeping unit (SKU) to the next; in other cases, they differ only by small yet crucial elements like an allergen declaration. Other applications involve matching labels to prefilled cans or other packages, which often means reading ink jet codes. This degree of difficulty is one of the key elements in determining how much functionality a vision system needs.
“We have seen over the last several years, and even more recently, a very strong interest in those types of [verification] applications,” says John Agapakis, director of business development for Microscan, a supplier of inspection equipment and software.
At its core, comparing two blocks of text and graphics involves two of the most basic functions of machine vision: optical character recognition (OCR) and optical character verification (OCV). John Lewis, market development manager for Cognex, explains that OCR is a sort of sub-application of OCV.
OCR is “reading what’s printed on the label,” Lewis says. “The machine vision system finds the label and the image, looks for characters and sees the order they’re printed in, and it just reports what it sees....[OCV] is one step further. Once it sees what’s on the label, it compares it with what should be printed on the label.”
Various factors in an application can increase the difficulty of OCR and OCV. One of the biggest is distortion.
“Distortion can result from a variety of causes, such as the printing process itself, the type of surface you’re printing on, whether it’s a metal can lid, or a plastic bottle that’s curved, or some type of flexible film,” Lewis says. “All these types of materials that are printed on can cause skew distortion or other types of distortion during the printing process.”
As distortion increases, the machine-vision system needs extra computing power to accommodate it, sometimes with special tools. One example of these is Cognex’s OCV Max, a software tool that is designed for extra distortion tolerance in applications where characters might have natural flaws.
Sometimes characters are hard to read because they’re simply not well defined. This is especially liable to happen with ink jet or laser codes that bear product identifying information, as well as dates and lot codes. Applications like this are common in canned foods, especially in operations that supply private label goods. These operations often can products unlabeled, as “bright stock,” with labels applied as orders come in. The only way to be sure of what’s in the unlabeled can is to read the coding accurately.
“In one instance, we had a customer mislabel clam chowder as cream of potato soup, and they’re very similar, but shellfish is a major allergen,” Lewis says. “There could have been major repercussions to that.”
Reading characters coded directly on packaging by ink jet and laser printing is more complicated than reading conventional printing on labels or paper. This can get worse if cans go by the printhead too fast. The difference between, say, an “8” and a “B” can come down to a couple of dots, Agapakis says.
There are “very, very small differences in placement of a couple of dots between the ‘8’ and the ‘B’ that still the vision system will identify, if properly trained,” Agapakis says.
Another common verification application is matching discrete elements of packaging together, most often bodies and lids. Again, the biggest factor is how much variation there is in packaging for the various SKUs.
Lewis recounts how an ice-cream packager in New Zealand needed help matching lids to tubs. They had been using a bar code reader, but a redesign removed the bar codes from the lids. An InSight system from Cognex was installed that checked lids by looking at the shape of the product name.
“They don’t even really read the lid,” Lewis says. “They actually just use pattern matching to look for graphics or design elements on the lid and see if they match. It may say vanilla on the side and on the lid, but the vision system doesn’t know it’s V-A-N-I-L-L-A. It’s just looking for the shape of that word.”
Looking for shapes is quicker than reading alphanumeric characters.
“OCV can be very tricky depending on different distortions in the printing,” Lewis says. “It’s a more complex application than matching a pattern in many cases.” Typically, in OCV, the system has to be shown multiple images of each letter as it might appear in various forms of distortion. “It’s a very tedious process to train on images of different letters. With pattern matching, it’s a much easier process, so a lot of people, rather than train on OCV, will just match patterns.” At a demo at last year’s Pack Expo, Cognex showed how its equipment could be used to pattern-recognize variations in the names of different kinds of canned tomatoes-“diced,” “stewed,” “Italian,” and so on.
This is a factor packagers may wish to consider during package design, especially for high-speed packaging lines.
“Most of the applications out there now, in order to maintain anything resembling a decent production line speed, are searching for something that can be applied in the label design that is large enough to read at a fairly good speed going by,” says Mark Traxler, senior marketing communications specialist for Omron Electronics.
Label verification is a crucial aspect of food safety. With the right preparation of both labels and equipment, verification can be done automatically, at high speeds, with great accuracy.
FOR MORE INFORMATION
The following companies contributed to the research of this article:
Cognex Corp.
877-264-6391; www.cognex.com
Microscan
603-577-5865; www.microscan.com
Omron Electronics
866-88-OMRON; www.oeiweb.omron.com
Verifying that the right label goes on the right package is a task increasingly trusted to machine vision.
PHOTO COURTESY OF COGNEX CORP.
One of the most basic functions of a package is to say what’s inside. It’s important to make sure your packages are telling the truth.
Mislabeled consumer products can have consequences ranging from annoyance to injury or worse, in the case of allergens. As speeds increase and automation removes human eyes from the packaging line, accurate labeling can be more of a challenge than ever. Mislabeling is one of the most common reasons for FDA food recalls.
But automation can hold the key to insuring that the right label, film or carton gets matched up with the right product. Machine vision systems can constitute a vital component in this aspect of quality control. Matching the proper amount of functionality with the application means that end users can get needed protection for the lowest possible price.
Verification applications vary widely in their degree of complexity. In some cases, labels or packaging differ greatly from one stock-keeping unit (SKU) to the next; in other cases, they differ only by small yet crucial elements like an allergen declaration. Other applications involve matching labels to prefilled cans or other packages, which often means reading ink jet codes. This degree of difficulty is one of the key elements in determining how much functionality a vision system needs.
“We have seen over the last several years, and even more recently, a very strong interest in those types of [verification] applications,” says John Agapakis, director of business development for Microscan, a supplier of inspection equipment and software.
At its core, comparing two blocks of text and graphics involves two of the most basic functions of machine vision: optical character recognition (OCR) and optical character verification (OCV). John Lewis, market development manager for Cognex, explains that OCR is a sort of sub-application of OCV.
OCR is “reading what’s printed on the label,” Lewis says. “The machine vision system finds the label and the image, looks for characters and sees the order they’re printed in, and it just reports what it sees....[OCV] is one step further. Once it sees what’s on the label, it compares it with what should be printed on the label.”
Various factors in an application can increase the difficulty of OCR and OCV. One of the biggest is distortion.
“Distortion can result from a variety of causes, such as the printing process itself, the type of surface you’re printing on, whether it’s a metal can lid, or a plastic bottle that’s curved, or some type of flexible film,” Lewis says. “All these types of materials that are printed on can cause skew distortion or other types of distortion during the printing process.”
As distortion increases, the machine-vision system needs extra computing power to accommodate it, sometimes with special tools. One example of these is Cognex’s OCV Max, a software tool that is designed for extra distortion tolerance in applications where characters might have natural flaws.
Sometimes characters are hard to read because they’re simply not well defined. This is especially liable to happen with ink jet or laser codes that bear product identifying information, as well as dates and lot codes. Applications like this are common in canned foods, especially in operations that supply private label goods. These operations often can products unlabeled, as “bright stock,” with labels applied as orders come in. The only way to be sure of what’s in the unlabeled can is to read the coding accurately.
“In one instance, we had a customer mislabel clam chowder as cream of potato soup, and they’re very similar, but shellfish is a major allergen,” Lewis says. “There could have been major repercussions to that.”
Reading characters coded directly on packaging by ink jet and laser printing is more complicated than reading conventional printing on labels or paper. This can get worse if cans go by the printhead too fast. The difference between, say, an “8” and a “B” can come down to a couple of dots, Agapakis says.
There are “very, very small differences in placement of a couple of dots between the ‘8’ and the ‘B’ that still the vision system will identify, if properly trained,” Agapakis says.
Another common verification application is matching discrete elements of packaging together, most often bodies and lids. Again, the biggest factor is how much variation there is in packaging for the various SKUs.
Lewis recounts how an ice-cream packager in New Zealand needed help matching lids to tubs. They had been using a bar code reader, but a redesign removed the bar codes from the lids. An InSight system from Cognex was installed that checked lids by looking at the shape of the product name.
“They don’t even really read the lid,” Lewis says. “They actually just use pattern matching to look for graphics or design elements on the lid and see if they match. It may say vanilla on the side and on the lid, but the vision system doesn’t know it’s V-A-N-I-L-L-A. It’s just looking for the shape of that word.”
Looking for shapes is quicker than reading alphanumeric characters.
“OCV can be very tricky depending on different distortions in the printing,” Lewis says. “It’s a more complex application than matching a pattern in many cases.” Typically, in OCV, the system has to be shown multiple images of each letter as it might appear in various forms of distortion. “It’s a very tedious process to train on images of different letters. With pattern matching, it’s a much easier process, so a lot of people, rather than train on OCV, will just match patterns.” At a demo at last year’s Pack Expo, Cognex showed how its equipment could be used to pattern-recognize variations in the names of different kinds of canned tomatoes-“diced,” “stewed,” “Italian,” and so on.
This is a factor packagers may wish to consider during package design, especially for high-speed packaging lines.
“Most of the applications out there now, in order to maintain anything resembling a decent production line speed, are searching for something that can be applied in the label design that is large enough to read at a fairly good speed going by,” says Mark Traxler, senior marketing communications specialist for Omron Electronics.
Label verification is a crucial aspect of food safety. With the right preparation of both labels and equipment, verification can be done automatically, at high speeds, with great accuracy.
FOR MORE INFORMATION
The following companies contributed to the research of this article:
Cognex Corp.
877-264-6391; www.cognex.com
Microscan
603-577-5865; www.microscan.com
Omron Electronics
866-88-OMRON; www.oeiweb.omron.com
Looking for a reprint of this article?
From high-res PDFs to custom plaques, order your copy today!
