Bayer filter versus three-chip

At the risk of turning this in to ColorMachineVision4Users I want to continue the Bayer versus three-chip camera comparison. Or rather, I want to let camera-maker JAI have the last word. This video, lifted from YouTube, provides an excellent summary and should appeal to those of you who don’t enjoy reading.
 

 

If you found that useful, may I suggest you look at some of the other camera videos on the JAI website. They’re informative and well-produced, and best of all, they’re not just promotional glitz.

A different approach to color imaging

Machine vision pros know it’s not always necessary to use a color camera for a color application. Often all that’s needed is to either match the wavelength of the illumination source to the target. Like colors lighten, so a red light will make a red target appear white(ish) to the camera. An alternative is to put a filter over the lens, so that only light of the target wavelength is allowed through.

But what if you need to look at several targets in several different wavelengths?

What you need then is a liquid crystal tunable filter. A product like VariSpec might do what you want, and something similar is available from Inno-Spec. And if you want to understand just how they might be used, I suggest you read “Tell-Tale Color Changes: Camera Can Find Age of a Bruise” on the BioPhotonics website, October 2012.

This fascinating article describes some important color imaging work going on in the medical field. I’m not going to steal their page views by telling you about it: click the link and read for yourself.

And last, if you want to understand more about how a liquid crystal tunable lens actually works, take a look at “Liquid Crystal Tunable Filters” on the Olympus Microscopy Resource Center site.

The argument for prism-based color cameras

There are three options when it comes to color linescan: a form of Bayer filter over the sensor, the trilinear sensor (red, blue and green side-by-side,) or the prism. The Bayer is clearly a budget approach but how do you compare trilinear and prism?

Well you could take a look at an animated movie put together by Danish camera-maker JAI.  This shows the basic operating principles of each and summarizes the weaknesses of the trilinear method. It seems pretty logical to me, although it might be worth noting that JAI have spent a lot of money on a prism-sensor assembly process. (Their NIR-visible cameras are a clever use of the technology.) But perhaps they did that because the prism approach is more robust.

How do I solve my imaging problem without a color camera?

When I wrote “Stay away from color!” it occurred to me that rather than just being entirely negative, I should try to offer some helpful advice, so here it is.

I believe that most color machine applications are really monochrome apps that just need optimized lighting and/or appropriate lens filters. Unless you actually need to differentiate between two close shades, lighting and filters can be used to selectively make objects in your image either dark or light. Let me give you a few pointers before sending you off to a web site I rate highly.

Pointer 1: match the wavelength of the light to the color of the target.
The rule to remember is “like colors lighten.” This means that if you shine a red light on a red target a lot of light will reflect back to a monochrome camera. Thus the red object will appear white in the image. Conversely, when red light falls on a green object very little will be reflected and the object will appear dark in the image.

Pointer 2: use a lens filter to determine what wavelengths reach the camera. Typically you’d use a bandpass filter to ensure detection of a narrow – maybe 20nm – spread of wavelengths. This way you can illuminate the object with white but set the camera up so that only say blue light reaches the sensor. As a side note, I like to use the filtering approach to cut out any ambient light that might find its way in to my system.

So where do you go for more information? Well for basics of using different colors of lighting, check out Advanced Illumination, and for filters there is only one place to go: Midwest Optical Systems Inc.

Stay away from color!

Those new to machine vision are often tempted to reach for a color camera to solve their inspection problems. This is probably because most of us see in color, but I don’t think it helps that the camera companies like to trumpet the arrival of every new color product from the rooftops. This just encourages the habit of grabbing the color cam rather than the monochrome. (Yes PPT, I am talking to you! “PPT VISION introduces twenty-six new IMPACT M-Series color cameras… ”)

So I’m here to tell you to avoid color at all costs. Color machine vision is really challenging, for more than a few reasons. Perhaps the biggest is that our human perception of color correlates poorly with the actual wavelength of reflected light. The color of a surface is also heavily dependent on it’s texture, it’s orientation to the incident light, and of course the spectrum of the light used for illumination. (And please don’t try to do color vision with a red LED lighting – yes I have seen it attempted!)

Then there are all the problems with processing color images, plus the loss in resolution: is the edge at the green pixel or the red one? Color is just a real headache and should only be attempted by those with years of experience. In fact to support this point, note that the AIA’s Certified Vision Professional program deals with color in the Advanced category!

Have I put you off or are you still leaning towards a color camera? If you are, then read “Color Vision: Luck Has Nothing to Do With It” by Winn Hardin and published on the AIA web site, March 7th, 2011. This is a good article that goes into more detail than I have time for. If you really want to work with color at least go into it with your eyes open.

Color camera technology

CCD and CMOS sensors put out electrons in response to photons that land on their silicon surface. They have no knowledge of the wavelength of the light that fell on them. So how, you might be wondering, does the camera know when to put out red, green or blue?

Well most color cameras use a Bayer filter over the CCD sensor. This red, green and blue mosaic filters photons by color so that some pixels only receive red, some blue and the rest green. In fact I thought that was the way all color cameras did things, but thanks to an article on the Basler web site, I now know better.

“Color Creation with Interlaced Sensors – How Does That Work?” describes how the Sony ICX409 sensor creates color images. What’s interesting is that it uses a four color filter – green, magenta (pink), cyan (blue) and yellow – and a complex binning process to produce color images. There’s some good detail in the article, so if this is something that interests you, take a look.

Why are machine vision people color-shy?

To those who know little to nothing about machine vision – like the average Operations V.P. - it seems obvious that vision systems should use color. After all, we see in color, the products we make are colored, so why not?

Well Ben Dawson of Dalsa does a good job of summarizing some of the challenges in “Color Machine Vision – Untouched by Human Eyes,” published in “rtcmagazine” back in January 2010. I’m not going to plagiarize Ben’s good work by telling you all the secrets here; you’ll have to click the link to learn the details. But as you read you’ll also learn about a family of Dalsa-based systems for inspecting baked products – muffins and the like.

These are built by Montrose Technologies Inc. out of Ottawa in Canada, and it seems they’ve found themselves a great niche for replicating inspection systems. I see no reason why their 3D, color inspection machines should be restricted to bakeries though, so if you need fast inspection of product placed randomly across a conveyor it might be worth firing off an email to these boys.

A 4-sensor line scan camera

I hear through the grapevine that JAI are launching a new color line scan camera. Unlike existing models of color line scan, which use either a bayer filter over the CCD or 3 sensors with individual color filters, this uses JAI’s prism technology. The major benefit is the superior color registration that results from a common optical path. This eliminates the alignment problems that make color line scan cameras such a pain to use.

Incidentally, the camera is available in only a 2k format, but the CCD sensors cover Red, Blue, Green and Near IR. This should make it useful for inspecting organic materials where IR can help uncover more data than is evident to the human eye. For an example of where this can be useful, take a look at this page on the JAI site about the similar multi-spectral AD-080 camera.

As for the LQ-200CL, which is what they’re calling this new camera, at the time of writing there are no details on the JAI site, but I’m sure that will change soon. Also absent is any pricing information. It won’t be cheap, but judging by the pricing of the AD-080 range I would expect this to be around $3,500.

And this is a 100 level class?

For my non-US readers, let me start with an explanation: Universities in the US rank their academic classes by three digit number where the leading digit signifies the difficulty of the class. Thus a 100 level class is generally taken by freshmen (first year students,) while a 500 level class is typically only encountered while studying for a Master’s degree. One consequence is that any introductory level is generally described as being “one-oh-one” as in Economics 101, or perhaps Machine Vision 101.

I start with this preamble because those of you who receive the print version of Quality Magazine, with the Vision & Sensors supplement will find, under the heading of “Machine Vision 101” in the May 2010 edition, an article titled, “Color Goes Mainstream.” (If you just use the web link you’ll notice the “101” subheading is absent.)

The article, penned by Henning Tiarks of Basler, provides a detailed discussion of the various flavors of color line scan, and in particular, addresses the challenges of aligning tri-linear color line scan cameras.

It’s a good article, and if you really need to acquire color line scan images, it should be required reading, but it is most definitely not 101 level. Line scan machine vision is complicated, and color line scan imaging is really complicated. Do not attempt this for your first machine vision project.

Color machine vision – tread carefully!

Color is a difficult beast to grapple with. My advice is to steer well clear unless you have absolutely no choice. What makes it so difficult? Well for a thorough overview of the challenges, I recommend you spend a few minutes reading, “Color Machine Vision – Untouched by Human Eyes,” penned by Ben Dawson of Dalsa and published in RTC magazine, January 2010.

In a very readable article, Ben explores why it’s so hard to be objective about color, the importance of illumination (white light does not contain equal helpings of all wavelengths,) and how a Bayer filter impacts camera resolution.

And if, after reading the piece, you’re wondering how to perform a color inspection task with a monochrome camera, well I have two words for you: filters and illumination wavelength. (OK, so that’s three words, but I’d like to count “illumination wavelength” as a single word since it is a single concept, which is probably …)

Why does machine vision prefer monochrome cameras?

Unless you’re actually doing something that requires color – matching shades for example – you’re better off using a monochrome camera. As we discussed previously, this is because monochrome provides higher resolution, and thus sharper edges.

And for most machine vision tools its edge sharpness that matters. Practically every vision tool uses edges in some way, whether to apply a caliper, to locate edges as a reference location or even in blob detection (yes, blobs have edges.)

So while the human eye would rather see in color, software is happier, and often more successful, when dealing with black and white.

There’s no such thing as a color camera

If you’re familiar with how a CCD or CMOS sensor detects light you’ll agree with my title statement. Our silicon-based sensors develop electrical charge in response to the arrival of photons of light. Photon arrives, charge generated. The silicon has no knowledge of the wavelength of the incoming light.

So how do we get color images, and are there benefits in staying with monochrome? Rather than fudge an answer, let me refer you to a fascinating paper by Francois Marçeau of Clemex Technologies entitled “Image Analysis in Black & White,” (Advanced Imaging, May 2009.) Francois does a terrific job of explaining why monochrome can give better fidelity to the actual scene being imaged than can color. His paper should be required reading for anyone who believes they need a color machine vision system.

Even more on color

Don’t make the mistake of thinking you need a color camera to inspect a colored target. It’s often the case that you can create appropriate contrast by matching the color of the light to that of the target.

Last month, in “More on color,” I mentioned that you can lighten a surface by illuminating it with light of the same color, while an opposite color (refer back to the color wheel to find the opposite,) will darken the target.

Well, courtesy of LED lighting specialists CCS Inc., here’s an illustration of what I mean.

And when you’ve finished browsing that particular page I would encourage you to look at the test of their Technical Guide. It has some good information about using LED lighting.

An insight into product planning at Cognex?

How many color machine vision jobs have you ever done? I’m guessing it’s a very small number. Color images are more complicated to work with, and quite frankly, there are few applications that need them. Even in cases where you need to differentiate between say red and green, you can get away with combinations of colored lights and lens filters. In short, I’m not sure the world is hungry for new color “smart cameras.”

So why have Cognex launched three different color In-Sight Micros? Yes, it will give some marginal sales growth, but at the cost of further complicating their inventory management processes. I’m just not convinced the market for color is big enough to warrant the increased business complexity.

I suggest there are some other motives at work. I think the real reason for the new cameras is that Cognex are incorporating some of the DVT color algorithms into the Explorer software. In short, this is all part of phasing out Framework, Intellect and the 500 series range of cameras.

Think about it: a range of color In-Sight Micros is one more reason not to offer the old DVT products. If I worked for Cognex I’d probably see that as a sensible decision. Color Micros expand the product range now but will ultimately permit some significant pruning, and presumably some cost reductions too.

I know the fans of scripting will be sobbing into their coffee once more, but it’s just business. Adapt to the spreadsheet or find an alternative product.

By-the-way, the specs on the new cameras are quite interesting, especially the 2 Mp color Micro. It won’t be cheap but it could be a great solution to a problem that absolutely must have a color camera.

Color application, and a bonus

From the pages of “Test & Measurement World,” (T&MW) comes an interesting report on a color machine vision system developed by EPICVision Solutions.

Now I don’t know about you, but I try to stay away from color. It’s not that I prefer a drab, monochrome world, but frankly, color gets so damn complicated to set up and maintain. My approach is to do as much as possible with colored lights and filters, leaving color cameras as a tool of last resort.

But, there are times when nothing else will do, so it’s prudent to be aware how others have approached these challenges. After reading the article linked above you might like to turn to The Imaging Source and review their white paper on “How Color Cameras Work” (opens as a pdf.)

And the bonus? I’d like to share this delightful pun from the opening paragraph of the T&MW article, “One color application that is gaining ground is “color matching,” according to … EPICVision Solutions, a St. Louis integrator that designs machine-vision systems for blue-chip customers …”

I wonder if those blue-chip customers have been color matched?

“Everything looks worse in black and white”

My May edition of Evaluation Engineering is hot off the press, and as always I turned straight to the machine vision article. This is an interesting piece by Robert Howison of Dalsa about the value of color in machine vision. It provides a reasonable overview of the issues but, to be quite frank, I thought it needed more detail.

The main point Mr. Howison is trying to make is that color is often not necessary in a machine vision system, which I agree with, but I think more explanation would have been helpful.

First, we need to remember that the CCD or CMOS sensor does not sense wavelength; it just catches photons of light that pass through the lens of the system. So all its detecting is the quantity of light.

Second, a filter will let us trap photons of certain wavelengths, so preventing them from reaching our sensor. For example, if we use a green filter we are allowing photons with wavelengths in the region of 500 to 570 nm to pass through, while absorbing all other wavelengths. This is how we detect green light.

So, it doesn’t take too much imagination to see that combinations of colored light and colored filters can be used to help a monochrome camera detect light of a specific wavelength. Thus the complexity of color is often unnecessary. (I should add that color image processing can get really ‘hairy’, but any further discussion is beyond the scope of this blog.)

It’s possible that everything actually looks better in black and white, (which is what Paul Simon sometimes sings, as opposed to the title of this piece, which is what he wrote.)

Enjoy the article, but treat it as an appetizer.