Slashdot Mirror
Using Commodity Hardware in Laboratories?
PhysicsTom asks: "I am a Senior Physics student who's final year project is based upon using common, easily available technology to replace parts of the aparatus used in various departmental labs. Currently, my main area of interest is trying to integrate certain computer peripherals (such as scanners and digital cameras) into experiments at an earlier stage, so that images gained from the experiments (such as difraction patterns, etc) can be analysed in a program such as MathCAD straight off, rather than the much less efficient methods we're using at the moment. The problem is that I am having trouble finding out about the way in which scanners and digital cameras work, and how this would affect their accuracy with respect to what I am aiming to do." Basically, how do the various hardware aspects of such devices affect their ability to accurately measure or scan the subject of the experiment?
"The information I am looking for includes things like: the resolution of their grey-scales, what degree of accuracy the motor steps at, how uniformly distributed the CCDs are in the arrays, and other issues that might affect accuracy. Just so that I can know how close to the 'real' picture what I get out of the scanner/camera is. If anyone can tell me all these boring facts for any suchequipment (preferably solutions currently available in the UK) then I would be very appreciative."
The grey scale should be in the documentation. CCD density can be calculated by width of scanner bar divided by number of CCD's. As for the rest, you may have to track down the manufacturers engineers. An alternative is to take apart a scanner and find out who manufactures the components, and contact them. Good luck.
Best Slashdot Co
And the answer is... You can't depend on it. You can't even depend on one camera being identical in specs to another. These devices are made for the consumer market and aren't meant for scientific use.
This doesn't mean you can't use them, though. What it does mean is that you'll need to select something you're pretty sure can handle what you want, and then devise procedures for calibrating the devices' output.
Have you tried contacting the manufactuer of the products you are using? I don't mean as joe student, I mean as a represenative of your university. You might be able to get more then just the specs to the equipment. You might be able to get donations.
...I learned how to make certain things out of easily-available common household items, but I doubt you could put any of them in a thesis.
- A.P.
"Remember when the U.S. had a drug problem, and then we declared a War On Drugs, and now you can't buy drugs anymore?"
If you're relying on your equipment to give you reproducable results, you're going to have measure what it's actually capable of and not rely on published specifications.
I stopped relying on spec sheets when I discovered they weren't very accurate. I've seen variances as high as 50% off spec.
I'm sorry to inform you but this information is illegal to discuss as it would enable you to use this device as you see fit.
Please continue to use the equipment as the manufacturer intended but please refrain from learning anything about it or using it for actual work.
Your friends in peace
USA
..which just shows that the human brain is ill-adapted for thinking and was probably designed for cooling the blood-T P
Keep that scanner glass clean and fresh!
Basically, how do the various hardware aspects of such devices [scanners and digital cameras] affect their ability to accurately measure or scan the subject of the experiment?
Well, it would all depend on the time quanta that you are measuring. Digital scanners for anything having velocity are right out the window, since it takes a notable amount of time to capture the image. Digital cameras are somewhat faster, but it would depend on the quality of the camera if you wanted to track moving objects at higher 'shutter speeds' and resolutions.
If a regular camera could capture the data you are collecting - and it seems that this is the case - the digital cameras should be fine. The important issue is that higher resolutions take longer to fix the image. Finding a happy medium between image resolution and image capture is what you're looking for. You might be able to get those specs from the manufacturer(?).
Everything that you are requesting is component dependent and is most likely availeable from manufacturers. This is the best source ( most dependable) of the info you seek
Have you considered just calibrating the equipment? You'll probably need this anyway since, even if you can get the specs, they'll be expressed as ranges and individual components can fall anywhere within the range (as well as changing physically over the life of the equipment). This is true of your custom hardware as well.
If you want to get an idea of how the equipment performs before you buy, just bring your test images and a laptop into the store and ask to try the demo model.
Talk to some of the researchers in your lab. They probably already have tests as well as software that will compensate for irregularities in a CCD based on the results of the calibration.
It is tempting, if the only tool you have is a hammer, to treat everything as if it were a nail. - Abraham Maslow
Checksumming a binary image may not be possible, although this works for other forms of expected output (like from printer drivers). But there must be some loss-analysis tools for images which could help you get a feel for the accuracy and deviation of any given device.
You can probably get some specifications from manufacturers, but I wouldn't really rely on them (you may for example get 8-bit grey scale, but it may be a non-linear response). WHy not calibrate it yourself: you have got the equipment. Also different models may vary slightly (some manufacturers will change components without telling you). Pay a bit more for good quality equipment, and bear in mind that for example the quality of a ccd camera depends a lot on the lens you use, and it may be better to buy greyscale equipment rather than use colour eqiupment in grey modes.
Also for image capture avoid anything that adds software artefacts (especially compression). firewire uncompressed cameras (we get ours from www.unibrain.gr, very good) are good for high framerate high res, with good Linux support.
It would be a good lesson in the real world - like the old aphorism
Envy my 5 digit Slashdot User ID!
It has been said before, but I'll say it again: test and calibrate.
Get a resolution bar chart. Any decently stocked camera store can help.
Vary it's location systematically. Lather, rinse, repeat.
Come back and do it again tomorrow, until you are happy that performance is not changing, or is predictable.
Calibrate before and after your measurments.
Check the capabilities of the driver software. You want to be able to set the actual parameters the scanner uses, not just "Color photo", "B&W Drawing", etc.
Use no compression or interpolation if you can avoid it. Again, the capabilities of the driver are important.
you call yourself a physics student, i wonder where? even the 100 level freshmen know how these puppies work here at caltech.
Anyway: This site on film scanners talks specifically about film scanners, but also about the technology associated with them. I also really liked the discussion on ink jet printers (which I knew nothing about). Good luck!
I think its a great idea to use cheap commercial hardware for shoestring budget experiments. But be aware that scientific grade measurement hardware are expensive for a reason, and you get what you pay for. For instance, I bought a cheap $200-$300 CCD camera for recording images from an optical microscope. It works fine if I want just pictures of what I'm seeing with my eyes, but as a measurement device, its terrible. The camera is color, but uses an array of filters such that every group of four pixels has two that record green, one red and one blue. An algorithm is then used to generate resolution and color for the image that is captured. This means that knowing the CCD pixel size doesn't mean you can take measurements in your image by simple counting pixels. You have to calibrate the system.
The same goes for your scanner. There are a ton of problems you will run accross when you try it, so just make sure that you compare the results you get in your traditional meaurement aproach with what you get using the scanner.
Good luck,
JD
hmm, sounds to me like you've got a very noble project on your hands; use technology to assist the scientific process. However, I think you're going to have a bit of trouble finding those kinds of deep technical specs for hardware devices that you may be re-purposing beyond their intended usage... I remember many a HP Chromograph system that had to be completely replaced (or radically upgraded) at great expense when a simple hardware hack could have provided a strap-on solution to the problem. In particular I'm recalling a heating/ventilation issue that some ingenious soul had resolved by building a simple hot-box inside the machine with a peltier, thermistor and some PICs to maintain even sample temperature for some device. HP's solution was a whole new piece of kit that cost more than a car, but someone rev-engineered it and saved some serious beer-money for the lab. HP, I'm sure, doesn't approve of these things (I'm not knocking HP, it's a corporate survival instinct). Now you may have the best intentions in the world, but I find it pretty hard to believe that the manufacturers are going to go out of their way to make this info available to you, so a little research is probably in order.
a) See if anyone in the open-source community is working on projects utilizing this hardware. Heck, see if you can find some uber-geek who's been involved in creating linux drivers for a bunch of scanners; maybe you'll find someone who has a great storehouse of eclectic ccd-centric knowledge they would be glad to dump on you.
b) Never underestimate the power of documentation. As I always was told, "documentation is like sex, even when it's bad, it's better than nothing..." Maybe they've listed component manufacturers in the crufty stuff they pack in the back of some of the user manuals.
c) Love your service technician. If you're working with equipment that requires outside support, get friendly with him/her and see if you can't wheedle yourself a set of old support/repair documentation. Most of these people are wage-slaves like us and may well be interested in your little projects.
Beyond that, keep your nose to the grindstone and good luck. Let us know how it goes.
"If I wanted your input on my pet project, I'd stick my hand up your ass and use you like a sock-puppet." - Muse
The best thing to do, when possible, is to do the measurements yourself. That way you know exactly what *your* device is capable of doing, and not the *average* device from the manufacturer. You shouldn't rely on manufacturer's spec sheets for this type of information.
For example, you can get a quick idea of the bit depth of a CCD by measuring the noise floor of the output of a null signal and compare it to the output of a saturated signal. You will find that most *consumer* or *security* CCD cameras will not give you a full 8-bits. Even scientific CCDs which state that they give a full 8-bits are only under certain conditions with a specific type of average or weighted measurement. Don't trust the spec sheet. Measure to make sure!
Then of course you could also use your head. How uniform are CCD arrays (spatially)? Think about how they are made. They are very uniform.
Finally, you should talk to your final project advisor. What you are doing isn't Physics, it's Engineering. Sure engineering is part of experimental science, but shouldn't be the prime focus of a "Physics" project, IMO (was a Physics undergrad myself).
Considering that even a great scanner can't handle a half-tone image without producing an interference pattern, this would be hard to judge. I suppose one factor would be the density of the images being scanned. Photos on paper have something like 32k pixels/inch, meaning there are no diffraction gratings inherently built in to the image. (Whatever you do don't scan a magazine image! A half-tone image is the result of the process used to print photos on magazine/newsprint so that they are economically and technically feasible. The image is masked to convert it to a series of dots at a frequency of perhaps 85 dots-per-inch or usually better for magazines. This clarifies the image by preventing inks from bleeding together.) Whatever the case it would be very difficult to devise an algorithm to for instance compare digital-camera images vs. something from another medium -- I am assuming that there would always be a need to compare data to external sources or other external experiments. One of the first obstacles would be to normalize all the images to reduce any equipment-introduced artifacts, such as the interference I am referencing above. Considering that there could be a wide-base of possible inputs, I believe that this technique will only work effectively if all the inputs can be controlled (such as specifying ONLY photo prints be scanned) and providing some sort of algorithm to for instance calibrate all different scanners in relation to one another. There is a name for the interference pattern produced by scanning images -- It is called a "Moire" pattern. Incidentally, all this is said without even talking about the mathematics or the meaning of how little blotches of white in an image can be falsely interpreted by a scanner to form a diffraction grating!
I prefer it this way: Theory and reality rarely coincide.
There was an article on a very similar subject in New Scientist a few weeks back.. Lemme see if I can get a URL....{time passes} Oh dear, it's in the archive and you'll need to register to see it. And registration requires a subscription to the magazine.. how very lame.
Anyway, the upshot was that a research group was using consumer-type Digital Cameras to help automate surveys of Rain Forest flora.. turns out that their estimates were VERY BADLY off because the prime descriminator used was colour (think: "shades of green"). And the cameras they used (specific make/models not mentioned) basically couldn't capture the range of greens required, or distorted them. Spherical Aberations from the el-cheapo lens on the cameras just made things worse. Bottom line: years of work needs to be re-done with more expensive, calibrated equipment.
--
I'd rather have a bottle in front of me than a frontal lobotomy
The SANE folks have gone to great efforts to get various scanner/camera devices to work in an open source environment. In some cases the manufacturer provided all the information needed to interface to the device; in other cases the interface has been found exclusively through reverse-engineering.
I highly recommend that you look closely at the list of supported SANE devices and choose a device known to work from the list. If you go into your local computer store and buy something off the shelf without looking at the SANE list, you are *very* likely to end up with a product that is completely unsupported in any useful environment.
Calibration is the ultimate tool of the scientist, in that it allows you to measure your measuring instruments! Generate test articles, measure them, and determine the accuracy of the instrument. Bing, bang, boom, done! This is a no brainer, you would need to do it with any scientific instrument, regardless of its origin (whether it came from bestbuy or from perkin-elmer).
Here a potted google search.
BugBear
Ignorance is curable. Stupid is forever.
I got my degree in physics, and you're right that off the shell hardware can be a great cost cutting measure. It's honestly disturbing how many times I've seen data collection run on something like a 386 using QBASIC.
The thing to learn though is that consumer hardware is not scientific hardware. There is rarely much quality control with regards to specs, even when they are available. If this hardware is going to be the dominant error source you probably shouldn't be using it in the first place. As tedious as it can be, it's a good idea to test the specification of ANY piece of hardware that you are adding to a research lab, whenever reasonable to do so. I still remember wasting two days of my life because the magnetometer was disturbingly off spec, and that was a serious research tool.
How do you test scanners and cameras? Clearly by scanning and photographing known objects. If you're just scanning diffraction patterns and stuff like that, then find a couple well known, well understood such effects and use them as your benchmark. It's also possible to buy high quality gray scales and precisely known grids to use as references.
The lesson here is, don't use cheap equipment when it will be the dominant error source (preferably use it in parts of the experiment that contribute neglibly to your overall error), and TEST all your equipment and quit relying on spec sheets for anything important. Publication retractions that read the equivalent of "Oops! There really isn't any effect here, but we were too lazy to get it right." are very funny, but won't do anything good for your career.
First you can check out How Things Work for the basics.
Second, off the shelf imaging devices are challenging to use for scientific data collection for a number of reasons. The main one being their response is usually designed to replicate the human eye rather than a true spectral response--the difference between photometry and radiometry.
For resolution tests, go to www edmundoptics com and check out the various testing targets available. The cheapest mylar USAF targets are pretty good for testing spatial resolution. Remember that when you get close to the resolution limit of the CCD, aliasing due to misalignment is going to be a factor. Your resolution could be up to a factor of 2X (per axis) better than you can test for, unless you're able to align the target with the pixels.
You should also try to figure out which CCD the device uses. Yahoo!'s Electronics Marketplace is a good place to search for components and there is usally a link to the manufacuter's spec sheet. Some spec sheets are quite detailed and will give you plenty of information regarding sensitivity, dark current, spectral response, etc.
Be skeptical of resolution claims. A flatbed scanner I have claims 9600 dpi or about 2.6e-6 m resolution. In reality, it's no better than about 5e-5 m.
Also, the picture you get out vs the "real" picture is highly dependent on the imager's software & firmware. Autoexposure and color correction functions are usually present and can play havoc with an attempt to figure out what the "real" image is. Again, test targets may help here--if you can control all the other variables in the system, you can do some calibration experiments to figure out what the imager is doing to your image.
Well, I hope this points you in the right direction.
I bet you wish someone would repond with usefull information. Here is what you should do. Make a "sample" to be measured. I don't know what you are trying to measure. Making up a perfect example is difficult because of that. But you will have accurate measurements of the "sample" because you have measured it with standard means. Scan the sample with the equipment you need to callibrate. (scanner, camera, etc..) Does the data from the flatbed scanner match what you know to be true about the sample? If it does so after a few tries, it is suitable for your purposes. It probobly won't. Experiment with different scanner settings and other software suites. You are probobly interested in the raw data from the scan. Many scanning programs will perform touch-ups on the image for photograpic purposes. This will throw your data off. It's possible that the equipment won't work for your purposes. But this procedure will help you find out.
astronomers use ccd cameras all the time to gather images from the sky. hence, they also have powerful software to analyze this (pixelized) data. common tools would be DS9 (http://hea-www.harvard.edu/RD/ds9/) and IRAF (http://iraf.noao.edu/iraf-homepage.html) which work with fits files (use convert to generate them from common formats). you could also do it simple, get a digital camera which runs under linux (e.g. nikon coolpix 900), download the images with photopc and then work from there. it depends on what exactly you want to do.
There was nothing terribly expensive about the physics laboratory equipment I worked with.
There are exceptions, such as specialty devices like the Michaelson-Morely apparatus, lasers with particular wavelengths, oscilloscopes, frequency analyizers... but none of that is going to be replaced by a general purpose computer.
You may be looking for entirely different kinds of experiments which can be done using computers and digital cameras or scanners... like "take this camera and use it to measure distance, speed and direction of motion", "determine the rate at which accuracy deteriorates", "move the camera or use two cameras to calcuate the distance of unknown objects, applying what was learned about the camera's accuracy and resolution to determine your confidence in the object's position" or "measure the colour response and accuracy of this scanner"
Other fun first year exercises might be to demonstrate the effect of various binary representations of numbers on the accuracy of data... all physics students need to know that stuff.
Forget about push-button dumps of information into Matlab or whatever. I hated when lab instructors would set up labs, you don't learn anything. It would be worse if I walked in and didn't even have to measure anything... just hit a button (god forbid touching the apparatus!), push the data into MatLab, follow the instructions, hope the OS doesn't crash, then hand in my results.
something you should go and check out is a trade show RSNA which shows up medical scanners and other imaging hardware that can be usefull to you.
RSNA is held in chicago.
-rev
Consumer imaging devices are great if what you want to record are spatial patterns of light in an image or over time. But if you need to record absolute light values, or measure differences within a still image, or compare values over time, you will need a scientific measuring device capable of maintaining a calibrated black level as well as a known response to light. Consumer scanners typically only maintain a calibration within a single scan cycle, then recalibrate for the next scan, in order to give consumers the best digitization for each individual target. Consumer photo and video CCDs let the black level 'float' over time so as to give the best overall exposure at any given instant. These design features make consumer equipment useless as measurement devices, but ideal as pattern recorders.
Keep in mind that good postprocessing can factor out all sorts of predicatable equipment shortcomings. When the Hubble Telescope went up with a seriously flawed mirror, good software made it possible to get scientifically valid results without replacing the flawed optics. A similar approach might be useful here, if you're interested in this aspect of the problem.
Also, keeping benchmarking data such as a color test image in field in each of you data images could allow for per-image calibration and factor out some of the unpredictability of consumer imaging. This could be easily automated in software.
There's plenty of room for commodity hardware in research labs.
Just look at any Fisher Scientific catalogue and you can see the rip-offs running wide and free.
Common household blenders get re-labeled with a 'scientific' label and sold for a huge markup.
Those cheap 'diaphram' fish tank air pumps suddenly become expensive scientific instruments at far higher prices.
And that doesn't even start to touch the root of the problem.
Abolish tenure.
Not only do I agree 100%, I would put it even more strongly.
Stop thinking like a freshman who expects to find the answers in the back of the book. Even if you find this information someplace, the nature of commodity (vs. scientific) gear is that the manufacturer can change it at any time to meet market needs.
You're a senior and need to start thinking like one. If you need calibration data, and you do, you should be thinking about how to get it for yourself using other commodity equipment. This is important today, critical with the improved hardware a decade or two from now.
A trivial example I would have killed for 20 years ago? A 600 DPI laser printer. With it you can easily produce high quality optical test patterns, including some basic grey scales. (A standard sized sheet of paper will have far more 'pixels' than the CCD element in the camera.)
A slightly more advanced example is what you can do with a cheap A/D card. 10-bits of accuracy doesn't sound like much, but if you're clever you can leverage it.
Finally, I would strongly recommend that you review the "Amateur Scientist" columns in Scientific American over the past four or five years. If you can construct a simple closed feedback loop (cheap op-amp chip) and monitor it with an A/D converter ($100), you can do some incredible experiments.
For every complex problem there is an answer that is clear, simple, and wrong. -- H L Mencken
If your goal is to reduce the cost of automating experiments that require an optical sensor, then consider the imaging equipment being used by amateur astronomers. These imagers are less expensive than the "professional grade" units, and are much more adaptable to being attached to equipment than are consumer units. Most of the amateur astronomy magazines have an assortment of ads for these units. As indicated by other folks, you'll need to develop or acquire physical calibration standards for noise, linearity, sensitivity versus exposure time, resolution, dark response, pattern sensitivity, repeatability and temperature stability, to name a few. It sounds like fun. Good Luck, Art
Artifacts from CCDs are bad enough - you don't need more caused my "corrective" software designed for human perception.
--
Evan
"$30 for the One True Ring. $10 each additional ring!" -- JRR "Bob" Tolkien
Does the scanner produce the same results each time? Will you have to recalibrate or measure the scanner's accuracy before and following each experiment? Calibrate for sure, but watch out for variations as things heat up or move around. I think you're going to have to spend extra time making sure your "consumer grade" tools are behaving as expected in every case.
or does this guy seem way to picky about his PORN SCANS?
-- www.globaltics.net
Political discussion for a new world
"The problem is that I am having trouble finding out about the way in which scanners and digital cameras work, and how this would affect their accuracy with respect to what I am aiming to do."
This can be learned by studying optics. It is not very staight forward, but any discussion relating to it would be far beyond the range possible in discussing on slashdot. Get a book and start reading.
"The information I am looking for includes things like: the resolution of their grey-scales, what degree of accuracy the motor steps at, how uniformly distributed the CCDs are in the arrays, and other issues that might affect accuracy. Just so that I can know how close to the 'real' picture what I get out of the scanner/camera is"
There is a LOT more that affect what "the real picture" is than these factors. Again, perhaps you need to go do some reading on optics.
Burn Hollywood Burn
I worked in a lab where the intensity reading of each pixel was important. We used scientific grade equipment, so that we could set the sensitivity, offset, and correction.
If you are interested in measuring the intensity of each pixel, read on:
First of all, I would think that a consumer camera with automatic exposure control would automatically set the gain (or sensitivity to light). You would need to be able to turn this off.
Secondly, the offset has to do with black noise (or error due to thermal energy within the ccd). On the camera's that I used, it was around 5 intensity levels out of 256 on an 8 bit camera. There is not a need of refining this on consumer equipment, so it probably doesn't get much better than that. You can buy cooled camera's for getting rid of this, but you want it cheap so this is not a great option. You could try cooling the camera w/ liquid nitrogen. I wondered about doing this myself. Alternatively, if you are taking images of something that doesn't change in time, you can take multiple images and average them. The black noise of the averaged image will decrease as the square root of the number of images.
Thirdly, the image correction -
Most consumer equipment uses a gamma correction curve b/c of similarities w/ film and video. Look it up if you don't know about it, it is interesting, and useful for taking pleasing photographs. For scientific purposes, though, you probably want a linear response. This will give you a constant sensitivity to light changes.
The last thing you should be concerned with it changing conditions / response with time. Some others have noted this. You will need to calibrate the device many times at different times of the day to make sure that changes over large times are reasonable. We utilized a calibration during the experiment to reduce this problem.
As for image formats readable by mathcad / matlab, etc. That sould be fairly easy once you get the device driver settled.
For these kinds of things just use normal 35mm camera film to record the difraction patern (or whatever) and then run it through a film scanner, flatbed scaners would be very arcane. A good film scanner will run for less than $500 and give you probably over 3000dpi of resolution (mine is about 2 years old and does 2720).
This has two major advantages. First you get something non-digital to archive (the film) which you may use later on for studying something completely unrelated that you never thought of. Secondly, you'll probably get much better quality. Film scanners are made for professional/semi-profession uses and are probably alot better built.
To test if the thing isn't distorting your images produce a well known pattern and mesure it digitaly, see if it checks out.
Pedro Côrte-Real.
I just asked a Professor here at the optical sciences center if he knew of any good publications on scanner technology.. He said to check out books/papers by Leo Beiser. He has apparently written some books on Optical Scanning technology and various SPIE papers as well. This may give you a starting point for a more rigorous look at the tech behind scanning.
First off, most commodity imaging hardware does not have specified shutter rates, so anything that's timing dependant may be a problem. On the other hand, at the slightly more money than commodity but still less than lab equipment point, are industrial CCDs... usually they can be had for sub $500, and have a lot more options and better specs.
Assuming that you are going with a commodity imaging device, calibration is going to be important. Do it yourself with test patterns. Also, try to work in greyscale if possible. Many CCDs, when in color mode, have some built in color / light level compensation that will kill all your chromatic accuracy.
I am disrespectful to dirt! Can you see that I am serious?!
consumer != science!!!
I've seen it multiple times, adhere to it! I graduated physics this summer and my CCD's cost approx. $30k - $50k EACH! And there's a reason for it! (well, you might not exaclty need _that_ costly an equipment, but still...)
You sure you're cut out for this field?
I mean come on - what are you going to do in your results - cite a random slashdot posting as a reference on the accuracy of your tools??
You need to test the hardware you have. Even if I have the same model, it could have a different CCD, DAC, etc.
It is just like assuming that just because we both have Dell 2550 servers that the hardware is identical. It probably isn't.
... you'll need this program to read the raw CCD data.
An easy way to calibrate the color and/or gray-scale response of a scanner ( or digicam ) is to go to your local high-end photo store ( one that deals in darkroom equipment ) and buy a color standard card. Kodak publishes them for use in calibrating color printing equipment and evaluating filter packs for color enlargers, and they are QUITE accurate. They also have a gray-scale reference card available. While this won't be as accurate as calibration against NIST primary standards, they don't cost as much either. These cards are NOT expensive, but the old, immutable rule is true. Cheap, fast, good ... pick two. Do NOT use a laser printer for grey scale standards. Even inexpensive scanners can be set to use interpolated scanning at high enough resolution to resolve the toner particles.
As for x/y positional calibration, I made up a template for fret placement on a guitar fingerboard, once upon a time, by computing and plotting the fret placement in AutoCad and printing it out on a laser printer. The finished home-built instrument played scales more accurately in tune than my commercially-built acoustic guitar did. Or, if your school has a machinist on-campus, see if you can obtain a set of Jorgensen blocks and scan them. They are sized accurately to, IIRC, 0.0001 inch, or so. If you decide to use a laser printed calibration chart, be SURE you use a grid, rather than, say, a rectangle of a certain size. This way you will be able to determine whether there are any non-linearities in the motion of the scan head.
Accurate calibration standards just aren't THAT hard to find.
utter rubbish
use something of known dimentions and scan it and try to measure the acuracy, you're a scientist, create a set of tests to measure color and distortion.
"The Most Fun Possible on 4 wheels" is at SunBuggy in Las Vegas
Slightly off-topic -- In general, I don't recommend PC-based measurement equipment. A company I used to work for bought a PC-based Oscilloscope trying to save some money. It was a cheap data acquisition card you plug in to an ISA slot and used software to look at waveforms. But in actuality the capture and rendering was ridiculously slow, and the controls hard to operate, so nobody was able to perform any real work with it. Also one must take into consideration the additional trouble of having to set up the PC near the lab bench, waiting for the PC to boot up, and the lack of isolation between the probe and PC.
In the end, the purchase was a short-lived expensive toy.
Download Mazes and Puzzles from www.puz.com
They have an interesting battery of tests for digital cameras. Take a look at some of their tests and that will give you an idea as of what you should be looking for.
BTW, the guys are right. Don't expect consumer hardware to be consistent. You are just not paying for that consistency. Also keep in mind these things are not built to take the abuse of continuous usage. Devices that are a bit more expensive are usually built better. For example, look at a $50 scanner and a $200 scanner. Now guess which one will still have a functional lid after a month of intense use.
Pedro
----
The Insomniac Coder
Although manual data collection is sometimes tedious, part of what the labs are supposed to do is teach data handling, error propagation etc. For the relatively straightforward experiments in 1st/2nd year physics labs, it's probably easier to do the error tracking when the data is measured by hand (The calipers are accurate to .01mm, the stopwatch measures to 1 ms, etc). The quantities that you measure are clearly and unambiguously related to the experimental variables. Errors can be calculated explicitly too. It's already hard enough to do this part of the experiment well. Face it - it's boring. Automating the collection makes the whole notion of experimental errors more abstract and more difficult to teach.
Not everyone's going to be an experimental physicist but the rigour still has value.
The proper spelling is "commode hardware in lavoratories." Please be more careful
Many people have already pounded home the idea that you need to calibrate your devices. Camera calibration has been well researched, and there are several well known methods available. Perhaps the best known calibration method for modelling the camera as a pinhole model is Roger Tsai's alg.
h tm l/vision.html
You can find this and various other info on:
http://www-2.cs.cmu.edu/afs/cs/project/cil/ftp/
A few notes:
1. There are two separate calibration issues. Intrinsic (focal length, pixel size, aspect ratio, image center, radial distortion) and extrinsic (position and orientation of the camera in space)
2. Images can't give you full 3D information, so euclidean measurement isn't possible with one camera. The images are a perspective projection of 3D onto 2D. (lengths in image space aren't comparable, angles aren't preserved, etc. )
3. Using these calibration routines means you will have to have a calibration target, and likely will have to pick point correspondences by hand.
I work surrounded by a bunch of microbiologists. A lot of our scanning work done here are on gel runs (DNA / protein sequences) and the data is so course (?) that you could do gel compairs using a two dollar ruler. For stuff like that we have no problem using off the shelf equipment.
Now seeing that your from a Physics lab, your probably going to want to be capturing high speed interactions, tracking single particles, and the like. I would suggest for you to stay away from commodity equipment.
I for one have noticed that calibration varies pending apon the position of the sun and how close the window is to the scanner. One guy thought I was crazy for putting a towel over the scanner to block out ambiant light! Not to mention what I noticed after some playing around. I think I still have some images from one scanner showing glass warp and the non-continious brightness of the lamp. Definite learning experience that was!
I have some experience in this, so i will interject.
Home scanners and digital cameras are definitly not suited for the task if you need very near digital reproduction of an object. The reasons for this are many, but mainly, the all interpolate colors between what imaging elemaents they have. And not that accuratly.
When you move up to the midrange of scanners/digital camers (~$2000), the problem can still be there, but its less pronounced. I worked on a project requireing digital photos of a very hard to photograph subject, and tthis range of cameras produced sub-par results for the task (the shots look incredible, but zoom in and youll see fuzziness and interpolated color).
Then, you have the ~$20,000+ cameras and scanners. This was eventually what we had to go with. One camera delivered particularly good results, and achived it through actually moving the CCD so that there were no interpolated pixles. It was accurate enough that if you shot a Greytag/Macbeth chart, right from the camera, the greys would be the same value for evey pixel.
As with all these [camera] setups, You need a very controlled lighting situation (ie. photo studio), but you can shoot just about anything.
As far as scanners, the same applys. You will need to get in toe the pricy professional line to get accurate pixles, and from that, better analysis.
Your test for any product should be if you scan a greyscale, if you go in to photoshop and look at the pixle color values, are they all the same value (like 125,125,125) and, is it consistant across the swatch (if you move your mose a few pixles over, does the value change?)
The other aspect you have to contend with is your computer and monitor and its interpretation of what your seeing. Again, if any amount of accuracy is needed, you will need a controled lighting setup. No direct sunlight, try not to wear clothing that will project a color cast on the monitor, a lightbox to properly illuminate the scaned subject for proper color editing, etc.
This is where you buy a macintosh. you dont need to do all the ColorSync stuff, just keep your monitor and scanner/camera in line.
So based on the three levels in imagry equipment (home, semi-pro, pro) you can determine what level of final output you need and judge your costs from there.
For full setup, id guess:
Home 5-8k Semi-Pro 10-15k Pro 20-40k
Some useful links:
Greytag Macbeth
Apple:ColorSync
Imacon 3020 is camera i mentioned above
megavision
leaf
Sinar
Phase One
Betterlight
This is mostly high end stuff, but, you should be a good starting point in findieng the mix of price/performance you are looking for for the overall project.
"Stuff... In my home!? NEVER!" - Zim on Invader Zim
"I want the toilet seat!" - Little Dog on Two Stupid Dogs
I agree with this completely, except that I read the original post as a question of reliability or accuracy of the commodity hardware.
Calibrating any device is important, and I love the laser printer idea for positional calibration. I think other posters responded to the intensity aspect of this.
One problem is that, for digital cameras, most are automatic and set F-stop, shutter speed, and do color corection and sharpening afterwards. You don't want any of this post-processing, and you need to know the shutter speed and F-stop (or their equivalents). I don't know enough about the market to say where these are available or not. I guess if you can't control this directly, then you'll have to include calibration data with every photo as part of the "scenery", as if each photo were taken with a completely different camera with different properties. I have a great book entitled "CCD Astronomy" which, while it isn't exactly on topic, covers the basics of image processinig and calibration very nicely. I don't have the book with me now, so I can't tell you what the author is. I used this book as an undergrad in physics while working on a CCD camera as a research project. It was very accessable to me.
But the main thing you're after, I think, is the accuracy of the device, after it's been calibrated. I think this is based on two factors, the actual pixel sizes/bit depth, and the accuracy of your calibration factors. Something like pixelsize/sqrt(12), added in quadrature to the calibration errors. Estimating the accuracy of the calibration is an interesting business, and there are a variety of ways you could do it. Taking multiple calibrations and finding the statistical spread (RMS) is an easy way, but only gives you a lower bound because there will be systematic errors in your calibration procedure. I'll leave it at that. Probably any intelligent, hand-waving approach to estimating these systematic errors is enough for a senior project. A really thorough approach could take much more time than you have, generally speaking.
Hope this helps!
http://www-2.cs.cmu.edu/afs/cs/project/cil/ftp/htm l/vision.html
/code. Computes a pinhole camera model from n (n > 8) matchpoints. Computes both the intrinsic and extrinsic parameters. The more matchpoints the better.
Look for Tsai's calibration program
Good luck
I'll chime in here just because I've seen *so* many computer-types foul this up. (No slam, just fact.)
The bottom line is this: Everything depends on how well you collect that initial data. This is TRUTH: No amount of signal conditioning, DSPs, FFTs, DCTs, quantum neural framulators, or anything else can make up for crappy sensing elements.
This is true whether you're talking about image sensors, temperature sensors, EM sensors, mechanical sensors (force, pressure, torque, etc.), or anything else. Remember that silly saying you learned in your first computer class: Garbage In, Garbage Out - It's still true, whether you like it or not!
Sadly, I can't tell you how many times I've seen really bright people ignore this simple fact of life (wasting countless millions of dollars in the process), confident that the rules don't apply to them, and that their computer can somehow create something from nothing. (My experience tells me thse people are more likely to be in academia or very large companies where "scientists" are more highly regarded than mere "engineers".)
BTW: I know a thing or two about this because my father's company (and no I did *not* do the web site) specializes in building high-quality mechanical sensors that provide laboratory precision in hellish environments. (Literally hellish: like downhole in oil wells.) Some customers are willing to pay for really good data, realizing that there's no alternative if you really need to know what's going on.
Get the sensing elements right, and the rest of your job will be much easier (and cheaper, too...)
"The future's good and the present is nothing to sneeze at." - Roblimo's last
For work like this, I like my QX3. Cheap and powerful.
There's a short review of its capabilities here, but this site has some amazing hacks that enable it to do darkfield, polarized, Rheinberg, or even simulated Hoffman modulation contrast viewing.
Da Blog
First off : you should be able to get info such as pixel sizes from the manufacturer (typical pixel sizes for CCDs are in the in the 10^-12 range for scientific grade CCDs) In arrays CCDs are fairly good. . .all CCDs have problems with charge transfer efficiency of course but if you're willing to gut your digi cam you'd find out about that. Also you will start to get quantum efficiency problems if it's not a good chip- so your detection of photons as a function of wavelength will be very poor. And then there biggest problem with digi cams are the read noise as CCDs don't do to well with out being cooled.
Software wise I'd recommend you go track down your local astronomer and ask them about IRAF or IDL. IRAF is free from the National Optical Astronomy Observatories and is made to work well with CCD images. The same goes for IDL but it costs arms and legs (say roughly $1000 for a license!). I think the best advice you can get is 1) cantact the manufacturers on the technical info they have and then 2) track down your local observational astronomer. They can tell you about gain, read noise, shot noise and tell you about fun effects like diffraction fringes at redder wavelengths that you can get with CCDs.
Okay, there will be no "probably" or "maybe" words used in this response. I have completed a contract which involved using a flatbed scanner to calibrate a laser marking system.
The system was used to mark/label ICs on PC boards in a manufacturing setting. The tiny print required high accuracy in the scanning system.
The system was commanded to make a square grid, which was at first aligned by eye and some plastic rulers. Then the system was told to make a square grid on a photographic negative which was scanned in using a flatbed scanner, and my software found all the line intersection points and constructed a file of calibration adjustments to make the grid accurately square.
For spacial accuracy, most consumer scanners are not sufficient. The plastic bed mechanism is simply too sloppy for good repeatability. Also, the scanner needs good PHYSICAL resolution instead of SOFTWARE resolution. Essentially, PHYSICAL resolution is the hardware resolution in pixels per inch. Using software and special processing algorithms, the physical accuracy can be increased by taking each pixel and dividing it up to make 4 pixels or more.
What you want is a scanner with good physical resolution and high repeatability (ie - a tight non-sloppy bed mechanism). There are consumer scanners out there which do this - HP makes a nice scanner with 2000 dpi physical resolution and a very tight bed mechanism - costs about $5000. This pricing information is probably old by now.
Then be careful of what your images are printed on. Paper tends to swell and shrink depending on the relative humidity, so put your images on plastic instead. Plastic also has some of these problems (ie - temperature dependence on size) but they will be much less than with paper.
Make a test image and scan it in twice. Take the two images and "subtract" them... anything that's left over will be your repeatability error. Many paint programs have a "subtract" feature built in, such as PaintShop Pro. Do this a number of times and you can make an analysis of the total spacial accuracy of your system. I found that our scanner had a bell curve accuracy of between 0, 1, and 2 pixels and was therefore accurate enough for our purposes. One pixel at 2000 dpi is really small.
Standard Twain drivers were used to command the scanner to read in a bitmap image, which was then analyzed as a rectangular array of pixels. A standard (although home-built) filter-match algorithm found the locations of all the artifacts of interest (ie - the line intersections) quite accurately.
So for spacial measurements, within the limits outlined above the answer to your question is "yes, it can be done".
From my experience with film scanners, you're not going to find what you're looking for, or at least your not going to make anything easier with "consumer" hardware. However, keep in mind that expensive digital photography gear IS held to a very high standard of accuracy and reliability. Professional photographers wouldn't have it otherwise. Not as strict standards as scientific or medical gear I'm sure, but much better than consumer stuff. The current technology in my area is capable of surpassing the quality of film imaging using film smaller than 6x6cm. For instance, if you were to put an Imacon Flexframe camera back (5 digit price tag) on a microscope with good optics you would have an image that I would be more comfortable with than a 4x5 color slide. If that corolates at all to what you're doing...
Simple experiment: use a laser printer to print 0.01" squares 0.02" apart in both directions (ie.25% grey) on a sheet of acetate. Do it twice, superimpose the sheets, and try to get an even tone of grey. It's impossible, and there are no published specs on the accuracy of positioning.
At the very least, include a graticule in every scan if accurate measurement is important to you.
When I was in college we did this same sort of thing (I think). I was once a physics guy, doing research for NASA, trying to build a temperature sensing device with no electrical components with in the sensor. Ultimately we found that using a sapphire fiber optic we could measure the temperature variance by relating the change in optic signal frequency to the diameter of the fiber, and then calcualting the temperature. Anyway, long story short, we needed percisioning devices for our testing. My main part of this project was the electronics and software to get data from them, so we bought up some CD players (specific ones which had lasers that had wavelengths with of something like .5 micrometers) got the scematics from the companies and started ripping them up to grab the laser and focusing electronics. CD lasers at the time used some interesting stuff which allowed the electronics to move the lense of the laser to maintain focus within a certain range of distance (read can focus on a wobbling CD). This made the cd laser perfect to measure percision of our systems, we knew that the laser's wavelength was half a micron so that was a rather exact margin of error, once the laser was focused on a target, we measure the signals on the electronics side and could develop our margin of error based on the movements of the lense with an accuracy of half a micron. KEWL! It was a fun project, and the use of CD's kept us on budget as opposed to buying research lasers for the same thing which could have been thousands of dollars.
Later that year, I used my project for that research to develop a rudimentary surface topology scanner, which made really kewl 3D computer images of pennies, dimes and quarters at a microsopic resolution. Not bad for 1993...
You're a Senior. You're about to go out into the world and you still haven't learned the basic skills of research. Experience is no free hand out. Go figure it out yourself. I'm sorry, but it's your project. Some Universities might consider this collaborating on your project and I don't know if you're allowed to solicit this kind of help. We're like your fellow students, so do your own.
One major aspect of using commercial digital cameras that will get you when trying to make quantitative measurements is what's known as anti-blooming. One of the problems with CCDs is that when you start saturating a particular pixel, the charge from that pixel can "leak out" onto other pixels, causing "blooming", where you get extra electrons in pixels around pixels that absored a lot of photons. The solution to this is to have hardware anti-blooming, which automagically compensates for this (I'm not sure how.) However, the end effect of this is that your image is no longer linear: the number of electrons is not strictly proportional to the number of photons, because some of the ones with larger numbers of photons have been siphoned off to prevent blooming. In scientific applications, blooming is prevented by using low light levels to keep the camera well below the saturation limit, but anti-blooming is used in commercial applications whree the light level isn't well defined, so that you always get pretty (if not accurate) pictures. This will kill you if you're trying to use a commercial camera to take spectroscopic data. One good mid-range solution is to use amateur astronomy cameras, like those produced by Starlight Xpress. These cameras have no anti-blooming, are internally cooled, and have very small pixels and large arrays. The down side is that they're slow to read out, but in many scientific applications (like spectroscopy) this isn't necessarily a huge problem.
* mild mannered physics grad student by day *
* daring code hacker by night *
http://www.silent-tristero.com
This is physics, and part of physics is researching, testing, gathering data and data analysis. You need to think about the parts of your experiment. Nobody can say whether a test component is sufficient. In some experiments, a low cost ccd/scanner could be ok, but for others they are not. But only you (the researcher) can find this out by doing the experment and seeing if the results are consistent.
And just because something costs alot doesn't mean it works.
I played with it for a while many months ago. After reading this Oreilly Net Article about it. The link is to page 2, because that's where the calibration stuff is.
This is how you can find out where all the pixels are pointing. I suspect there is code for calibrating intensities, but I didn't use it.
My remote sensing professor had a setup using off-the-shelf video cameras for an airborne scanning system. Three "identical" cameras with filters to record three bands.
In the end, he discovered that there were substantial geometry differences between the cameras that made registering the three bands quite a challenge. That's the difference between a 12" format aerial photo camera and a 35mm SLR, the former costs a small fortune because a lot of effort has gone into ensurng that the film is held as flat as possible (on a plate rather than stretched between two spindles).
Then there's colour response. Same deal. I'd guess "scientific" quality sensors would be moe consistent in response to being hit by light.
Bascially if you can account for the effects of using more variable components (by calbration, experimental design or whatever), where's the problem? Regardless of which way you go, you will need to *know* how your instrumnts take their measurements so you can know how they influence your results.
Xix.
"Everything is adjustable, provided you have the right tools"
The Edmund Scientific catalog has some helpful optics accuracy hints for its products... biologists do a lot of this stuff as well, you might want to examine the NIH image software and associated resources.