Backyard Aquaponics
A place to discuss aquaponics

No one is picking on anyone, I pointed out why it wouldn't work and then explained the reasons why.

You would be better off contributing to a thread rather than snipping it.

I think it is a great idea as I am colour blind too. I have to keep asking my wife to help to interpret my tests. I'll definitely give it a try. Great work Ria!

You are aware that there are commercial water testing kits that use your phone to read the results? More on the pool side, but it works.

ps: How do you know this doesn't work? Have you tested it? Or do you just "know it won't work"?

This is a legitimate question.


This is a good response.

This bit, not so much.

Fair point

I wasn’t aware of commercial water testing kits using your phone but I would imagine the colour charts would be setup in such a way as too not have other colours interfere with the camera’s digital compensation. The API colour chart has too many colours influencing that compensation process and depending on which colours were involded would give different outcomes. The colours are too close together and depending on how many of those colours you had in the photo area would possibly well and truly outnumber the single test colour that you are trying too match.

The left side of the API colour chart favours the blue spectra scale and the right side favours the red, if you’re trying too match a pH of 6.0 -6.4 and have all the right hand side of the test card in your photo calculations there would have more of the red colour spectra being calculated than the single square that you are trying too match.

If one person framed their photo favouring the left hand side of the test card and someone else favoured the right hand side of the test card, the camera would have 2 different sets of colour values being calculated and depending how much difference was calculated between each side you would have 2 totally different values calculated.

The issue here is that if a swimming pool has a ph is below 6.0 or above 8.0 I don’t think you would die if you swam in it but in a fish living in a fish tank that’s a different story.

This information was too be gathered for research and if you read the posts the people taking the photos even they have their doubts about its accuracy.

Their quotes:

4x4Eric:
But remember that if you have led or fluro lights the colours in your pic can be changed quite dramatically.

When I take a pic of my tests on my phone I put the test tube on the colour I think it matches best then click and look on the screen I often find that I'm a colour match out which with PH means 7.4 to 8 could be the difference between minor inconvenience to a HSM.

Ria:
I found that its very tricky getting the exact colour and the colour can change a little each time you take it, maybe the test tube is on a slight angle, or you are getting different external light sources. But after a lot of testing and noting that even if the colour didn't seem exactly right or a bit different each time the it small differences didn't make any difference to the result.

I completely agree that computer screens and electronic devices don't give a perfect colour match. I get this fairly constantly when you present a powerpoint in a seminar the colours you chose on the powerpoint are not at all like those appearing on the big screen.

I've found with the tricky pH values (ie 6.4-7.2) even if the colours that are displayed on the screen on the chart don't look quite like I expected them to, the matching value corresponds to the value I get when I check it on the original. printed chart.



How I know it will not work in its present form; it’s not just my personal opinion.
http://www.cambridgeincolour.com/tutori ... tering.htm

32 years of lighting film taught me that if I have a white card and I add red light to it then it’s no longer a white card and it will photograph as a red card, I cannot explain it any simpler than that, if you’re still struggling with that explanation get back to me and I’ll try and work out a way simplifying it further.

The other way you can go about it is leave the theory behind.

Get samples of known pH and test them. Then use a range of different cameras with a range of different lighting conditions. Once you have a measure of the variation in measurement post the results here so that either:

1. we know your tool is useful and despite Joblow's valid theoretical concerns the error while present is not significant; or

2. we know your tool doesn't work and anyone who sees this thread will know that the idea has been tried before.

Either way it would be a valuable contribution.

Ok, so your objections are based in theory. Got it. Nobody is talking about shining a red light on test tubes, or blindly listening to the computer. Ria said that in their testing, it worked when compared to the original printed chart. In addition, you talk about photographing the test tube in from of the API chart and those colors messing up the reading. It's my understanding that you would take the photo of the test tube in front of a piece of white paper. The computer already knows what the colors are, and having a large white area allows the digital camera to easily white balance

What I think you would need to do is set a standard so everyone did exactly the same thing or you wouldn't be comparing apples with apples, for example a different amount of light reflected off the white background would vary the colour, light or shade on the white background would also alter the colour.

I don't know how far out it would throw the calculations but they would differ, and I don't if it would be too much but IMO I think it would not be a true comparison, it might be near enough but I would try too get the correct or close too the correct comparisons.

I'm currently using both a Sera & API test kits to take my pH readings and I'm finding with out doubt the Sera test is FAR, FAR much easier to read than the API. It has a lot more separation between the colour samples and it is viewed on white from above with both a larger area of chemical solution and colour sample for comparison than the API test kit.

It uses the same 5mls volume of test water but because it uses a wider test tube there is far less side light hitting the test water. I think most people just hold their API test tube up against the sample card for their tests, with the Sera test you can't do that so you're not looking at any glass curvature or stray light trying to match the color.

In the API tests notice the the nitrate test tube on the right in my photo looks darker than the ammonia test tube, it was actually lighter in colour and again the nitrite test tube is darker on the top half and lighter on the bottom half.

Those test tubes have some background and side colour influence but the pH test tube has a much lighter background but looks much darker than the nitrate test tube but were they actually much closer in colour than what the photo shows, that is the point I was trying to make.


IMO I think the Sera test or testing method would suit Ria's research better and have less problems than photographing a curved test tube, I think even with the test tube photographed against a white background the light hitting the test tube curvature from slightly different positions would show as different colours and different camera angles the same thing. It would only need too be a few degrees either way and it would make a difference in colour.

I think being able too use your moblie phone would be fantastic and I hope Ria can succeed with it.

The app is really nice but the quality of the camera of the phone will affect its colours. Many phones are not same.

Thank you everyone for your comments. This has been an interesting discussion, and you are quite right, I should have performed an experiment to test it out. So I have done some experimenting.


I used a scientific grade Ph meter from the university (Martini instruments) to get an accurate pH level of each test sample.

I created 5 solutions with different pH levels, as measured by the pH meter. I marked 5 test tubes with a different coloured dot to uniquely identify the test tubes (and to give no hint of the actual pH level)

I put the solution into my usual test tubes, and the API indicator solution as per API instructions.

I then asked 7 individuals, ages ranging from 24 to 62 to determine the pH levels of each of these 5 solutions and record their results against the colour on the dot.

I also used the camera on my Samsung phone, and RGB colour picker app to work out the colour code of these solutions in 4 different lighting situations – outside in the sun, outside in the shade, inside with fluorescent lights, and inside in front of my computer screen, which showed a white background.

The following table gives the results


The first obvious but unexpected conclusion was that none of the people testing the solutions managed to get 100% correct, and neither did the camera.

Worryingly, only one person managed to get more than half the levels correct. However nearly all their results were within one scale.

Also interesting in the variation in accuracy of the camera approach based on light source. Using the computer screen as a light source in this test was equivalent to the best performing person.

These results are interesting, but lead to many more questions that need to be answered. Why did the people not perform better? Do they need training to help them improve their success rate? Are their physical characteristics such as colour blindness that affects them? I don't have any answers.

Regarding the camera results, a full test would need to check different lighting characteristics, rather than the broad examples on this test. I was intrigued that the computer screen seemed to be consistently better. Yet more questions - does this depend on the screen, is what on a computer screen the same for each screen? etc etc.

I like the idea of calibrating the values, but then you would need to make sure that each time you were in the same setup.

On the other hand, it may not matter if you are one level out. I have noticed that many people have diaries with what I would consider high pH levels, and still have a system that works.

Without resorting to the use of lasers: What if you created a "light box" inside which the lighting conditions were the same each time, unaffected by any ambient-light conditions outside the box? I'd consider a box with white insides, a 20W incandescent light and an translucent baffle to prevent "hot" light directly hitting the subject, and a wire frame to ensure the subject is in the same location every time with the same orientation; drill a hole in the top of the box through which the photo will be taken.

I would imagine that photographing the water sample end-on and averaging the chromatic values for a specific number of sample points inside a region not affected by the meniscus or any curvature of the vessel (test tube) to obtain a single value would yield the best (repeatable) results? I also think you'd have to filter the water to ensure any turbidity (suspended particles) don't affect the sample.

I'm interested in how you're mapping the chromatic attributes to a value, and if you're using a self-organising map what training data you're using. I've never plotted the colours of the test cards to see how they lie in relation to each other, or figured if it's indeed possible to use a map to draw sample points to the correct node? I think you should also test how many dimensions are needed for reliability and the ability of your software to identify the values associated with those dimensions; may be you need more than just RGB? CMYK? More?

Ria - fabulous job! I'm quite impressed with your work.

Just as a matter of curiosity, what would happen if you went the opposite way and reduced it to shades of grey (no humor intended). Then would it just be a matter of counting? dpi?

I m using the colour distance calculation to find the shortest "distance"between the colours

This is done by dividing the colours into the red/green/blue components, and then finding the distance using the formula sqrt((r1-r2)**2 +( g1-g2)**2+(b1-b1)**2)).

I compare, for example, the pH value sampled with each of the colours on the API card, and then select the shortest distance as the closest colour and display this.

I will add the ability to calibrate your standard values and see if this helps.

Ria