LabVIEW
NI Vision: Color Spectrum
This is somewhat related to the original question.
I need to calculate the area of certain colors in an image.
The colors are not pure RGB, and are very close in the HSL and RGB color spaces.
I have struggled to find good ways of detecting them, so that I could calculate the areas, but yet I have not found a method that can match the colors as well as the build in LV_ColorMatch() and LV_ColorLearn combination.
my problem is that I cannot do it fast enough.
The colors represent different layers of a material, that is randomnly damaged, as a result the areas that I am looking for occurs unstructured. The camera resolution is matched to the task, so that 1 pixel after bayer bilinnear colour estimation represents the area resolution the customer is looking for.
The images are precaptured and the ROI is found by the capture system, that produces the images for this vi.
The input images are 3000w by 300w [px]
when I attempt to do it like this double for loop, I have to iterate 300x3000 times ~900'000 times - and that takes about a minute.
However I need to do this in about 3 seconds.
The image below contains a screenshot of a simplified version of my software, that illustrates the algorithm that actually works well enough to provide a decent count of target colours.
(I know the mm/px ratio in both axis, so I can multiply the count with that and get the area)
This is very frustrating as the build in histogram / historgraph uses a few milliseconds, and even though they only look at 1 colour space, I guess they must access the image data in another way than the double for loop pattern.
Zcuba, do not hijack other threads !!
As a rule of thumb, if you are looking for a region (different layers of material) first figure out the region !
You are using functions, that will give you the result, but as you figured out, it takes loong.
Your source code can be replaced by faster alternative using IMAQ color treshold. You should get your 3 seconds.
If you dont fancy HSL, just use the RGB model. For images of your size, the treshold should run in about 50ms (guess).
Try something like this :
Hi Bublina
It was not intended as a Hi-Jack
The question I tried to formulate, is so closely related, (on how to use the color learn / color match efficiently) that I thought creating a new thread would merely make someone link to this one. 
- I have no problem with using the Hue-Sat space, in fact I already tried using the IMAQ color treshold on RGB space with terrible results, and have tried without achieving a robust result with the HSL space.
The reason for abandoning the strategy (probably to fast) was due to the problems I faced with manually determining the optimal color spectrums for each layer. 
I am basicly looking for 3 shades of pink, and one gray layer that have similar intensities as one of the pink shades..
I will re-visit the idea after you'r comment, and use a little more time on calibrating color spectrums, but have little confidence that I can eperate the layers robustly using only 3 color space threshold attributes.
It seems that in tests the combination of using colorLearn / colorMatch can seperate the layers robustly.
So I am very interested in the original question, and may have formulated my own in a slightly wrong way.
What I tried to ask was that I would very much be able to use the "color spectrum" from colorLearn to do a histogram (like you'r example shows for the HSL space) but I want to do the histogram on the entire parameter set in color spectrum, as this seems to seperate colors better than I could do manually.
Hence I also, like the original questioneer, have interest in knowing what the parameters in the Color Spectrum Array, are indicating.
I did not get much knowledge from running the example linked to...
thank you for the help though 
