calculate color mask

[Madhubalan]

Hi alexderjuengere,
Definitely .

Thank you

[alexderjuengere]

Kudos would be nice

[Madhubalan]

Hi alexderjuengere,

                               Can you please send me the labview.vi file for the segmentation(color mask file).

THank you

[alexderjuengere]

Hi,

you can use the .png snippet via drag and drop!

http://www.ni.com/white-paper/9330/en/

[Madhubalan]

Hi alexderjuengere,

                              When using k-mean clustering to find the dominant color in an image,for example,if i had given classes as 5,in the array there will 5 colors as output,each time when i execute the code with the same input image file,the colors in the array changes.Do you have any any solution for this or any other method to find the dominant color?.

I have attached the image regarding the change in color in the arrax color while executing the same input image for various time.

Thank you

[alexderjuengere]

hi madhubalan,

k-means in a nutshell:

1. Ask user how many

clusters (classes) they'd like.

(e.g.. k=5)

2. Randomly guess k

initial cluster center

locations

3. Each datapoint finds

out which Center it’s

closest to. --> classification!

4. Each Center finds

the centroid of the

points it owns...

5. ...and jumps there

6. ...repeat until

terminated

compare: http://www.autonlab.org/tutorials/kmeans11.pdf

one quick solution:

have you tried to reduce the number of classes? 

--------------

K-means is an unsupervised learning technique.

A k-means algorithm works best, if the initial cluster centers ("ground truth" would this be called in supervised learning techniques) are decently guessed by the algorithm or "teached-in" by the user.

I haven't installed this clustering toolkit on this machine, but I think this double-array constant was related to those initial cluster centers:

If this constant is empty, there is an automated guess, I guess.

Regards,
Alex

[alexderjuengere]

by the way, I prefer this machine learning toolkit, because it also got visualization in 2d (XY Graph) and 3d (3d picture controls)

[alexderjuengere]

Do you have any any solution for this or any other method to find the dominant color?

The screw you want detect is evidently quite bright due to it's well-reflecting surface and the chosen illumination.

Total reflex ion causes rather white colors, so the 3 rgb values of those particular pixels will be rather on the right side of your brightness histogram graph.

Speaking of histogram graphs, you should do some exploratory data analyses on your input pictures: http://www.ni.com/white-paper/4158/en/

This will definitely help you to localize the range of your thresholds. Regarding to thresholds: I recommend to use more than one.

Segmentation is a tricky task with real world data.

Regards,
Alex

[Madhubalan]

Hi alex,

           I could not able to find any other method for dominant color determination.If you have any method ,then let me know it.

But ,if i use only class 1,then dominant color remains the same for many execution step for the given input image.

This the link for downloading the clustering tool.https://decibel.ni.com/content/docs/DOC-4063

I have attached a vi file for color mask,in this file,how to determine the percentage of dominant color present in the given input image.(i have attached a input image)

Thank you

[alexderjuengere]

But ,if i use only class 1,then dominant color remains the same for many execution step for the given input image.

and does this give you the expected result/color? are you fine with this?

I have attached a vi file for color mask,in this file,how to determine the percentage of dominant color present in the given input image.(i have attached a input image)

now, this sounds to be trivial:

                                 number of dominant color pixel

dominant color in % = ---------------------------------------------

                                 total number of pixel

have you given a try to my previous post at:

http://forums.ni.com/t5/LabVIEW/calculate-color-mask/m-p/2472546#M756827

?

a will look into your example later this day

Regards,
Alex