Normalize a clustered color space

[liaw82]

Hi.. Sorry as I dunno which forum section should I post this question...

I have a number of fleshy tone samples and I found out the Cr color plane for each sample. Then I drew out a histogram which clusters all the histogram of each Cr color plane. The result of the clustered histogram is shown in the attachment.

I wish to normalize the clustered histogram so that the graph will become a normal gaussian waveform which range from 0 to 1. I found out the difficulty of doing this as it is a clustered histogram rather than just a normal histogram. Does anyone have ideal of doing this?

p.s. I need the normalized histogram to transform my image into skin tone probability image..

[RManion]

Hi again,

It sounds like you are working on a pretty interesting project! Let me see if I understand your question correctly.

You have an image representing the Cr Color plane of the image. Is this in reference to the chromatic VI we worked on earlier? If so, that means there is an image that expresses the percentage of any particular color plane present. Is this the image we are working with?

Also, I am not sure what you mean by a clustered histogram. I see the image you have attached, and it appears as though you have a large number of plots with varying quantities of low intensities. I am unfamiliar with the term 'clustered histogram', but if you explain it, I may be able to help you normalize it.

Thanks,

Robert

[liaw82]

Hi...

I'm not using that chromatic vi but another vi which was developed by mike.. I took out part of the work of him to just seperate a RGB image to seperated Y, Cr, and Cb images.

I have made 93 samples of my own face color for tesing. Each is 40x40. (if i consider Cr plane only) I find the histogram of each Cr samples. Hence i have 93 histogram. Each histogram is an array of Cr intensity. Then I clustered 93 arrays and display in the graph as u have seen. There were overlapping actually. Since this 93 clustered histograms are representing my skin color, I wish to normalize the clustered one into only one gaussian with range of [0,1].

If my explaination is still not very clear, u may have my program for further understanding...thx really

p.s. my way of doing it was to normalize each array by using "normalize vector" vi. Dunno y d output of each array is not btwn 0 and 1. so i divided each array with the maximum no. to make it 0 and 1. Then I average each normalized array to develop a single wave. I'm not sure this idea works anot..

thousand of thx again..

[RManion]

Hello again liaw82,

Thanks for explaining the clustered histogram. I now have a much clearer idea of what you are working with.

Without knowing more about your application, and the algorithms that are associated with it, I may not be the best resource. In most applications I have dealt with, it is possible to normalize data by dividing the set by the maximum entry. In those situations, the distribution being Gaussian was actually a property of the data. This wasn't something that we could cause without ruining the data.

That said, this type of data manipulation is not my expertise. If you are working on this as a project, I might recommend asking for assistant from whomever has created or selected your processing algorithm.

Good luck,

Robert

[liaw82]

thx..and thx again for ur hardly consideration..

it's ok n i will figure out myself.. I will post up if I get the idea of doing it..