Local Binary Patterns

lbp() returns a histogram of features, so that positions X corresponds
to how many pixels had code X.

HTH
--
Luis Pedro Coelho | EMBL | http://luispedro.org

Post by Solimyr
Hi to everyone!
I'm trying to use the mahotas features to analyze some images. In
particular I'm using the Local Binary Patterns (LBP).
My idea is to try with image with radius 1 and 8 point and so, a 3x3
window. So I take a 3x3 window and apply the mahotas lbp. I was
expecting a vector/histogram of 8 values with just 0 and 1 but, what I
get, is a 36 vector-size with elements from zero to more than one. I
don't know if I'm stucked with some old theories of LBP or what.
Why don't I get a 8 vector size with elements from 0 to 1? My idea is
afterthat, convert in decimal and get the decimal value. Well, I'm
pretty confused with the result and the theory that I get from internet.
Could you please help me to understand?
Thank you very much for your support!
Solimyr
--
You received this message because you are subscribed to the Google
Groups "pythonvision" group.
To unsubscribe from this group and stop receiving emails from it, send
For more options, visit https://groups.google.com/groups/opt_out.

Luis Pedro Coelho

2013-05-01 19:59:31 UTC

I took this opportunity to better the documentation on the function:

http://mahotas.readthedocs.org/en/latest/lbp.html

(It's nice that Read the docs automatically updates after a github push)

HTH,

--
Luis Pedro Coelho | EMBL | http://luispedro.org
--
You received this message because you are subscribed to the Google Groups "pythonvision" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pythonvision+unsubscribe-/JYPxA39Uh5TLH3MbocFF+G/***@public.gmane.org
For more options, visit https://groups.google.com/groups/opt_out.

Solimyr

2013-05-01 21:02:30 UTC

Post by Luis Pedro Coelho
lbp() returns a histogram of features, so that positions X corresponds
to how many pixels had code X.
HTH
--
Luis Pedro Coelho | EMBL | http://luispedro.org

Solimyr

2013-05-01 21:12:27 UTC

Thank you for your answer but sorry, it's still not clear to me and of
course, you don't have to explain to me anything! It's my fault!

I have my pixel in the middle. Let me say that I'm taking a R=1 and P=8 so
it could be represented as 3x3 window, right?

What I was expecting from the function is something like this (I will try
to explain it with an example):

I have this 3x3 window:

10 13 12
8 9 22
1 3 5

So, I'm expecting this result:

1 1 1
0 1 --> 11110000 = 240
0 0 0

Using the mahotas.features.lbp(image,1,8), I get:

[2, 1, 6, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]

I'm start to thinking that are two different stuff or from the second I
could get 240?

I'm sure that is a dummy question so don't worry about answering...

Thank You,
Solimyr

Il giorno mercoledì 1 maggio 2013 16:53:58 UTC-3, Luis Pedro Coelho ha

Post by Luis Pedro Coelho
lbp() returns a histogram of features, so that positions X corresponds
to how many pixels had code X.
HTH
--
Luis Pedro Coelho | EMBL | http://luispedro.org

Luis Pedro Coelho

2013-05-02 06:45:15 UTC

Post by Solimyr
Thank you for your answer but sorry, it's still not clear to me and of
course, you don't have to explain to me anything! It's my fault!
I have my pixel in the middle. Let me say that I'm taking a R=1 and P=8
so it could be represented as 3x3 window, right?
What I was expecting from the function is something like this (I will
10 13 12
8 9 22
1 3 5
1 1 1
0 1 --> 11110000 = 240
0 0 0

Yes, but that's one pixel, so we'd count it like this:

[0,0,...., 1, 0, 0 ...].

The position of the 1 is the position corresponding to 240. It is not
position 240 because most of the values are unused (00001111 is
equivalent to 11110000 rotionally).

Your image has nine pixels, so you get a histogram which sums to 9.

--
HTH
Luis

Post by Solimyr
[2, 1, 6, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
I'm start to thinking that are two different stuff or from the second I
could get 240?
I'm sure that is a dummy question so don't worry about answering...
Thank You,
Solimyr
Il giorno mercoledì 1 maggio 2013 16:53:58 UTC-3, Luis Pedro Coelho ha
lbp() returns a histogram of features, so that positions X corresponds
to how many pixels had code X.
HTH
--
Luis Pedro Coelho | EMBL | http://luispedro.org

what I

Post by Solimyr
get, is a 36 vector-size with elements from zero to more than one. I
don't know if I'm stucked with some old theories of LBP or what.
Why don't I get a 8 vector size with elements from 0 to 1? My

idea is

Post by Solimyr
afterthat, convert in decimal and get the decimal value. Well, I'm
pretty confused with the result and the theory that I get from

internet.

Post by Solimyr
Could you please help me to understand?
Thank you very much for your support!
Solimyr
--
You received this message because you are subscribed to the Google
Groups "pythonvision" group.
To unsubscribe from this group and stop receiving emails from it,

send

Post by Solimyr
For more options, visit https://groups.google.com/groups/opt_out

<https://groups.google.com/groups/opt_out>.
--
You received this message because you are subscribed to the Google
Groups "pythonvision" group.
To unsubscribe from this group and stop receiving emails from it, send
For more options, visit https://groups.google.com/groups/opt_out.

Solimyr

2013-05-02 14:06:09 UTC

Thanks, I think that I'm starting to get it (I hope :D).

Post by Luis Pedro Coelho
[0,0,...., 1, 0, 0 ...].

So I get 240 for that pixel and I'm increasing the frenquency of that value
in my histogram. So, when I see on internet image like this:

Loading Image...

It's because they just use one pixel to make the new image, right? In
mahotas the LBP works with all the pixel in the window, e.g. 3x3 pixel use
the 9 pixel, right? Is this correct?

Thank you very much,
Solimyr

Il giorno giovedì 2 maggio 2013 03:45:15 UTC-3, Luis Pedro Coelho ha

[0,0,...., 1, 0, 0 ...].
The position of the 1 is the position corresponding to 240. It is not
position 240 because most of the values are unused (00001111 is
equivalent to 11110000 rotionally).
Your image has nine pixels, so you get a histogram which sums to 9.
--
HTH
Luis

corresponds

Post by Solimyr
to how many pixels had code X.
HTH
--
Luis Pedro Coelho | EMBL | http://luispedro.org

3x3

Post by Solimyr
window. So I take a 3x3 window and apply the mahotas lbp. I was
expecting a vector/histogram of 8 values with just 0 and 1 but,

what I

Post by Solimyr
get, is a 36 vector-size with elements from zero to more than

one. I

Post by Solimyr
don't know if I'm stucked with some old theories of LBP or what.
Why don't I get a 8 vector size with elements from 0 to 1? My

idea is

Post by Solimyr
afterthat, convert in decimal and get the decimal value. Well,

I'm

Post by Solimyr
pretty confused with the result and the theory that I get from

internet.

Post by Solimyr
Could you please help me to understand?
Thank you very much for your support!
Solimyr
--
You received this message because you are subscribed to the

Google

Post by Solimyr
Groups "pythonvision" group.
To unsubscribe from this group and stop receiving emails from it,

send

Post by Solimyr
For more options, visit https://groups.google.com/groups/opt_out

Luis Pedro Coelho

2013-05-03 09:41:25 UTC

Post by Solimyr
So I get 240 for that pixel and I'm increasing the frenquency of that
http://www.mathworks.com/matlabcentral/fx_files/37781/2/LBP_face.jpg
It's because they just use one pixel to make the new image, right? In
mahotas the LBP works with all the pixel in the window, e.g. 3x3 pixel
use the 9 pixel, right? Is this correct?

Solimyr

2013-05-03 13:30:37 UTC

Well, to answer your question, my point is that LBP value and conseguently
that image, represents in some way a new texture of that image.

Il giorno venerdì 3 maggio 2013 06:41:25 UTC-3, Luis Pedro Coelho ha

K.-Michael Aye

2013-05-03 14:48:52 UTC

Well, to answer your question, my point is that LBP value and conseguently that image, represents in some way a new texture of that image.

If you want a minimalistic but identifiable representation of your image, maybe you want a Radon transform? I don't know if mahotas has it, but scikit-image has it.

Michael

--
You received this message because you are subscribed to the Google Groups "pythonvision" group.
For more options, visit https://groups.google.com/groups/opt_out.

Solimyr

2013-05-03 15:29:04 UTC

Don't know what is a Radon transform, I will check it! Thank you!!!

Thanks,
Solimyr

Post by Solimyr
Well, to answer your question, my point is that LBP value and conseguently
that image, represents in some way a new texture of that image.
If you want a minimalistic but identifiable representation of your image,
maybe you want a Radon transform? I don't know if mahotas has it, but
scikit-image has it.
Michael

--
You received this message because you are subscribed to the Google Groups
"pythonvision" group.
To unsubscribe from this group and stop receiving emails from it, send an
For more options, visit https://groups.google.com/groups/opt_out.

Luis Pedro Coelho

2013-05-23 08:25:05 UTC

Hi,

I finally understood why this could be useful: in order to build a
pixel-level classifier, you might think it useful.

I just committed support for it:

https://github.com/luispedro/mahotas/commit/d01a02800548beaa03be49a4964812e45808ad2e

HTH
--
Luis Pedro Coelho | EMBL | http://luispedro.org

Post by Solimyr
Well, to answer your question, my point is that LBP value and
conseguently that image, represents in some way a new texture of that image.
Il giorno venerdì 3 maggio 2013 06:41:25 UTC-3, Luis Pedro Coelho ha

Post by Solimyr
So I get 240 for that pixel and I'm increasing the frenquency of

that
http://www.mathworks.com/matlabcentral/fx_files/37781/2/LBP_face.jpg
<http://www.mathworks.com/matlabcentral/fx_files/37781/2/LBP_face.jpg>

Post by Solimyr
It's because they just use one pixel to make the new image,

right? In

Post by Solimyr
mahotas the LBP works with all the pixel in the window, e.g. 3x3

pixel

Post by Solimyr
use the 9 pixel, right? Is this correct?

Ah, that kind of image with each pixel being assigned its LBP value
cannot be computed with mahotas (why would you want it? Serious
question, why would you like to have that image?)
What mahotas computes is the histogram of that image (except that
certain pixel values which are impossible are kept out of the histogram).
HTH
--
Luis Pedro Coelho | EMBL | http://luispedro.org
--
You received this message because you are subscribed to the Google
Groups "pythonvision" group.
To unsubscribe from this group and stop receiving emails from it, send
For more options, visit https://groups.google.com/groups/opt_out.

Solimyr

2013-05-23 12:35:44 UTC

Eheheheheh, that's what I was trying to say. What does 'I just committed
support' mean? My english is awful!!

Il giorno giovedì 23 maggio 2013 05:25:05 UTC-3, Luis Pedro Coelho ha

Post by Luis Pedro Coelho
Hi,
I finally understood why this could be useful: in order to build a
pixel-level classifier, you might think it useful.
https://github.com/luispedro/mahotas/commit/d01a02800548beaa03be49a4964812e45808ad2e
HTH
--
Luis Pedro Coelho | EMBL | http://luispedro.org

Post by Solimyr
Well, to answer your question, my point is that LBP value and
conseguently that image, represents in some way a new texture of that

image.

Post by Solimyr
Il giorno venerdì 3 maggio 2013 06:41:25 UTC-3, Luis Pedro Coelho ha

Post by Solimyr
So I get 240 for that pixel and I'm increasing the frenquency of

that

Post by Solimyr
value in my histogram. So, when I see on internet image like

http://www.mathworks.com/matlabcentral/fx_files/37781/2/LBP_face.jpg
<

http://www.mathworks.com/matlabcentral/fx_files/37781/2/LBP_face.jpg>

Post by Solimyr
It's because they just use one pixel to make the new image,

right? In

Post by Solimyr
mahotas the LBP works with all the pixel in the window, e.g. 3x3

pixel

Post by Solimyr
use the 9 pixel, right? Is this correct?

--
Luis Pedro Coelho | EMBL | http://luispedro.org

Luis Pedro Coelho

2013-05-23 12:38:41 UTC

Post by Solimyr
Eheheheheh, that's what I was trying to say. What does 'I just committed
support' mean? My english is awful!!

I meant to write

"I just pushed a git commit which includes support for this."

This means that it is available on the github version, but not in any
released version (ATM).

If you get it from https://github.com/luispedro/mahotas, you will get
this functionality.

It also means that it is considered "experimental" and open to changes
(for example, I have already changed the name of one of the parameters
from the initial commit, something I would never ever do if it was a
released version).

HTH
Luis

Post by Solimyr
Il giorno giovedì 23 maggio 2013 05:25:05 UTC-3, Luis Pedro Coelho ha
Hi,
I finally understood why this could be useful: in order to build a
pixel-level classifier, you might think it useful.
https://github.com/luispedro/mahotas/commit/d01a02800548beaa03be49a4964812e45808ad2e
<https://github.com/luispedro/mahotas/commit/d01a02800548beaa03be49a4964812e45808ad2e>
HTH
--
Luis Pedro Coelho | EMBL | http://luispedro.org

Post by Solimyr
Well, to answer your question, my point is that LBP value and
conseguently that image, represents in some way a new texture of

that image.

Post by Solimyr
Il giorno venerdì 3 maggio 2013 06:41:25 UTC-3, Luis Pedro Coelho ha

Post by Solimyr
So I get 240 for that pixel and I'm increasing the

frenquency of

Post by Solimyr
that

Post by Solimyr
value in my histogram. So, when I see on internet image

http://www.mathworks.com/matlabcentral/fx_files/37781/2/LBP_face.jpg
<http://www.mathworks.com/matlabcentral/fx_files/37781/2/LBP_face.jpg>
<http://www.mathworks.com/matlabcentral/fx_files/37781/2/LBP_face.jpg <http://www.mathworks.com/matlabcentral/fx_files/37781/2/LBP_face.jpg>>