bwtraceboundary in MATLAB is it the ultimate soultion help

Re: bwtraceboundary in MATLAB , is it the ultimate soultion ? (help needed)

> But if for example I don't have any problems of disconnected  dots  ?
> and all my contours are connected as it should, is there any need to
> go to Hugh transform ?
> Is there other reasons I am missing here?
>
> Thanks
> E

Why don't you say what you are really trying to do instead of asking
these open-ended questions?

Re: bwtraceboundary in MATLAB , is it the ultimate soultion ? (help needed)

On Dec 20, 3:38=A0am, Bob  wrote:
> > But if for example I don't have any problems of disconnected =A0dots =
=A0?
> > and all my contours are connected as it should, is there any need to
> > go to Hugh transform ?
> > Is there other reasons I am missing here?
>
> > Thanks
> > E
>
> Why don't you say what you are really trying to do instead of asking
> these open-ended questions?

No problem, I just didn';t want to bother other people who already
read the other thread.

In any case, I am looking for a fast method to capture, road signs, it
seems there are many  anachronistic ways to do that, which are spread
all over the net  with some academic papers, but yesterday I found out
that in one line in Matlab you can solve this problem very fast.
I wanted to make sure I am not missing something....

my pictures are BW, I don't think I have a discontinuities in the
boundaries, and it seems that matlab is doing it blazing fast, so
unless I am missing something very crucial here , it looks like a good
fast method.

It will be interesting to know, though, what method they are using in
order to implement the bwboundaries.
If anyone know, I would be glad to read it .

Re: bwtraceboundary in MATLAB , is it the ultimate soultion ? (help needed)

On Dec 20, 5:04=A0am, Eizo  wrote:
> On Dec 20, 3:38=A0am, Bob  wrote:
>
> > > But if for example I don't have any problems of disconnected =A0dots =
=A0?
> > > and all my contours are connected as it should, is there any need to
> > > go to Hugh transform ?
> > > Is there other reasons I am missing here?
>
> > > Thanks
> > > E
>
> > Why don't you say what you are really trying to do instead of asking
> > these open-ended questions?
>
> No problem, I just didn';t want to bother other people who already
> read the other thread.
>
> In any case, I am looking for a fast method to capture, road signs, it
> seems there are many =A0anachronistic ways to do that, which are spread
> all over the net =A0with some academic papers, but yesterday I found out
> that in one line in Matlab you can solve this problem very fast.
> I wanted to make sure I am not missing something....
>
> my pictures are BW, I don't think I have a discontinuities in the
> boundaries, and it seems that matlab is doing it blazing fast, so
> unless I am missing something very crucial here , it looks like a good
> fast method.
>
> It will be interesting to know, though, what method they are using in
> order to implement the bwboundaries.
> If anyone know, I would be glad to read it .

---------------------------------------------------------------------------=
---------------------------------------
Eizo:
Well just look at the source code!  That will tell you, in
excruciating detail, exactly what they are doing.  Just put the cursor
on the word bwboundaries in your source code, then type control-d to
bring up the source code of bwboundaries in its own window.  The code
for all or most of the image processing toolkit is available this way.
Regards,
ImageAnalyst

Re: bwtraceboundary in MATLAB , is it the ultimate soultion ? (help needed)

On Dec 20, 9:43 am, ImageAnalyst  wrote:

> Well just look at the source code!  That will tell you, in
> excruciating detail, exactly what they are doing.

If you look in the source code, it turns out to be implemented as a
mex dll i.e. compiled code with no source code provided:
>B = bwtraceboundarymex(BW,P,CONN,FSTEP,N,DIR);

This is a real problem with Matlab. It is essentially a closed
environment. Heaven help you if you want to port your code to
distribute it.

On Dec 21, 1:51=A0pm, Bob  wrote:
> On Dec 20, 9:43 am, ImageAnalyst  wrote:
>
> > Well just look at the source code! =A0That will tell you, in
> > excruciating detail, exactly what they are doing.
>
> If you look in the source code, it turns out to be implemented as a
> mex dll i.e. compiled code with no source code provided:
>
> >B =3D bwtraceboundarymex(BW,P,CONN,FSTEP,N,DIR);
>
> This is a real problem with Matlab. It is essentially a closed
> environment. Heaven help you if you want to port your code to
> distribute it.

--------------------------------------------------------------------
Why "heaven help you"?  If it works fine for you then why does it
matter that you can't see every source code line.  The Mathworks
usually gives references for papers if they're following a published
algorithm.  I don't know about you but I've used many libraries over
the past 30 years and if the library routine does it's job then I
don't feel any urge to dig into it further.  Why waste my time?  I bet
lots of us use libraries all the time without concern as to what's in
them.  Even if you're using open source code libraries, how many of us
actually dig into them to see how they do an erosion, or a histogram,
or whatever?  If it works for you, then you just use it without
worry.  If you want to distribute your code, then MATLAB lets you
distribute it royalty free and it's pretty easy - just two files the
run-time installer, and your executable.  So easy that I don't need
any help from heaven to do it.
ImageAnalyst

On Dec 22, 2:50=A0am, ImageAnalyst  wrote:
> On Dec 21, 1:51=A0pm, Bob  wrote:
>
> > On Dec 20, 9:43 am, ImageAnalyst  wrote:
>
> > > Well just look at the source code! =A0That will tell you, in
> > > excruciating detail, exactly what they are doing.
>
> > If you look in the source code, it turns out to be implemented as a
> > mex dll i.e. compiled code with no source code provided:
>
> > >B =3D bwtraceboundarymex(BW,P,CONN,FSTEP,N,DIR);
>
> > This is a real problem with Matlab. It is essentially a closed
> > environment. Heaven help you if you want to port your code to
> > distribute it.
>
> --------------------------------------------------------------------
> Why "heaven help you"? =A0If it works fine for you then why does it
> matter that you can't see every source code line. =A0The Mathworks
> usually gives references for papers if they're following a published
> algorithm. =A0I don't know about you but I've used many libraries over
> the past 30 years and if the library routine does it's job then I
> don't feel any urge to dig into it further. =A0Why waste my time? =A0I be=
t
> lots of us use libraries all the time without concern as to what's in
> them. =A0Even if you're using open source code libraries, how many of us
> actually dig into them to see how they do an erosion, or a histogram,
> or whatever? =A0If it works for you, then you just use it without
> worry. =A0If you want to distribute your code, then MATLAB lets you
> distribute it royalty free and it's pretty easy - just two files the
> run-time installer, and your executable. =A0So easy that I don't need
> any help from heaven to do it.
> ImageAnalyst

Pretty simple . I need the code, because I am using it in my thesis,
and  I want to be able to explain (in case someone will ask) , why I
choose this algorithm   over the other one exist in the  academic
paper , and I want to make sure that the algorithm they are using is
ok, and was the best solution I can come up with.
Also plain curiosity.
.
you are right in the case when someone is programming a code in some
company and they don't care on "how", they just focus on the : " make
it work"

Btw, I also saw that it was compiled.
and to start digging into the CPP code, that's a whole new ball game.
I just hope they care to put some comments :)

On Dec 21, 4:50 pm, ImageAnalyst  wrote:

> Why "heaven help you"?  If it works fine for you then why does it
> matter that you can't see every source code line.  The Mathworks
> usually gives references for papers if they're following a published
> algorithm.  I don't know about you but I've used many libraries over
> the past 30 years and if the library routine does it's job then I
> don't feel any urge to dig into it further.  Why waste my time?  I bet
> lots of us use libraries all the time without concern as to what's in
> them.

Well, it looks like you have never tried to make a shippable i.e.
salable product based on your work. You need to port the code to
another language since the matlab compiler is a joke. It requires an
~100 Mbyte dll,  the dll changes every time they release a new version
and the different versions are incompatible, and the execution time is
no faster than the matlab code.

Even if you are an academic, some researchers want their code to be
open and reproducible. After all, that is the purpose of all
publications, to show others what you did and how you did it so they
can reproduce it and see for themselves. As the OP noted, you cannot
reproduce a black box no matter how satisfied YOU are with it.

And as to references, most publications, in the words of Jon Claerbout
of Stanford, are 'advertisements' and not sufficient to reproduce the
results. The scholarship in software research (and I include image
processing in this topic) is the code itself. It is almost impossible
to reproduce results based on the description in a paper. Suppose 200
years from  now, someone wants to reproduce an algorithm that you
documented as matlab code. Mathworks has gone the way of all companies
and is defunct and gone with it is matlab. Your results are
essentially useless.

Steve Eddins
> We ship the C and C++ source code for almost all of the MEX-files in the Image Processing Toolbox.
Thanks. I had not noticed that. Do you provide source for other
functions such as rand? I have not followed up the references but see
my comments above about reproducible research.

Bob

On Dec 22, 10:08=A0pm, Bob  wrote:
> On Dec 21, 4:50 pm, ImageAnalyst  wrote:
>
> > Why "heaven help you"? =A0If it works fine for you then why does it
> > matter that you can't see every source code line. =A0The Mathworks
> > usually gives references for papers if they're following a published
> > algorithm. =A0I don't know about you but I've used many libraries over
> > the past 30 years and if the library routine does it's job then I
> > don't feel any urge to dig into it further. =A0Why waste my time? =A0I =
bet
> > lots of us use libraries all the time without concern as to what's in
> > them.
>
> Well, it looks like you have never tried to make a shippable i.e.
> salable product based on your work. You need to port the code to
> another language since the matlab compiler is a joke. It requires an
> ~100 Mbyte dll, =A0the dll changes every time they release a new version
> and the different versions are incompatible, and the execution time is
> no faster than the matlab code.
>
> Even if you are an academic, some researchers want their code to be
> open and reproducible. After all, that is the purpose of all
> publications, to show others what you did and how you did it so they
> can reproduce it and see for themselves. As the OP noted, you cannot
> reproduce a black box no matter how satisfied YOU are with it.
>
> And as to references, most publications, in the words of Jon Claerbout
> of Stanford, are 'advertisements' and not sufficient to reproduce the
> results. The scholarship in software research (and I include image
> processing in this topic) is the code itself. It is almost impossible
> to reproduce results based on the description in a paper. Suppose 200
> years from =A0now, someone wants to reproduce an algorithm that you
> documented as matlab code. Mathworks has gone the way of all companies
> and is defunct and gone with it is matlab. Your results are
> essentially useless.

Exactly.

Bob wrote:
> On Dec 20, 9:43 am, ImageAnalyst  wrote:
> 
>> Well just look at the source code!  That will tell you, in
>> excruciating detail, exactly what they are doing.
> 
> If you look in the source code, it turns out to be implemented as a
> mex dll i.e. compiled code with no source code provided:
>> B = bwtraceboundarymex(BW,P,CONN,FSTEP,N,DIR);
> 
> This is a real problem with Matlab. It is essentially a closed
> environment. Heaven help you if you want to port your code to
> distribute it.

Keep looking.  In the folder where MATLAB is installed, see the file 
toolbox\images\images\private\bwtraceboundarymex.cpp.

We ship the C and C++ source code for almost all of the MEX-files in the 
Image Processing Toolbox.

---
Steve Eddins
http://blogs.mathworks.com/steve/

On Dec 19, 5:00=A0pm, Eizo  wrote:
> On Dec 19, 11:06=A0pm, "Mark Andrews (The Other One)"
>
>  wrote:
> > Eizo wrote:
> > > I wonder why to use the Hugh transform, or any other complicated
> > > mathematical method to extract boundaries of shapes, while one can
> > > use this simple function..
>
> > The Hough and Radon transforms have applications that cannot be handled
> > by boundary tracers. Specifically, boundary tracing routines usually
> > require the boundary pixels to be contiguous (no breaks) whereas the
> > Hough transform can extract useful line information even when the pixel=
s
> > are not contiguous. For example, if you have a sequence of disconnected
> > dots in a straight line then Hough is useful, boundary tracing is not.
>
> > --
> > Mark.
>
> But if for example I don't have any problems of disconnected =A0dots =A0?
> and all my contours are connected as it should, is there any need to
> go to Hugh transform ?
> Is there other reasons I am missing here?
>
> Thanks
> E

---------------------------------------------------------
E:
If you have a binary image with contiguous blobs, then yes, you can
use the bwboundaries function in MATLAB.  The Hough (not Hugh)
transform is one possible method that you can use on your way to
getting a binary image with blobs in it.  Yes, you are missing
something, but continued learning in the area will benefit you.  Don't
worry, everyone starts out in the area with no knowledge at all and
gains more and more as the years go by.
Regards,
ImageAnalyst