On Dec 1, 1:38=A0am, Eizo <odpe...@[EMAIL PROTECTED]
> wrote:
> On Nov 30, 2:18=A0pm, Manuel Matias <manuel.mat...@[EMAIL PROTECTED]
> wrote:
>
>
>
> > On Nov 29, 9:41=A0pm, Eizo <odpe...@[EMAIL PROTECTED]
> wrote:
>
> > > On Nov 29, 10:19=A0pm, serg271 <serg...@[EMAIL PROTECTED]
> wrote:
>
> > > > On Nov 29, 7:00=A0pm, Eizo <odpe...@[EMAIL PROTECTED]
> wrote:> Hi,
> > > > > I am looking for some (fast) algorithm to find circles in image.
> > > > > I understand that there is the usual Matlab routine that use
Hugh
> > > > > transform, but it is slow, and I need some other fast solution.
> > > > > any ideas?
>
> > > > I would suggest just brute-force it. Make edge detection, or
> > > > thresholding/segmentation. After that for each contour (closed
edge=
or
> > > > segment boundary) check if it circle or not - for example
calculate
> > > > center (average) and average radial error. Works pretty well for
> > > > convex polygons.
>
> > > Well, I am using a picture, that isn't super clear (e.g could be
> > > places where the edge isn't closed )
> > > also I think that computing whatever a contour is closed or not is
> > > very time consuming ( correct me if I Am wrong here..)
> > > and I need something that work =A0very fast.
>
> > You don't really need to check if the contour is closed.
> > The strategy Eizo suggested will work well on a contour
> > that is _almost_ closed, lacking a small segment
> > (but not if there is a large segment missing or if the
> > circunference is broken in two or more places).
>
> > Also, you can improve the results in several ways:
> > - Preprocess the original image to remove noise
> > - Choose well the edge detector that is better suited
> > =A0 for your task (sounds like Canny's algorithm might
> > =A0 help).
> > - After segmentation, use morphological operations
> > =A0 (closing should be useful) to have more compact,
> > =A0 better connected components.
>
> > Also, posting an example image will help us with
> > the specifics of your problem.
>
> Thanks,
>
> the work concentrate mainly on road signs. as such it has to be fast,
> also the image is not very good quality, but you can imagine a gray
> scale image of a sign, and =A0a little bit of background.
>
> I will try to arrange some good sample.
> in the meanwhile if you have more ideas I would love to hear.
>
> E
I think hough transformation combined with RANSAC algorithm will
accelerate the speed of execution.
google random hough transformation and RANSAC may help you.
|
