Re: Image detection
- From: "moodamultani" <siddiqi.mahmood@xxxxxxxxx>
- Date: 24 Apr 2006 06:46:13 -0700
Hi SaM,
[As it may be relevant to what your doing,] May I just mention here
that detecting text in images (like car number plates) has been
researched since quite a few years? Several factors that differentiate
the text from the non-textual part of the image include:
1. the fact that text is distinct from its background (Of course, this
is why you obtain the edge image).
2. Text components mostly have dimesions that are fairly proportional.
For example, I do not expect a text component to have a width of 5
pixels and length of 500 pixels.
2. the problem of discarding the non-textual components of the image
may be solved by deleting the connected components that are too large
or too small. One way is to look for the mean area of all connected
components and reject those whose area is greater than 3 times the
standard deviation greater (or smaller) the mean area. Before starting
the recognition part, one can reject these unwanted components and
obtain the 'candidate' textual components. I have tried this and am
fairly satisfied before I can find a better scheme.
3. There may also be an issue of skew in the image or characters may be
slanted. How can we take care of that?
4. Illumination problems can also affect results.
HTH,
Regards,
MA
.
- Follow-Ups:
- Re: Image detection
- From: SaM
- Re: Image detection
- References:
- Image detection
- From: SaM
- Re: Image detection
- From: Greg Heath
- Re: Image detection
- From: SaM
- Re: Image detection
- From: Greg Heath
- Re: Image detection
- From: SaM
- Image detection
- Prev by Date: Re: IIR Filter in Matlab
- Next by Date: Re: genetic algorithm for an objective function under certain constraints
- Previous by thread: Re: Image detection
- Next by thread: Re: Image detection
- Index(es):
Relevant Pages
|