| Re: Distinguishing between point-like and curve-like features [message #82810] |
Wed, 16 January 2013 02:11 |
Mats Löfdahl
Messages: 263
Registered: January 2012
|
Senior Member |
|
|
Den onsdagen den 16:e januari 2013 kl. 10:54:23 UTC+1 skrev karo03de:
> Am Dienstag, 15. Januar 2013 17:46:08 UTC+1 schrieb Mats Löfdahl:
>
>> Then I clean the mask with a morph_open operation to get rid of some raggedness of the edges of the "holes" in the mask.
>
>> This has worked fine, but I now have some images that have, in addition to the small features that I want to take out, also some high-amplitude curved fringes with similar width that I do _not_ want the mask to take out.
>
>
> What about applying top_hat operation, segmenting (threshold), (morph-) closing and size selection (histogram of labeled result) of the smaller conglomerates ? If that is not too time-consuming.
Yes, another good point. The curve-like features are usually larger in area so that makes for another criterion, once I have the regions.
Thanks.
/Mats
|
|
|
|
| Re: Distinguishing between point-like and curve-like features [message #82811 is a reply to message #82810] |
Wed, 16 January 2013 01:54  |
karo03de
Messages: 21
Registered: March 2007
|
Junior Member |
|
|
Am Dienstag, 15. Januar 2013 17:46:08 UTC+1 schrieb Mats Löfdahl:
> Then I clean the mask with a morph_open operation to get rid of some raggedness of the edges of the "holes" in the mask.
>
>
>
> This has worked fine, but I now have some images that have, in addition to the small features that I want to take out, also some high-amplitude curved fringes with similar width that I do _not_ want the mask to take out.
>
What about applying top_hat operation, segmenting (threshold), (morph-) closing and size selection (histogram of labeled result) of the smaller conglomerates ? If that is not too time-consuming.
Karsten
|
|
|
|
| Re: Distinguishing between point-like and curve-like features [message #82813 is a reply to message #82811] |
Wed, 16 January 2013 01:49 |
Mats Löfdahl
Messages: 263
Registered: January 2012
|
Senior Member |
|
|
Den tisdagen den 15:e januari 2013 kl. 17:54:24 UTC+1 skrev David Fanning:
> Mats Löfdahl writes:
>
>
>
>>
>
>> I'm making image masks that are supposed to take out certain small point-like features. These features are usually just a few pixels wide but sometimes as much as about ten or fifteen pixels wide. Because there are other, more large-scale variations in the image, I base the mask on a unsharp masked version of the image, to make the features I'm interested in stand out more. Then I clean the mask with a morph_open operation to get rid of some raggedness of the edges of
>
> the "holes" in the mask.
>
>>
>
>> This has worked fine, but I now have some images that have, in addition to the small features that I want to take out, also some high-amplitude curved fringes with similar width that I do _not_ want the mask to take out.
>
>>
>
>> I'm wondering if anyone can suggest a strategy for automatically (and reasonably fast) distinguishing between the point-like and the curve-like features. The masks I'm making either take both out or none, depending on the choice of parameters.
>
>>
>
>> I'm not looking for code here, but maybe some good ideas.
>
>>
>
>> (If you want the background, I'm trying to make bad-pixel masks for some CCD cameras based on flat fields that have significant interference fringes. The small features I want to mask out are clusters of bad pixels, both from the detector itself and due to near-focus dust particles. The purpose of the mask is to identify pixels with no information, where I have to interpolate to get useful values in the science images. Useful in the sense that those pixels do not cause
>
> ringing artifacts when I do deconvolution.)
>
>
>
> Maybe you could fit the features with an ellipse. It is possible, I
>
> suppose, that curved features will have more elliptical character than
>
> point-like features.
>
>
>
> http://www.idlcoyote.com/ip_tips/fit_ellipse.html
Good point! I've been thinking in terms of operations on the whole mask or image but of course I get more control if I do label_region and work on the holes one by one. Then checking for elongatedness shouldn't be too hard, by fitting ellipses or by other means.
Thanks!
/Mats
|
|
|
|
| Re: Distinguishing between point-like and curve-like features [message #82832 is a reply to message #82813] |
Tue, 15 January 2013 08:54 |
David Fanning
Messages: 11724
Registered: August 2001
|
Senior Member |
|
|
Mats Löfdahl writes:
>
> I'm making image masks that are supposed to take out certain small point-like features. These features are usually just a few pixels wide but sometimes as much as about ten or fifteen pixels wide. Because there are other, more large-scale variations in the image, I base the mask on a unsharp masked version of the image, to make the features I'm interested in stand out more. Then I clean the mask with a morph_open operation to get rid of some raggedness of the edges of
the "holes" in the mask.
>
> This has worked fine, but I now have some images that have, in addition to the small features that I want to take out, also some high-amplitude curved fringes with similar width that I do _not_ want the mask to take out.
>
> I'm wondering if anyone can suggest a strategy for automatically (and reasonably fast) distinguishing between the point-like and the curve-like features. The masks I'm making either take both out or none, depending on the choice of parameters.
>
> I'm not looking for code here, but maybe some good ideas.
>
> (If you want the background, I'm trying to make bad-pixel masks for some CCD cameras based on flat fields that have significant interference fringes. The small features I want to mask out are clusters of bad pixels, both from the detector itself and due to near-focus dust particles. The purpose of the mask is to identify pixels with no information, where I have to interpolate to get useful values in the science images. Useful in the sense that those pixels do not cause
ringing artifacts when I do deconvolution.)
Maybe you could fit the features with an ellipse. It is possible, I
suppose, that curved features will have more elliptical character than
point-like features.
http://www.idlcoyote.com/ip_tips/fit_ellipse.html
Cheers,
David
--
David Fanning, Ph.D.
Fanning Software Consulting, Inc.
Coyote's Guide to IDL Programming: http://www.idlcoyote.com/
Sepore ma de ni thue. ("Perhaps thou speakest truth.")
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
