Sign in with
Sign up | Sign in
Your question

Image Restoration to improve image detail

Last response: in Digital Camera
Share
Anonymous
January 8, 2005 8:43:40 PM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

HI
I have put together a web page showing the improvement
one can get with Adaptive Richardson-Lucy image restoration,
and comparison to Unsharp Mask methods. Please take a look
at: http://clarkvision.com/imagedetail/image-restoration1
and let me know what you think. I include a link on the
page to the original cropped 16-bit tif file in case
you have a method that can improve over what I present.
If you do make an improvement, I am interested in the results
and how you did it.

Roger Clark
Anonymous
January 9, 2005 1:22:16 AM

Archived from groups: rec.photo.digital (More info?)

On Sat, 08 Jan 2005 17:43:40 -0700, "Roger N. Clark (change username
to rnclark)" <username@qwest.net> wrote:

>HI
>I have put together a web page showing the improvement
>one can get with Adaptive Richardson-Lucy image restoration,
>and comparison to Unsharp Mask methods. ...

Your discussion of sharpening by deconvolution is indeed informative
and valuable. Along with your example, you asked whether others could
achieve equal or superior sharpness. I can report that "Image
Analyzer" (http://meesoft.logicnet.dk/Analyzer/index2.php), which is
freeware for Win32, can indeed provide equal quality, and do so rather
quickly. For the small section of the Red Fox image made available,
choosing Filter | "Image restoration by deconvolution", "Model:
circular blur", Radius 1.5, and "Do iterations: 5", gave a result that
I judged similar to yours. Choosing "Model: Gaussian blur" with
slightly different settings also worked well.

It might be worth noting that digital camera images are often less
noisy that ones scanned from negative film, and because of that yield
significant improvements in apparent sharpness through deconvolution.
Images with noise do not sharpen well.
January 9, 2005 3:36:02 PM

Archived from groups: rec.photo.digital (More info?)

James Kendall wrote:

> On Sat, 08 Jan 2005 17:43:40 -0700, "Roger N. Clark (change username
> to rnclark)" <username@qwest.net> wrote:
>
>
>>HI
>>I have put together a web page showing the improvement
>>one can get with Adaptive Richardson-Lucy image restoration,
>>and comparison to Unsharp Mask methods. ...
>
>
> Your discussion of sharpening by deconvolution is indeed informative
> and valuable. Along with your example, you asked whether others could
> achieve equal or superior sharpness. I can report that "Image
> Analyzer" (http://meesoft.logicnet.dk/Analyzer/index2.php), which is
> freeware for Win32, can indeed provide equal quality, and do so rather
> quickly. For the small section of the Red Fox image made available,
> choosing Filter | "Image restoration by deconvolution", "Model:
> circular blur", Radius 1.5, and "Do iterations: 5", gave a result that
> I judged similar to yours. Choosing "Model: Gaussian blur" with
> slightly different settings also worked well.

Thanks, that's a very cool program (though I did crash my computer
running it <g>). So is that the same thing Roger discusses? Lots of
different options for various techniques of sharpening, noise reduction,
etc. I wonder if NeatImage is still better for noise reduction though
because it uses the camera's profile.

I didn't see what program they used for the Adaptive Richardson-Lucy
image restoration.
Related resources
Anonymous
January 9, 2005 5:35:27 PM

Archived from groups: rec.photo.digital (More info?)

paul wrote:

> James Kendall wrote:
>
>> On Sat, 08 Jan 2005 17:43:40 -0700, "Roger N. Clark (change username
>> to rnclark)" <username@qwest.net> wrote:
>>
>>
>>> HI
>>> I have put together a web page showing the improvement
>>> one can get with Adaptive Richardson-Lucy image restoration,
>>> and comparison to Unsharp Mask methods. ...

> I didn't see what program they used for the Adaptive Richardson-Lucy
> image restoration.

I used ImagesPlus: http://www.mlunsold.com
It is a full image processing system.

Roger
Anonymous
January 10, 2005 2:39:22 AM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

Hi All:

The ideal is interesting....I am thinking whether it is possible to restore
images from motion blur or hand shake if we could model the bluring (i.e. by
observing a point source in the image), provided the noise it not so high
and the we are not overdemanding.

And...should it be easly to introduce DOF I guess ?

Regards

David Zou

"Roger N. Clark (change username to rnclark)" <username@qwest.net> wrote in
message news:41E07E3C.4060207@qwest.net...
> HI
> I have put together a web page showing the improvement
> one can get with Adaptive Richardson-Lucy image restoration,
> and comparison to Unsharp Mask methods. Please take a look
> at: http://clarkvision.com/imagedetail/image-restoration1
> and let me know what you think. I include a link on the
> page to the original cropped 16-bit tif file in case
> you have a method that can improve over what I present.
> If you do make an improvement, I am interested in the results
> and how you did it.
>
> Roger Clark
>
Anonymous
January 10, 2005 2:39:23 AM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

David Zou wrote:

> Hi All:
>
> The ideal is interesting....I am thinking whether it is possible to restore
> images from motion blur or hand shake if we could model the bluring (i.e. by
> observing a point source in the image), provided the noise it not so high
> and the we are not overdemanding.

Yes. The ImagesPlus software allows you to put in a custom
point spread function, up to 9x9 pixels. You can do this by
hand or click on a smeared point source in an image.
I haven't needed it though, so I haven't tried it.
(Not that I don't get lots of motion blurred images ;-)

> And...should it be easly to introduce DOF I guess ?

I don't know how you would do this.

Roger

> Regards
>
> David Zou
>
> "Roger N. Clark (change username to rnclark)" <username@qwest.net> wrote in
> message news:41E07E3C.4060207@qwest.net...
>
>>HI
>>I have put together a web page showing the improvement
>>one can get with Adaptive Richardson-Lucy image restoration,
>>and comparison to Unsharp Mask methods. Please take a look
>>at: http://clarkvision.com/imagedetail/image-restoration1
>>and let me know what you think. I include a link on the
>>page to the original cropped 16-bit tif file in case
>>you have a method that can improve over what I present.
>>If you do make an improvement, I am interested in the results
>>and how you did it.
>>
>>Roger Clark
>>
>
>
>
Anonymous
January 10, 2005 4:08:20 AM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

Hi

Sorry pls forgot that....

Since the out of focus will blur the image via some defined function and it
is possible to restore it somehow (true of not?)...I guess it accordingly.
But I realise there is no way for the software to know the corresponding
distance of a point on the image....

Unless we define some coordinate in the image and the software work out
every point by interpolate.......but this seems to be neither practical and
nor advantageous.


Regards

David Zou

"Roger N. Clark (change username to rnclark)" <username@qwest.net> wrote in
message news:41E1CCB6.9080102@qwest.net...
> David Zou wrote:
>
>> Hi All:
>>
>> The ideal is interesting....I am thinking whether it is possible to
>> restore images from motion blur or hand shake if we could model the
>> bluring (i.e. by observing a point source in the image), provided the
>> noise it not so high and the we are not overdemanding.
>
> Yes. The ImagesPlus software allows you to put in a custom
> point spread function, up to 9x9 pixels. You can do this by
> hand or click on a smeared point source in an image.
> I haven't needed it though, so I haven't tried it.
> (Not that I don't get lots of motion blurred images ;-)
>
>> And...should it be easly to introduce DOF I guess ?
>
> I don't know how you would do this.
>
> Roger
>
>> Regards
>>
>> David Zou
>>
>> "Roger N. Clark (change username to rnclark)" <username@qwest.net> wrote
>> in message news:41E07E3C.4060207@qwest.net...
>>
>>>HI
>>>I have put together a web page showing the improvement
>>>one can get with Adaptive Richardson-Lucy image restoration,
>>>and comparison to Unsharp Mask methods. Please take a look
>>>at: http://clarkvision.com/imagedetail/image-restoration1
>>>and let me know what you think. I include a link on the
>>>page to the original cropped 16-bit tif file in case
>>>you have a method that can improve over what I present.
>>>If you do make an improvement, I am interested in the results
>>>and how you did it.
>>>
>>>Roger Clark
>>>
>>
>>
>>
>
Anonymous
January 10, 2005 5:34:26 AM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

"David Zou" <xzous@yahoo.co.uk> wrote in message
news:crskib$1r4$1@gemini.csx.cam.ac.uk...
> Hi
>
> Sorry pls forgot that....
>
> Since the out of focus will blur the image via some defined function
> and it is possible to restore it somehow (true of not?)...I guess it
> accordingly. But I realise there is no way for the software to know
> the corresponding distance of a point on the image....

It would require at least two images focussed at different distances,
or a user supplied depth map.

Bart
Anonymous
January 10, 2005 3:16:18 PM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

Roger

Trying to throw another couple of variables in your tests. I would think to
get the absolute best sharpness in an image, depends on how you processed
the RAW image and maybe what your in camera settings were.

I know the RAW files hold the all original DATA, but if you straight process
them. They will process with the in-camera sharpness settings.

I am sure it is the same with Canon RAW files, but with Nikon RAW and the
application "Nikon Capture" you can change the "Sharpness" using the pure
RAW data. It has six different settings from "None" to "High." There is a
lot of difference in these settings.

Even Photoshop CS "Camera RAW" will allow different sharpening values when
importing RAW Files.

I believe I have read somewhere that either of these RAW actions should
yield better results than any after sharpening actions.

In my own "very limited" tests this appeared to be true. And I use it to
varying degrees on most of my images.

Have you put these different options into your tests?

PWW
--
PWW (Paul Wayne Wilson)
Over 1,000 Photographs Online at,
http://PhotoStockFile.com



> On 1/8/05 7:43 PM, in article 41E07E3C.4060207@qwest.net, "Roger N. Clark
> (change username to rnclark)" <username@qwest.net> wrote:

> HI
> I have put together a web page showing the improvement
> one can get with Adaptive Richardson-Lucy image restoration,
> and comparison to Unsharp Mask methods. Please take a look
> at: http://clarkvision.com/imagedetail/image-restoration1
> and let me know what you think. I include a link on the
> page to the original cropped 16-bit tif file in case
> you have a method that can improve over what I present.
> If you do make an improvement, I am interested in the results
> and how you did it.
>
> Roger Clark
>
Anonymous
January 10, 2005 5:08:46 PM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

"Roger N. Clark (change username to rnclark)" <username@qwest.net>
wrote in message news:41E07E3C.4060207@qwest.net...
> HI
> I have put together a web page showing the improvement
> one can get with Adaptive Richardson-Lucy image restoration,
> and comparison to Unsharp Mask methods.

Great, I've been looking into similar things, including various
resampling methods, and came to similar conclusions.
An example of R-L restoraton on pictorial (terrestrial) imaging
can be found here:
<http://heim.ifi.uio.no/~gisle/photo/interpolation.html&...; .

> Please take a look at:
> <http://clarkvision.com/imagedetail/image-restoration1&g...;
> and let me know what you think. I include a link on the
> page to the original cropped 16-bit tif file in case
> you have a method that can improve over what I present.
> If you do make an improvement, I am interested in the results
> and how you did it.

Restoration will also try to restore square pixels into jaggies, so
the interpolation method used for going to 200% becomes more
important as well.
If you start with a (depending on the amount of rescaling) step
interpolation of 10% per resampled image to reach the 2x
enlargement, or one of the better interpolation methods used by
Qimage, the results may improve a bit more, depending on
image content.

Different rescaling, will probably call for modified R-L parameters,
but this is what happens when using the same R-L PSF 7x7 20
iterations as you used, but now on a Qimage "Pyramid"
interpolated version:
<http://www.xs4all.nl/~bvdwolf/temp/foxQP+IP.png&gt;
or after rescaling to 8x and then down to 2x:
<http://www.xs4all.nl/~bvdwolf/temp/foxQP8to2+IP.png&gt;
which reduces jaggies even further, but may lose too much
resolution, depending on the subject. You can see it produces
slightly less pronounced jaggies, and a bit more detail in the
almost mushy areas.

Here is the 2x Qimage interpolated version before restoration /
sharpening: <http://www.xs4all.nl/~bvdwolf/temp/foxQP.png&gt;
if you want to try different R_L settings (which I sometimes
follow-up with an IP Multiresolution Finest step).

I think Qimage's "Pyramid" method would do well on most
subjects in general, while "Vector" is better for keeping smooth
gradients. And of course, Qimage can also be used to prepare
final 8-bpc output for the Lightjet.

The Adaptive Richardson-Lucy restoration is a time-consuming,
but worthwhile operation if you want to squeeze the final bits of
information out of the data.

Bart
Anonymous
January 10, 2005 5:56:03 PM

Archived from groups: rec.photo.digital, rec.photo.technique.nature (More info?)

David Zou wrote:

> The ideal is interesting....I am thinking whether it is possible to
restore
> images from motion blur or hand shake if we could model the bluring
(i.e. by
> observing a point source in the image), provided the noise it not so
high
> and the we are not overdemanding.

A while ago, I de-blurred the graphic on the side of a moving van from
a 35 pixel long smear to the point of readability. The result was
useful for "forensic" work, but little else. The main problem was PSF
estimation and noise.

> And...should it be easly to introduce DOF I guess ?

Probably not, given that the various deconvolution codes (RL, etc)
typically assume a spatially invariant PSF.
Anonymous
January 11, 2005 12:47:25 AM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

PWW wrote:

> Roger
>
> Trying to throw another couple of variables in your tests. I would think to
> get the absolute best sharpness in an image, depends on how you processed
> the RAW image and maybe what your in camera settings were.
>
> I know the RAW files hold the all original DATA, but if you straight process
> them. They will process with the in-camera sharpness settings.
>
> I am sure it is the same with Canon RAW files, but with Nikon RAW and the
> application "Nikon Capture" you can change the "Sharpness" using the pure
> RAW data. It has six different settings from "None" to "High." There is a
> lot of difference in these settings.
>
> Even Photoshop CS "Camera RAW" will allow different sharpening values when
> importing RAW Files.
>
> I believe I have read somewhere that either of these RAW actions should
> yield better results than any after sharpening actions.
>
> In my own "very limited" tests this appeared to be true. And I use it to
> varying degrees on most of my images.
>
> Have you put these different options into your tests?
>
> PWW

Paul,
You are correct, that does add more variables. Also to be added:
resampling to a larger image at the raw conversion step.
I experimented with these two things a little using the
photoshop converter. The initial images are quite different
when coming out of this step. However, I was quite surprised
when the adaptive Richardson-Lucey algorithm produced
esentially the same results for the final output regardless
of the quite variable input, even down to the same
artifacts. I could make the photoshop CS output closer to
the adaptive Richardson-Lucey result. Perhaps a combination
of the two will result in better processing time, allowing one
to reduce the adaptive Richardson-Lucey very long compute time.

Thanks for the suggestion!

Roger
Anonymous
January 11, 2005 12:51:07 AM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

Bart van der Wolf wrote:

>
> "Roger N. Clark (change username to rnclark)" <username@qwest.net> wrote
> in message news:41E07E3C.4060207@qwest.net...
>
>> HI
>> I have put together a web page showing the improvement
>> one can get with Adaptive Richardson-Lucy image restoration,
>> and comparison to Unsharp Mask methods.
>
>
> Great, I've been looking into similar things, including various
> resampling methods, and came to similar conclusions. An example of R-L
> restoraton on pictorial (terrestrial) imaging can be found here:
> <http://heim.ifi.uio.no/~gisle/photo/interpolation.html&...; .
>
>> Please take a look at:
>> <http://clarkvision.com/imagedetail/image-restoration1&g...;
>> and let me know what you think. I include a link on the
>> page to the original cropped 16-bit tif file in case
>> you have a method that can improve over what I present.
>> If you do make an improvement, I am interested in the results
>> and how you did it.
>
>
> Restoration will also try to restore square pixels into jaggies, so the
> interpolation method used for going to 200% becomes more important as
> well. If you start with a (depending on the amount of rescaling) step
> interpolation of 10% per resampled image to reach the 2x enlargement, or
> one of the better interpolation methods used by Qimage, the results may
> improve a bit more, depending on image content.
>
> Different rescaling, will probably call for modified R-L parameters, but
> this is what happens when using the same R-L PSF 7x7 20 iterations as
> you used, but now on a Qimage "Pyramid" interpolated version:
> <http://www.xs4all.nl/~bvdwolf/temp/foxQP+IP.png&gt; or after rescaling to
> 8x and then down to 2x:
> <http://www.xs4all.nl/~bvdwolf/temp/foxQP8to2+IP.png&gt;
> which reduces jaggies even further, but may lose too much resolution,
> depending on the subject. You can see it produces slightly less
> pronounced jaggies, and a bit more detail in the almost mushy areas.
>
> Here is the 2x Qimage interpolated version before restoration /
> sharpening: <http://www.xs4all.nl/~bvdwolf/temp/foxQP.png&gt; if you want
> to try different R_L settings (which I sometimes follow-up with an IP
> Multiresolution Finest step).
>
> I think Qimage's "Pyramid" method would do well on most subjects in
> general, while "Vector" is better for keeping smooth gradients. And of
> course, Qimage can also be used to prepare final 8-bpc output for the
> Lightjet.
>
> The Adaptive Richardson-Lucy restoration is a time-consuming, but
> worthwhile operation if you want to squeeze the final bits of
> information out of the data.
> Bart

Bart,
Thanks. Very interesting.
May I add these images to my page? I'll credit you
unless you request otherwise.

Roger
Anonymous
January 11, 2005 1:18:34 AM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

"Roger N. Clark (change username to rnclark)" <username@qwest.net> wrote

> I have put together a web page showing the improvement
> one can get with Adaptive Richardson-Lucy image restoration,
> and comparison to Unsharp Mask methods. Please take a look
> at: http://clarkvision.com/imagedetail/image-restoration1

TTTT, I like the original: all the sharpened ones look overly
pixelated and make the fox look like it needs to use a creme
rinse after it shampoos.

--
Nicholas O. Lindan, Cleveland, Ohio
Consulting Engineer: Electronics; Informatics; Photonics.
Remove spaces etc. to reply: n o lindan at net com dot com
psst.. want to buy an f-stop timer? nolindan.com/da/fstop/
Anonymous
January 11, 2005 1:18:35 AM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

Nicholas O. Lindan wrote:
> "Roger N. Clark (change username to rnclark)" <username@qwest.net> wrote
>
>
>>I have put together a web page showing the improvement
>>one can get with Adaptive Richardson-Lucy image restoration,
>>and comparison to Unsharp Mask methods. Please take a look
>>at: http://clarkvision.com/imagedetail/image-restoration1
>
>
> TTTT, I like the original: all the sharpened ones look overly
> pixelated and make the fox look like it needs to use a creme
> rinse after it shampoos.
>

Interesting. Well, some people like grain, some do not.
Please try printing Figure 7 at 300 ppi and see if you still like the
original. Sometimes I think that full resolution images on a
monitor show too much detail compared to a nice print.
I am interested in what you think then (for everyone reading).

Roger
Anonymous
January 11, 2005 9:42:59 PM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

"Roger N. Clark (change username to rnclark)" <username@qwest.net>
wrote in message news:41E35B3B.3050004@qwest.net...
SNIP
> May I add these images to my page? I'll credit you
> unless you request otherwise.

Sure, even if you don't credit me, go ahead.

Bart
Anonymous
January 17, 2005 2:25:25 AM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

"Bart van der Wolf" <bvdwolf@no.spam> wrote in message
news:41e1dba1$0$6213$e4fe514c@news.xs4all.nl...
>
> "David Zou" <xzous@yahoo.co.uk> wrote in message
> news:crskib$1r4$1@gemini.csx.cam.ac.uk...
> > Hi
> >
> > Sorry pls forgot that....
> >
> > Since the out of focus will blur the image via some defined function
> > and it is possible to restore it somehow (true of not?)...I guess it
> > accordingly. But I realise there is no way for the software to know
> > the corresponding distance of a point on the image....
>
> It would require at least two images focussed at different distances,
> or a user supplied depth map.
>
> Bart
>

Or a stereo pair?


Peter
Anonymous
January 17, 2005 5:53:25 AM

Archived from groups: rec.photo.digital,rec.photo.technique.nature (More info?)

"Bandicoot" <"insert_handle_here"@techemail.com> wrote in message
news:1105920505.46896.0@despina.uk.clara.net...
> "Bart van der Wolf" <bvdwolf@no.spam> wrote in message
> news:41e1dba1$0$6213$e4fe514c@news.xs4all.nl...
SNIP
>> It would require at least two images focussed at different
>> distances, or a user supplied depth map.
SNIP
> Or a stereo pair?

Indeed, either refocused with the same Field of View (allowing to
calculate, or being given, a depth map) or by using a stereo pair.

Bart
!