| Re: Image error calculation [message #70283] |
Thu, 01 April 2010 00:42  |
Maxwell Peck
Messages: 61
Registered: February 2010
|
Member |
|
|
I have a feeling there is probably a 'proper' way to do this but
perhaps you could use something like Sobel edge detection filter
(http://idlastro.gsfc.nasa.gov/idl_html_help/SOBEL.html) which
calculates the magnitude of the gradients in the image. Sharper lines
I would think have a larger gradients than blurred areas so perhaps a
difference between the Sobel detected 'improved' image and the Sobel
detected original image may give some indication? I'm not sure if
you'd have to average the result somehow because of the blurring
itself though...
Max
On Apr 1, 2:07 pm, Suguru Amakubo <sfa2...@googlemail.com> wrote:
> Sorry about that, what I will consider to be a better quality image is
> that the details of the structures (DNA) could be identified better.
>
> So in a nutshell if I see the new image and see more details that was
> previously unidentifiable (due to partial bluring) that I consider to
> be a better image. A 'sharper' image will probably best describe it.
> However the problem lies in quantifying it. (Since saying this image
> looks better just won't do. It needs to be: e.g. x % better than the
> original image).
>
> As for the how I made the image, I basically used one image as a
> 'base' and the took 22 different images of the same DNA that was taken
> immediately after each other and then split the new image into smaller
> subset images and mathematically found a point that is considered to
> be similar and placed it on top of it (then divided to get the end
> image).
>
> My aim therefore is to compare the base image with the new image and
> determine quantitatively by what degree the image has improved.
>
> Sorry about the lack of explanation. Please tell me if the above needs
> explaining further.
>
> Suguru
|
|
|
|
|
|
|
|
| Re: Image error calculation [message #70366 is a reply to message #70283] |
Thu, 01 April 2010 09:22  |
Brian Daniel
Messages: 80
Registered: July 2009
|
Member |
|
|
Image quality is a current field of research. Check out SPIE or OSA
journals if you're really interested in more detailed metrics. If
you're not, I suggest taking the root mean squared error between your
images. IDL's element by element algebra is very handy for this.
Best,
Brian
On Apr 1, 3:42 am, Maxwell Peck <maxjp...@gmail.com> wrote:
> I have a feeling there is probably a 'proper' way to do this but
> perhaps you could use something like Sobel edge detection filter
> (http://idlastro.gsfc.nasa.gov/idl_html_help/SOBEL.html) which
> calculates the magnitude of the gradients in the image. Sharper lines
> I would think have a larger gradients than blurred areas so perhaps a
> difference between the Sobel detected 'improved' image and the Sobel
> detected original image may give some indication? I'm not sure if
> you'd have to average the result somehow because of the blurring
> itself though...
>
> Max
>
> On Apr 1, 2:07 pm, Suguru Amakubo <sfa2...@googlemail.com> wrote:
>
>
>
>> Sorry about that, what I will consider to be a better quality image is
>> that the details of the structures (DNA) could be identified better.
>
>> So in a nutshell if I see the new image and see more details that was
>> previously unidentifiable (due to partial bluring) that I consider to
>> be a better image. A 'sharper' image will probably best describe it.
>> However the problem lies in quantifying it. (Since saying this image
>> looks better just won't do. It needs to be: e.g. x % better than the
>> original image).
>
>> As for the how I made the image, I basically used one image as a
>> 'base' and the took 22 different images of the same DNA that was taken
>> immediately after each other and then split the new image into smaller
>> subset images and mathematically found a point that is considered to
>> be similar and placed it on top of it (then divided to get the end
>> image).
>
>> My aim therefore is to compare the base image with the new image and
>> determine quantitatively by what degree the image has improved.
>
>> Sorry about the lack of explanation. Please tell me if the above needs
>> explaining further.
>
>> Suguru
|
|
|
|
| Re: Image error calculation [message #70367 is a reply to message #70285] |
Thu, 01 April 2010 09:15 |
Craig Markwardt
Messages: 1869
Registered: November 1996
|
Senior Member |
|
|
On Mar 31, 11:07 pm, Suguru Amakubo <sfa2...@googlemail.com> wrote:
> Sorry about that, what I will consider to be a better quality image is
> that the details of the structures (DNA) could be identified better.
>
> So in a nutshell if I see the new image and see more details that was
> previously unidentifiable (due to partial bluring) that I consider to
> be a better image. A 'sharper' image will probably best describe it.
> However the problem lies in quantifying it. (Since saying this image
> looks better just won't do. It needs to be: e.g. x % better than the
> original image).
>
> As for the how I made the image, I basically used one image as a
> 'base' and the took 22 different images of the same DNA that was taken
> immediately after each other and then split the new image into smaller
> subset images and mathematically found a point that is considered to
> be similar and placed it on top of it (then divided to get the end
> image).
>
> My aim therefore is to compare the base image with the new image and
> determine quantitatively by what degree the image has improved.
>
> Sorry about the lack of explanation. Please tell me if the above needs
> explaining further.
If you have a signal-free region of your image, you could calculate
the image noise "before" and "after," and show that the noise was
reduced.
However, you mentioned the use of JPEG formatted images. Since the
data values of JPEG images are not calibrated, you will have a hard
time quantifying the amount of exact improvement. Or rather, I should
say that JPEGs are calibrated to human perceptual levels which are non-
linear (sRGB, Adobe RGB, etc), rather than photometric levels. Better
to use TIFF, or at the very least, use JPEG with some agreement with
the maker about what the data values mean from a photometric
standpoint.
Craig
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
