Scoring image
Locked
 
|
Suppose I take a sequence of 50 b/w images of an eagle on a branch with a telephoto lens.
Now, sometimes the bird is in focus, sometimes not, as my hands on the camera jiggle in and out. Now I can use stackreg to align the images to say key image 1. But how can I get a score for each of the 50 images to how well they match key image 1? "Template matching" looks for little things, "Block matching" was too course. Sift is sort of 1D. I guess it is sort of like facial recognition questions, but black and white. Suggestions? Ethan Ethan Cohen, Ph.D. Div of Biomedical Physics, WO62 Rm 1204 Office of Science and Engineering Labs, FDA Center for Devices and Radiological Health White Oak Federal Res Ctr. 10903 New Hampshire Ave. Silver Spring, MD 20993 -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html |
|
|
Good day Ethan,
your description appears being insufficient. Please post at least two *typical* images in the original PNG- or TIF-format (JPG is not suited for scientific purposes), so we can see what the actual geometric deviations look like and if they are really purely geometric and of a simple kind. I fear that you may have to deal with projective geometry which is far from easy to handle... Regards Herbie :::::::::::::::::::::::::::::::::::::::::::: Ethan Cohen wrote > Suppose I take a sequence of 50 b/w images of an eagle on a branch with a > telephoto lens. > > Now, sometimes the bird is in focus, sometimes not, as my hands on the > camera jiggle in and out. > > Now I can use stackreg to align the images to say key image 1. > > But how can I get a score for each of the 50 images to how well they match > key image 1? > "Template matching" looks for little things, > "Block matching" was too course. Sift is sort of 1D. > I guess it is sort of like facial recognition questions, but black and > white. > > Suggestions? > > Ethan > > > > Ethan Cohen, Ph.D. > Div of Biomedical Physics, WO62 Rm 1204 > Office of Science and Engineering Labs, > FDA Center for Devices and Radiological Health > White Oak Federal Res Ctr. > 10903 New Hampshire Ave. > Silver Spring, MD 20993 > > > -- > ImageJ mailing list: http://imagej.nih.gov/ij/list.html -- Sent from: http://imagej.1557.x6.nabble.com/ -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html |
|
|
I think the bigger question is what are you looking for?
I find it is common for researchers to come to me with groups of images and say they want to quantify them. But I first need to hear the story of their research. Often the images don't address the questions they have and we need to go back to redesign the experiments to generate relevant images. Usually there need to be different conditions. Measuring KO cells without looking at the controls and rescues doesn't tell you anything. A score is kind of useless unless you know what the score measures and whether the measured features tell you something about the phenomena you are studying. For instance, is you want to know about camera focus, you could segment the bird as an object and then look for high vs low frequency within the segmentation. You can generate a score. This would be an assay for instrument or user performance. You don't need different sets of images; the differences will be included in the single dataset and probably the highest frequency image would be defined as best focused. But this tells you nothing about the bird. If you want to know about the bird's rotation, maybe area covered by the bird is enough; large area, in profile vs small area facing forwards or back. Maybe you could add to this measurements for features such as tail vs head and left vs right and include in the score as a sign (+/-) or cos(theta) generated by function of full profile image. But as pointed out in an earlier response to your post, maybe different images or far more complex approaches are required. Maybe you need to grab from the metadata and include time in the score? Perhaps these questions clarify your question? Michael Cammer, Sr Research Scientist, DART Microscopy Laboratory NYU Langone Health, 540 First Avenue, SK2 Microscopy Suite, New York, NY 10016 [hidden email]<mailto:[hidden email]> http://nyulmc.org/micros http://microscopynotes.com/ Voice direct only, no text or messages: 1-914-309-3270 and 1-646-501-0567 ________________________________ From: Herbie <[hidden email]> Sent: Saturday, March 21, 2020 4:39:42 AM To: [hidden email] Subject: Re: Scoring image [EXTERNAL] Good day Ethan, your description appears being insufficient. Please post at least two *typical* images in the original PNG- or TIF-format (JPG is not suited for scientific purposes), so we can see what the actual geometric deviations look like and if they are really purely geometric and of a simple kind. I fear that you may have to deal with projective geometry which is far from easy to handle... Regards Herbie :::::::::::::::::::::::::::::::::::::::::::: Ethan Cohen wrote > Suppose I take a sequence of 50 b/w images of an eagle on a branch with a > telephoto lens. > > Now, sometimes the bird is in focus, sometimes not, as my hands on the > camera jiggle in and out. > > Now I can use stackreg to align the images to say key image 1. > > But how can I get a score for each of the 50 images to how well they match > key image 1? > "Template matching" looks for little things, > "Block matching" was too course. Sift is sort of 1D. > I guess it is sort of like facial recognition questions, but black and > white. > > Suggestions? > > Ethan > > > > Ethan Cohen, Ph.D. > Div of Biomedical Physics, WO62 Rm 1204 > Office of Science and Engineering Labs, > FDA Center for Devices and Radiological Health > White Oak Federal Res Ctr. > 10903 New Hampshire Ave. > Silver Spring, MD 20993 > > > -- > ImageJ mailing list: https://urldefense.proofpoint.com/v2/url?u=http-3A__imagej.nih.gov_ij_list.html&d=DwICAg&c=j5oPpO0eBH1iio48DtsedeElZfc04rx3ExJHeIIZuCs&r=hUBj2D5n6oKThx2L01qn8IORZb5f-ruLVXPmQ1zQNnM&m=ARFgiWAuFn2R2VVmH4BWe3HJS7TYOcOebwmOegQgFvI&s=pw4letC1ZhQ06lN7j_KnX3r4gAr_nzd7UuX6n9cHqsc&e= -- Sent from: https://urldefense.proofpoint.com/v2/url?u=http-3A__imagej.1557.x6.nabble.com_&d=DwICAg&c=j5oPpO0eBH1iio48DtsedeElZfc04rx3ExJHeIIZuCs&r=hUBj2D5n6oKThx2L01qn8IORZb5f-ruLVXPmQ1zQNnM&m=ARFgiWAuFn2R2VVmH4BWe3HJS7TYOcOebwmOegQgFvI&s=xFy2GE6MsnO-LPTba9YYQ5Y78SkD22OYIn9SbeaFQa4&e= -- ImageJ mailing list: https://urldefense.proofpoint.com/v2/url?u=http-3A__imagej.nih.gov_ij_list.html&d=DwICAg&c=j5oPpO0eBH1iio48DtsedeElZfc04rx3ExJHeIIZuCs&r=hUBj2D5n6oKThx2L01qn8IORZb5f-ruLVXPmQ1zQNnM&m=ARFgiWAuFn2R2VVmH4BWe3HJS7TYOcOebwmOegQgFvI&s=pw4letC1ZhQ06lN7j_KnX3r4gAr_nzd7UuX6n9cHqsc&e= -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html |
|
|
Dear Michael,
widely agreed... For me however, some of the OP's questions appear slightly more transparent. 1) I don't think that the OP uses "focus" in the sense you and I commonly understand it. 2) A score for the matching of a bird (silhouette?) in several images is desired with respect to a reference bird-image. 3) The images are taken without a tripod or the like, i.e. if possible they must be registered to stabilize the bird most likely without compensation of the bird's appearance. It is far from clear whether image registration (how many degrees of freedom?) is possible and to which degree. As I wrote, without seeing typical images, there is little chance to judge whether the task is possible or if it makes sense at all. We know nothing about the environment of the bird, the distance from the camera, the image quality, the kind of appearance (movements) of the bird in the various images, if color is essential, etc. We shall see... Best Herbie ::::::::::::::::::::::::::::::::::::::::::::: Cammer, Michael-3 wrote > I think the bigger question is what are you looking for? > > > I find it is common for researchers to come to me with groups of images > and say they want to quantify them. But I first need to hear the story of > their research. Often the images don't address the questions they have and > we need to go back to redesign the experiments to generate relevant > images. Usually there need to be different conditions. Measuring KO > cells without looking at the controls and rescues doesn't tell you > anything. > > A score is kind of useless unless you know what the score measures and > whether the measured features tell you something about the phenomena you > are studying. > > For instance, is you want to know about camera focus, you could segment > the bird as an object and then look for high vs low frequency within the > segmentation. You can generate a score. This would be an assay for > instrument or user performance. You don't need different sets of images; > the differences will be included in the single dataset and probably the > highest frequency image would be defined as best focused. But this tells > you nothing about the bird. > > If you want to know about the bird's rotation, maybe area covered by the > bird is enough; large area, in profile vs small area facing forwards or > back. Maybe you could add to this measurements for features such as tail > vs head and left vs right and include in the score as a sign (+/-) or > cos(theta) generated by function of full profile image. But as pointed > out in an earlier response to your post, maybe different images or far > more complex approaches are required. > > Maybe you need to grab from the metadata and include time in the score? > > Perhaps these questions clarify your question? > > > > Michael Cammer, Sr Research Scientist, DART Microscopy Laboratory > > NYU Langone Health, 540 First Avenue, SK2 Microscopy Suite, New York, NY > 10016 > Michael.Cammer@.nyu > <mailto: > Michael.Cammer@.nyu > > http://nyulmc.org/micros http://microscopynotes.com/ > > Voice direct only, no text or messages: 1-914-309-3270 and 1-646-501-0567 > > > > ________________________________ > From: Herbie < > l16@ > > > Sent: Saturday, March 21, 2020 4:39:42 AM > To: > IMAGEJ@.NIH > Subject: Re: Scoring image > > [EXTERNAL] > > Good day Ethan, > > your description appears being insufficient. Please post at least two > *typical* images in the original PNG- or TIF-format (JPG is not suited for > scientific purposes), so we can see what the actual geometric deviations > look like and if they are really purely geometric and of a simple kind. I > fear that you may have to deal with projective geometry which is far from > easy to handle... > > Regards > > Herbie > > :::::::::::::::::::::::::::::::::::::::::::: > > Ethan Cohen wrote >> Suppose I take a sequence of 50 b/w images of an eagle on a branch with a >> telephoto lens. >> >> Now, sometimes the bird is in focus, sometimes not, as my hands on the >> camera jiggle in and out. >> >> Now I can use stackreg to align the images to say key image 1. >> >> But how can I get a score for each of the 50 images to how well they >> match >> key image 1? >> "Template matching" looks for little things, >> "Block matching" was too course. Sift is sort of 1D. >> I guess it is sort of like facial recognition questions, but black and >> white. >> >> Suggestions? >> >> Ethan >> >> >> >> Ethan Cohen, Ph.D. >> Div of Biomedical Physics, WO62 Rm 1204 >> Office of Science and Engineering Labs, >> FDA Center for Devices and Radiological Health >> White Oak Federal Res Ctr. >> 10903 New Hampshire Ave. >> Silver Spring, MD 20993 >> >> >> -- >> ImageJ mailing list: >> https://urldefense.proofpoint.com/v2/url?u=http-3A__imagej.nih.gov_ij_list.html&d=DwICAg&c=j5oPpO0eBH1iio48DtsedeElZfc04rx3ExJHeIIZuCs&r=hUBj2D5n6oKThx2L01qn8IORZb5f-ruLVXPmQ1zQNnM&m=ARFgiWAuFn2R2VVmH4BWe3HJS7TYOcOebwmOegQgFvI&s=pw4letC1ZhQ06lN7j_KnX3r4gAr_nzd7UuX6n9cHqsc&e= > > > > > > -- > Sent from: > https://urldefense.proofpoint.com/v2/url?u=http-3A__imagej.1557.x6.nabble.com_&d=DwICAg&c=j5oPpO0eBH1iio48DtsedeElZfc04rx3ExJHeIIZuCs&r=hUBj2D5n6oKThx2L01qn8IORZb5f-ruLVXPmQ1zQNnM&m=ARFgiWAuFn2R2VVmH4BWe3HJS7TYOcOebwmOegQgFvI&s=xFy2GE6MsnO-LPTba9YYQ5Y78SkD22OYIn9SbeaFQa4&e= > > -- > ImageJ mailing list: > https://urldefense.proofpoint.com/v2/url?u=http-3A__imagej.nih.gov_ij_list.html&d=DwICAg&c=j5oPpO0eBH1iio48DtsedeElZfc04rx3ExJHeIIZuCs&r=hUBj2D5n6oKThx2L01qn8IORZb5f-ruLVXPmQ1zQNnM&m=ARFgiWAuFn2R2VVmH4BWe3HJS7TYOcOebwmOegQgFvI&s=pw4letC1ZhQ06lN7j_KnX3r4gAr_nzd7UuX6n9cHqsc&e= > > -- > ImageJ mailing list: http://imagej.nih.gov/ij/list.html -- Sent from: http://imagej.1557.x6.nabble.com/ -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html |
«
Return to ImageJ
|
1 view|%1 views
| Free forum by Nabble | Edit this page |