users
users > RE: Registering to a blurred reference brain
Jul 18, 2014 04:07 PM | Torsten Rohlfing
RE: Registering to a blurred reference brain
Hi Ashley:
To some degree, registration with a blurry image will always lead to compromises in registration accuracy. A blurred average image is that way, after all, because the input images were not accurately aligned before averaging, so each point in the average represents a mixture of true anatomical locations. Thus, you wouldn't expect a single-specimen image to align "perfectly", simply because there is no perfect alignment with an ambiguous image.
Looking at your data, it seems to me that also blurring the other image would not lead to improvements, but likely make things worse.
Alternatively, you might want to "sharpen" the blurred image (ideally it should have been created from better-aligned specimens in the first place, but ah well). With the caveat that I haven't really thought this through, not tried it before, one possible approach to sharpening might be to subtract a further-blurred copy of the average image from that average image (I think this is similar to Photoshop's "Unsharp Masking" tool).
Like I said - not thought through, but maybe worth giving it a shot. CMTK's "convertx" tool has a Gaussian filter and "imagemath" can subtract two images, so that may be all you need to play with the idea.
Best,
Torsten
To some degree, registration with a blurry image will always lead to compromises in registration accuracy. A blurred average image is that way, after all, because the input images were not accurately aligned before averaging, so each point in the average represents a mixture of true anatomical locations. Thus, you wouldn't expect a single-specimen image to align "perfectly", simply because there is no perfect alignment with an ambiguous image.
Looking at your data, it seems to me that also blurring the other image would not lead to improvements, but likely make things worse.
Alternatively, you might want to "sharpen" the blurred image (ideally it should have been created from better-aligned specimens in the first place, but ah well). With the caveat that I haven't really thought this through, not tried it before, one possible approach to sharpening might be to subtract a further-blurred copy of the average image from that average image (I think this is similar to Photoshop's "Unsharp Masking" tool).
Like I said - not thought through, but maybe worth giving it a shot. CMTK's "convertx" tool has a Gaussian filter and "imagemath" can subtract two images, so that may be all you need to play with the idea.
Best,
Torsten
Threaded View
Title | Author | Date |
---|---|---|
Ashley Manton | Jul 17, 2014 | |
Torsten Rohlfing | Jul 18, 2014 | |