Dolores, The true PSF of an imaging device is the point source broadening in the continuous, or analog, image before it is spatially sampled by the detector array. (This PSF should include some additional broadening due to the finite area of the detectors in the array, as opposed to point sampling.)
Gaussian is a good approximation to PSF in many cases, but in others, it may not be. For the airborne imaging we did, we found that fitting the sampled edge spread function (ESF) to an error function yielded RER values (see below) that were erroneously low, for the same reason as above, namely that the PSF was not actually Gaussian but had tails that fall off more slowly than a Gaussian. If one can apply physical reasoning to construct a more realistic PSF or ESF function (with parameters adjustable for the fit), one might get a more accurate portrayal of the underlying PSF of the imaging system. Various papers have described results with various forms, e.g. pairs of 'Fermi' distributions for ESF. I've been wanting to try a Voigt (atomic lineshape) function for PSF. I'm actively working this problem in Matlab. More information on request. When measuring ESF and RER, we take all the pixels in a rectangular ROI spanning a clean edge. If the edge is tilted slightly from horizontal or vertical, the pixels and the ROI will sample the ESF in finer intervals that the array sampling interval (pixel spacing), which is necessary when the frequency spectrum (MTF) of the true PSF exceeds the Nyquist frequency of the sampling device. Good luck. Jeff -----Original Message----- From: Michael Schmid <[email protected]> Sent: Thursday, June 26, 2025 11:25 PM To: [email protected] Subject: Re: Problems with MTF Calculation Hi Dolores, sorry, I know nothing about the COQ plugin. Assuming you have a Gaussian PSF, there is a very simple solution that requires no plugin: You can take a line profile over an edge and fit an error function with Data>Add Fit... The log window then tells you the parameters. Divide parameter 'd' by sqrt(2) to get the sigma of the Gaussian. It would be rather easy to write a macro that does it with a single keystroke. Michael -----Original Message----- From: [email protected] <mailto:[email protected]> <[email protected] <mailto:[email protected]> > Sent: Thursday, June 26, 2025 2:44 PM To: '[email protected]' <[email protected] <mailto:[email protected]> > Subject: RE: Problems with MTF Calculation Sra. Morillas, For airborne image sensors we use a metric called relative edge response (RER) which is derived from the edge spread response (ESF). Satellite image sensors also use this concept. We use RER both to focus a sensor in flight and to evaluate its performance. I'm interested in finding out more about COQ and trying it out. Are you analyzing horizontal and vertical edges to focus a sensor, measure performance or both? Thanks. Jeff Stokes Solid State Physicist San Diego, CA USA ________________________________________________________________ On 26.06.25 12:11, M Dolores Morillas Perez wrote: > Hello. > > I am a new user of image J > > I have problem with MTF calculation with plugin COQ. > > When I have analized an horizontal edge with plugin COQ, the values of > MTF (Y AXIS) is NaN > > Could you help me? > > Kind Regards, > > Mª Dolores Morillas > > Medical Phisicist > > H.U.V.V > > Málaga > > -- > ImageJ mailing list: http://imagej.nih.gov/ij/list.html -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html
