We have been trying to find an answer to this question amongst ourselves. We know of a number of papers who have used Gaussian/low-pass filters to 'remove high spatial frequencies' but largely the methodology behind applying the filter seems to be rather slapdash. In an ideal world I think we should be using Fourier transform on the frequency spectrum of our images. I have attempted to do this using ImageJ.
Exactly, Dan. Sorry, I should have been more specific, Mark. Ideally, I/we would like to be able to demonstrate through sound mathematics and theory that we have removed the correct spatial frequencies that are critical in word/object recognition (approx. 2-4cpd). Others that have used this similar methods in our field state in their methods that they used a filter but don't justify the use of or explicitly state why the cut-off was chosen. Any advice you can give would be appreciated, Mark.
Thanks, Josh
Hi Josh,
Normally, sound editing software offers the possibility to fix the frequency band that you want to remove. It is a bit unclear what you mean by "spatial frequencies" as Mark pointed out. You can remove a band of frequencies, but if you want to know the spectral structure of the band of frequencies that you have removed you have to analyze the sound into frequency partials and then see what is left after you have removed a band of frequencies. If your sounds are periodic, the spacing between the partials should normally be more or less the same as they are multiples of the fundamental. Listening tests could be of help before you think of bringing in mathematics. Mind you, I'm not really sure of what your project is about.
Apologies. I don't seem to have been very clear with what we are looking for and I think that is a product of not really knowing the exact terminology to use. Essentially, we want to remove high spatial frequencies above 6cpd from a series of digital images and words. We want to explore whether we can simulate the reading behaviours and object recognition patterns of patients who present with Letter-by-letter reading following left hemisphere posterior stroke - patients show a drop-off on visual sine wave tests around 6cpd (Roberts et al., 2013). Now, in 2006, two studies were published that professed to do the same thing (for words only) however, they simply used photoshop to apply a Gaussian low pass filter without justification of what frequency band they had removed. I have done some reading around the subject with regard to images and we need to use a FFT to create the frequency spectrum and apply a low-pass Gaussian filter with a cutoff at 6 cpd. However, scaling from the pixels of the image and converting this to 6cpd does not appear to be straightforward. Any information you (or anybody else) could provide on this latter part would be much appreciated.
Dear Josh,
Thank you for the further clarification that you have made regarding your project. I am replying because I would feel uncomfortable (and perhaps you too) to leave in silence. I am not clear whether you are working with sounds or with images. Besides, I am not clear how you can convert digital images to audio frequencies.as to remove part of it. Furthermore, you are still referring to "spatial" frequencies; I do not know how this relates to sound or images. I would suggest that you do some more reading in the literature dealing with your field as to more precisely formulate your hypothesis and methodology. I'm sorry I've not been able to help.
Akpan.
Hi Josh, did you get an answer to this question ? I am looking for a similar answer on the exact spatial frequency that are allowed by High Pass filters of Photoshop. Thanks !
Dear Josh S. Payne , The Fourier transform of Gaussian is an another Gaussian. The "sigma_t" parameter in the time domain is the inverse of "sigma_f" in the frequency domain (sigma_f=1/sigma_t), the more resolution in time you get, the less resolution in the frequency domain. You may adjust the Gaussian standard - deviation to adjust low-pass filter bandwidth in the frequency domain. Thanx to Nabin Paudel , Akpan Jimmy Essien , Daniel Roberts , Mark Haefner for useful answers.
- Badges
- Science topic