Talk:Reconstruction filter
 | This article is rated C-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects: | ||||||||||
| |||||||||||
Is this in reference to CT reconstruction from DT samples?
Not anti aliasing
editIts not to avoid aliasing, its to remove atrifacts of the digital signal--Light current 11:52, 22 December 2005 (UTC)
- I agree. In addition I think an important point that is not generally appreciated is that images from cameras and video, as commonly viewed, never leave the digital domain - we view the samples, not the signal they represent, which results in loss of sharpness as well as 'twittering' on moving detail and Moire fringing, which is not strictly aliasing but is often taken to be. Attempts to test cameras for aliasing by looking for these effects without upconverting are invalid on patterns below the Nyquist frequency for this reason. In photography this is easily overcome when printing images (essentially analog domain) by upconverting, for example in Photoshop to a higher resolution before printing. With video we need to view on a higher resolution screen, as no optical reconstruction method has yet been devised for screens. -Lindosland (talk) 09:27, 29 April 2011 (UTC)
Ideal Reconstruction Filter
editI think this text is misleading for the case of a DAC output with zero-order hold (i.e. stair stepping). In fact, a zero order hold has a frequency response equal to a sinc function with its first zero at the update rate of the DAC. At the Nyquist rate (update rate of the DAC / 2), the frequency response of the system as described is 2/pi or ~0.636. In order to correct, the ideal frequency response of the reconstruction filter should be 1/sinc with a brick wall at Nyquist. So, in fact, peaking is needed with a zero-order hold system in order to assure a flat passband performace.
- I agree, this is an important point. We should also note that when we view photographs or video we usually view the samples directly as square pixels, not points of light, which imposes this 0.63 loss again (though confused by 'twittering' and Moire fringing) unless some sharpening is applied in an attempt to compensate. -Lindosland (talk) 09:27, 29 April 2011 (UTC)