Search Results - PentaxForums.com

The Nyquist-Shannon sampling theorem states the frequency required (2x highest signal frequency) required in order to allow an exact reconstruction of the signal. I haven't seen a form of it which accounts for an admissable loss in contrast of the signal.

But you are right, in my thought experiment the number of observed lines is less than the actual number of lines which must not happen.

Assuming that actual lines are spaced apart with a distance that is equal to their width then one cannot resolve more LW/PH than the sensor has vertical pixels (and even that requires B&W RAW processing). One could reduce contrast by choosing smaller lines but it would not be admissible to reduce the spacing between the lines accordingly (only by a minute factor, < * 1 / #vertical pixels).

Right, otherwise one may not recognise spurious resolution.

Forum: Pentax Lens Articles 02-10-2009, 07:22 PM

Sticky: DA lenses on Full Frame: Test Shots thread
Posted By Class A

Replies: 523
Views: 446,410

How did you calculate that?

The K20D's sensor has 3104 visible vertical pixels.

I'm assuming that 3000 LW/PH means that 3000 line widths are resolved across the vertical length of the picture with a 50% loss of relative contrast.

I'm further assuming that 1.3 x the number of line widths will cause a contrast loss of 50%. If I see 1.3 N line width through a N line width grid, then each grid line blurs 1 + 1/3 line widths so the contrast between adjacent grid line widths should be 1/3 : 2/3 = 0.5 (50%).

That would give me 4035 LW/PH but only for a black and white image and assuming that RAW data is evaluated (i.e., omitting any demosaicing).

Due to the bayer array structure of the sensor, the colour resolution will be reduced. The minimum should be 2017 LW/PH but the real LW/PH figure should be somewhere between 2017-4035 values as interpolation can recover information. For instance, for say image positions occupied by red pixels by using information from adjacent green pixels. How much colour resolution can be resolved will depend on the demosaicing algorithm.

All this omits the anti-aliasing filter which will reduce the LW/PH figure.

Search took 0.00 seconds | Showing results 1 to 7 of 7

Forum Jump