50mm Prime with Resolution for a D800? Come on Nikon/Sigma!

Started Jan 4, 2014 | Questions thread
Scott McMorrow Regular Member • Posts: 366
Re: 50mm Prime with Resolution for a D800? Come on Nikon/Sigma!

Mister Hairy

You misunderstood me. When I say contrast, I meant contrast of the final image. If we start with 100% unit step cycle-to-cycle contrast of the original image (i.e. black lines separated by white space) the final resulting image is the convolution of the 2-dimensional PSF of the lens, and the sensor array. By sensor array I include the sensor, Bayer filter, and AA-filter (if there is one.) For a measurement of resolution in lp/mm, we can plot the transfer curve of final image contrast vs. resolution.

As we know, the Fourier components of the original step image repeated in cycles for a resolution measurement contains harmonics well above the cycle frequency (resolution). We also know that all physical waveguides (which includes the 2-dimensional waveguide known as a lens) have some finite signal amplitude until we are beyond the propagation mode of the waveguide. In the case of a lens, there is finite signal (image) amplitude throughout the visible spectrum. As such, no matter how poor the lens is, there will be some image energy beyond Nyquist for any lens/sensor combination. We also know that AA-filters are imperfect and have some leakage just beyond Nyquist, and being a periodic filter, have additional lobes at repeating intervals above Nyquist.

Finally, the process of aliasing (moire) is an extremely sensitive one to our eyes for two reasons. First, it produces a beat frequency repeating pattern against the original high frequency pattern, which has reinforced peaks and troughs. Second, it produces color artifacts due to the de-mosaicing algorithm that also stand out to our visual system. Because of this, small amplitude leakage of image energy above Nyquist is easily seen visually, and is dependent upon the amplitude of the signal seen at the sensor. The amplitude of the signal seen at the sensor for a repeating black/white cyclical pattern is the measured contrast between black and white. At the peaks and troughs of the aliased moire pattern, contrast is doubled, and overall peak-to-trough contrast is quadrupled. In effect, moire is a strong amplifier of the pattern that gets through, which is why even a bad lens, with sharp resolution cutoff, can still show moire. Contrast that has dropped to lets say 5% between black and white, which is bit over 4 EV difference, is barely visible, except by pixel peeping. However, through aliasing it is boosted by 20% to a 2 eV difference, which is quite visible.

You are correct that AA-filtered sensors are a poor choice for making lens resolution measurements. Non-AA filtered sensors will provide much more information to work with. However, within the bandpass of the AA-filter, we can recover much of the original signal by sharpening, or by applying an optimal mathematical filter as Imatest can be asked to do. In doing so, we can plot contrast vs. resolution in lp/mm with fairly good accuracy. (Accuracy is, of course, better with a non-AA filtered sensor. At this point in time, the D7100 represents the highest resolution Nikon F-mount sensor, when measured in normalized lp/mm, and makes a good candidate for lens measurement, at least in the DX portion of the frame.) When these measurements are made across multiple lenses on a D800, it is quite clear that some lenses are superior to others, in that the final image contrast vs. lp/mm resolution is higher than others. Some lenses have a sharp cutoff due to aberrations, others do not. This becomes especially clear as we move from the center out to the corner of the frame.

Another aside.  Your contention is that any image information that gets through indicates the "resolution" of the lens.  This would be akin to saying that the bandwidth of an analog RC low pass filter is infinite. In analog we use attenuation criteria such as 3dB bandwidth or 6dB bandwidth to specify the bandwidth of an amplifier or filter.  In the case of optics, we can use lp/mm at MTF50, or MTF30 to specifiy the resolution of a lens or system.  Resolution at MTF10 contrast levels is not generally very useful, but it does still contribute strongly to alias artifacts.

Post (hide subjects) Posted by
(unknown member)
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow