08.01.2013 Views

DigitalVideoAndHDTVAlgorithmsAndInterfaces.pdf

DigitalVideoAndHDTVAlgorithmsAndInterfaces.pdf

DigitalVideoAndHDTVAlgorithmsAndInterfaces.pdf

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

A raised cosine distribution is roughly<br />

similar to a Gaussian. See page 542.<br />

Schreiber and Troxel suggest reconstruction<br />

with a sharpened Gaussian<br />

having σ = 0.3. See their paper cited<br />

in the marginal note on page 192.<br />

structure is likely to be visible. If an external optical<br />

element such as a lens attenuates high spatial frequencies,<br />

then a box distribution might be suitable. A simple<br />

and practical choice for either capture or reconstruction<br />

is a Gaussian having a judiciously chosen halfpower<br />

width. A Gaussian is a compromise that can<br />

achieve reasonably high resolution while minimizing<br />

aliasing and minimizing the visibility of the pixel (or<br />

scan-line) structure.<br />

Spatial (2-D) oversampling<br />

In image capture, as in reconstruction for image display,<br />

ideal theoretical performance would be obtained by<br />

using a PSF with a sinc distribution. However, a sinc<br />

function cannot be used directly in a transducer of light,<br />

because light power cannot be negative: Negative<br />

weights cannot be implemented. As in display reconstruction,<br />

a simple and practical choice for a direct presampling<br />

or reconstruction filter is a Gaussian having<br />

a judiciously chosen half-power width.<br />

I have been describing direct sensors, where samples<br />

are taken directly from sensor elements, and direct<br />

displays, where samples directly energize display<br />

elements. In Oversampling, on page 174, I described<br />

a technique whereby a large number of directly<br />

acquired samples can be filtered to a lower sampling<br />

rate. That section discussed downsampling in one<br />

dimension, with the main goal of reducing the<br />

complexity of analog presampling or reconstruction<br />

filters. The oversampling technique can also be applied<br />

in two dimensions: A sensor can directly acquire a fairly<br />

large number of samples using a crude optical presampling<br />

filter, then use a sophisticated digital spatial filter<br />

to downsample.<br />

The advantage of interlace – reducing scan-line visibility<br />

for a given bandwidth, spatial resolution, and<br />

flicker rate – is built upon the assumption that the<br />

sensor (camera), data transmission, and display all use<br />

identical scanning. If oversampling is feasible, the situation<br />

changes. Consider a receiver that accepts progressive<br />

image data (as in the top left of Figure 6.8, on<br />

page 59), but instead of displaying this data directly, it<br />

CHAPTER 18 IMAGE DIGITIZATION AND RECONSTRUCTION 193

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!