Kramer SP-14, Руководство по эксплуатации

Страница: 18 / 33

18
Cadence preservation can be switched on and off. When set to on filmic stutter of 3:2 or 2:2 or 23.98 or
24Hz material is preserved by applying a 2:2 (50Hz o/p) or 3:2 (50.94 and 60Hz o/p) cadence to the
converted material.
A zone can be defined to exclude an area from the MEMC processing. When changing the values for
the borders a white rectangular will mark the area being set up. The border can be switched on or off
with the menu item Display Border. When Enable MEMC Mask is set to Invert rather than On, the outer
of the area is masked out from the MEMC processing.
With the menu item MEMC Demo Mode the display can be tiled and one area is being MEMC
processed, the other not. This allows a side by side comparison of the processing algorithms.
4.5.2. Advanced Settings
These parameters allow user optimisation of the Motion Estimation, Motion Compensation process,
according to the nature of the video being converted.
By default, these settings are at “Auto” which means the values used are selected internally according to
the MEMC/FRC level setting (Low, Medium and High), and sometimes changed dynamically according
to various statistics that are measured frame by frame.
The settings and threshold values can be individually overridden from the automatic value to stages
denoted as Minimum, Very Low, Low, Medium, High, Very High, Maximum. Changing these
parameters may give better conversion results for certain isolated scenes depending on content.
The SP-14 algorithm for Motion Estimation is based on a technique called Phase Plane Correlation.
This uses FFT analysis of corresponding blocks in two successive video frames, to quickly extract
accurate details of motion quantity and direction. The PPC is applied repeatedly over a grid of
overlapping blocks to generate a field of motion vectors. Typically, the majority will be very similar as
they represent background motion, which is the apparent motion due to movement of the camera, the
remainder being the independent motion of any (foreground) objects.
However, with real video scenes, there can be lots of different causes of confusion or ambiguity, so the
Motion Vectors need to be individually checked and verified to avoid artefacts appearing in the output
video. Problematic scenes have repeating structures, such as fences and railings, windows in a tower
block, trees on a mountainside, ocean waves. Occlusions can be problematic, such as objects being
hidden or revealed by the motion of other objects, or perhaps a door opening or closing.
The following terms and abbreviations are used in the menu descriptions:
PPC Phase Plane Correlation
MV Motion Vector, the direction and quantity of movement of a block within an image
SAD Sum of Absolute Differences, which is an algorithm for measuring the similarity between
two blocks of video image
BG BackGround: that portion of a scene that only moves because the camera is moving
FG ForeGround: objects or people in a scene that are moving independently of the camera
Outlier A part of the image that is moving independently of the dominant motion
Cost Cost factors are relative weightings used when assessing and applying different classes
of candidate MVs
The following menu items can be changed:
Robust SAD threshold
Coring level for pixel noise filtering. Adjust this to improve robustness of MV detection especially when
images are complex.
SAD Cost Factor
Default cost weighting
Temporal/Spacial Balance
Cost balance between matching motion vectors to the previous set of MVs or matching to neighbours in
the current set.