NeuroMotive™ User’s Manual
LB-0336 Rev. 1.00
Page 35
High -
Used by the Blu-ray Disc storage format. High definition image, but very
costly for computing resources and an even higher risk of data/frame loss.
o
I
-
Frame Interval -
The I-Frame is a key frame that does not depend on other frames in
this encoding for information. Having a shorter I-Frame Interval means that the video file
is more robust against corruption, is faster to seek through (fast forward, rewind, etc.) and
should generally have a higher quality picture. A smaller interval, however, also means a
larger file size.
o
B-Frame Rate -
B-Frames are a type of frame that can look in both directions of a file
during compression for portions of frames that do not change and so only need to be
noted once in the video file, reducing file size or conversely allowing more detail to be
stored in the same amount of space. The frame rate indicates how many B-Frames will be
inserted between I-Frames and P-Frames (like B-Frames, but can only look in one
direction of a file during compression). The higher the rate, the better the compression-to-
detail ratio of a file, but there is also more time needed during file seeking and less
defense against corruption. A frame rate of 2 is considered sufficient.
o
Reference Frames -
This values determines how many frames a B or P-Frame may
search through looking for frame redundancies during compression. The higher the
number of frames, the better the compression and lower file size/better quality. But in
exchange decompression (playback and seeking) and compression times are increased
and more memory will be needed upon playback for decompression.
o
Slices -
During encoding (compression) a frame can be split into regions (slices) that can
be searched for redundancies against the same regions in other frames. On a multi-core
processor regions can be assigned to different cores and searched in parallel, greatly
increasing compression speed. The tradeoff is that some redundancies may be missed
because they merely shift to a different region (like in a panning shot), this loss results in
a lower quality video.
o
ME Method -
The Motion Estimation Method contributes to reducing the size of the
video and improving video quality in transmission by removing redundant information
between frames during the encoding process. It is the most computationally intensive
portion of encoding and thus has many algorithms that attempt to reduce this cost while
maintaining quality. NeuroMotive™ currently supports 6 different algorithms with the
H.264 encoding format that vary in quality vs. computational time required to perform.
All include a threshold optimization and are as follows:
(1) 2D Logarithmic Search -
This approach navigates a search area by
evaluating five equally spaced points (blocks) in a cross shape that shrinks when
the center candidate most closely matches the reference block (it is the best
candidate) and shifts when it does not. This one is a good balance of speed and
quality for encoding.
(2) Enhanced 2D Logarithmic Search -
This is an enhanced version of the
above Logarithmic Search that expands the number of blocks evaluated at a time
to nine, but retains the same speed. A better quality version of the Classic
Logarithmic Search; this is the default method for encoding.
(3) EPZS Search -
The Enhanced Predictive Zonal Search (EPZS) algorithm is
similar to the logarithmic searches, but it attempts to achieve the speed and
precision of those searches while also trying to ensure that the entire search area
is analyzed for the best candidate. By starting with a larger search area (1/4 of
the entire search region), but using the same number of points, it trades off some
precision for a better chance at finding the best global candidate.
(4) One-At-A-Time Search -
This algorithm first checks along the center x-axis
of the search area for the best candidate point and then along the y-axis from
that point. This is very fast and uncomplicated, but not necessarily accurate in
finding the globally best block of pixels in a search area for motion estimation.