Information Technology Reference
In-Depth Information
In order to reduce the level of complexity, the 3DRS algorithm is employed for
all types of frames. The motion vectors are initialized with zero for those frames,
whose reference frame motion vectors are not evaluated yet or whose reference
frame is I frame for temporal prediction. For the remaining frames in the GOP, the
3DRS algorithm is invoked where initially the motion vector array of the reference
frame is stored in memory. This array is then updated from left to right in a raster
scan order. The best vector found is further refined by conducting the one pixel
local diamond search around the position corresponding to the best motion vector.
The best vectors found are up to one-pixel accuracy. The sub-pixel refinement and
mode decision process is exactly the same as in the original Dirac algorithm.
7.3 Region of Interest Coding (Foveated Image Processing)
Region of interest coding allows you to code the certain portion of video with
more resolution and the rest of the video portion with less resolution and hence
help in efficient conservation of bit budget. This feature is already implemented
in JPEG2000. Region of interest (ROI) coding is important in applications where
certain parts of an image are of a higher importance than the rest of the image. In
these cases the ROI is decoded with higher quality and/or spatial resolution than
the background (BG). Example includes: Client/server applications where the
server initially transmits a low quality/resolution version of an image. The client
then selects an area of the image as a ROI and the server transmits only the data
needed to refine of that ROI. Similarly, when browsing a digital photograph album
it is often the case that we are looking for, or most interest in, the people/faces in
those photographs. Using an automated face detection algorithm the region(s) of
an image that contain faces can be coded as ROI's and therefore stored with more
accuracy than non-face sub-images.
There are two approaches to Foveated Image Processing.
1)
Object Based Foveation
2)
Motion Based Foveation
7.3.1 Object Based Foveation
In this approach, we do Foveation, based on the object. Particular object is
detected and then that region is sent with higher resolution and the remaining
object is sent with lower resolution. So the main task is to detect the object. For
example, in a video of a Football match, football is to be detected and that region
where football is present would be sent with higher resolution and the other region
would be sent with lower resolution.
7.3.2 Motion Based Foveation
In motion based Foveation, the region of interest is estimated based on its motion
vectors. If motion vectors have a high value, means that there is more motion at
Search WWH ::
Custom Search