51
edits
SIMD/Uses/SAD: Difference between revisions
Jump to navigation
Jump to search
→Uses of SAD
Maikmerten (talk | contribs) (Uses of SAD) |
Maikmerten (talk | contribs) |
||
| Line 17: | Line 17: | ||
A major use case for SAD is [http://en.wikipedia.org/wiki/Motion_estimation motion estimation], which is an integral part of most video compression schemes. Instead of transmitting complete frames, video codecs usually try to copy similar frame data from previously transmitted frames. The positions from which to copy parts of a previous frame are coded as [http://en.wikipedia.org/wiki/Motion_vector motion vectors]. This works well, as in video sequences most objects can be found in several consecutive frames, albeit they may have moved. | A major use case for SAD is [http://en.wikipedia.org/wiki/Motion_estimation motion estimation], which is an integral part of most video compression schemes. Instead of transmitting complete frames, video codecs usually try to copy similar frame data from previously transmitted frames. The positions from which to copy parts of a previous frame are coded as [http://en.wikipedia.org/wiki/Motion_vector motion vectors]. This works well, as in video sequences most objects can be found in several consecutive frames, albeit they may have moved. | ||
SAD is often used to determine these motion vectors. | SAD is often used to determine these motion vectors. Usually square regions of the current and a preceding frame are compared. Each region can be understood as a vector of pixel values. Generally speaking: the lower the SAD "score", the better the regions match. The computed motion vector is a vector pointing to the region in a preceding frame with the best match. | ||
== Code example == | |||