This technique should apply to both normal video (consequtive sequences of pictures of real world) and animations (sequences of images drawn by human or generated by computer)
This technique is supposed to perform better in lossy compression and by default it should use that mode but it does not preclude the option of lossless mode.
Overview
High level system diagram V1
[Motion Pictures] -> [Preprocessing] -> [CV module (cross frames)] -> ( CV restore : Elements/contours/directions/(inferred)object texture/descriptions )
| ______________|_____________
| v |
|--------------------------------> [ Image error analysis module] |
| |
v v
( difference ) -----> [ Sub image/Entropy coding]
Motion images compression and restoration based on computer vision
原文：http://www.cnblogs.com/quanben/p/4049911.html