Fast Motion Estimation for a High Efficiency Video Coding (HEVC) Encoder
G.P. Maier · Utrecht University Repository (Utrecht University) · 2013
This thesis discusses various concepts aiming at decreasing computational complexity regarding motion estimation with particular reference to a HEVC compliant encoder. The study was conducted at the Fraunhofer Heinrich Hertz Institute in the course of the Game and media technology master's programme at Utrecht University. For this purpose, algorithms regarding both integer-pixel search and sub-pixel search as implemented in a reference software encoder as well as strategies proposed in the literature are reviewed and their possible application is discussed. Drawing from existing ideas, new combinations of strategies that have proven to work well are developed and implemented in a corresponding encoder. Regarding sub-pixel search, a new context-based approach is introduced. Results are verified by means of experimentation, using a test set of 36 video sequences as well as two significantly different encoding profiles. In this course, several options to speed-up the motion search process at a reasonable loss of encoding efficiency are identified. Results regarding integer-pixel precision motion search can be understood as a framework, allowing for possible adaptation suiting a particular field of application. With respect to fractional search, it is shown that approaches may result in a significant amount of gain in coding efficiency as provided by sub-pixel motion estimation while making almost the entire computation process obsolete.