Inter-frame Coding Techniques Optimization For Video Coding Standard HEVC

High definition video contents are becoming big challenges for transferring andstorage. The latest video coding standard HEVC introduces more and more codingtechniques to achieve higher ultimate compression efficiency. But at the same timethe encoder complexity is greatly increased. When available resources areinadequate to apply all the coding techniques, compression efficiency will drop. It isthe very problem we try to solve in this paper that how to enable video encoder to fitconstrained and varying computational recourses.To solve the problem, we proposed a video coding framework based oncost-performance priority and designed some suitable algorithms. The frameworkmakes the video encoding optimally scalable. Optimally scalable means the encodercan fit to any varying resources and achieve relatively best performance. There arethree major contributions in this paper:1. We have proposed a video coding framework consisting of four hierarchicallevels which include algorithm level, block level, frame level and sequence level.This framework can simplify the cost-performance predicting and ranking problem,which makes the framework compatible with existing fast algorithms.2. Based on the framework, in algorithm level, we have proposed an optimallyscalable fractional-pixel motion estimation algorithm for HEVC. We predicted thecost-performance ratios of each fractional-pixel search point by analysis andstatistical learning. As a byproduct, this algorithm can be used as fast algorithmwhich can reduce the fractional-pixel motion estimation time by53%with negligibleloss on compression efficiency. 3. Based on the framework, in frame level, we have built an optimally scalablemulti-mode selection algorithm for HEVC. We developed a new video featuretermed “Motion Collision Count (MCC)” to predict the cost-performance ratios foreach coding unit (CU). We then apply multi-mode selection to those CUs withhighest cost-performance ratios. As a byproduct, this algorithm can be used as fastalgorithm which can reduce the total encoding time by43%with negligible loss oncompression efficiency.The first and second contributions have been published on IFTC2012andICASSP2013. The third contribution have been submitted to ICIP2013.