Inter-Picture Prediction for Video Compression using Low Pass and High Pass Filters

Most of the IP network traffic today consists of video content. In 2015 the videotraffic was estimated to be 70% of the Internet traffic and it is expected to rise past80% by 2020. The demand for high quality video, including high definition and ultrahigh definition, is steadily increasing with high resolution TV and computer monitorsbeing available to the general consumer. The video content streamed over theInternet have typically been compressed using a video encoder to reduce the bitsrequired to store and send the video. The increase of high resolution video contentalso requires updates to the video compression techniques. AVC is one of the mostcommonly used video compression standards, and its successor, HEVC, was able toachieve a 50% reduction of bitrate for the same perceived quality. Today there isongoing work to improve the compression efficiency of HEVC even further. Interpicture prediction, available in AVC and HEVC, is used to find redundant pixelsbetween adjacent frames in a video sequence and describe the movement of thepixels using motion vectors. This thesis focuses on exploring possible improvements to the inter prediction of thesuccessor to HEVC, currently under development, using low and high pass filteringwithin motion estimation and motion compensation to find potentially betterpredictions. The evaluation of the implemented filtering extension shows that themotion estimation and compensation filtering result can yield small benefits in somevideo sequences, with most of the video sequences in the test set resulting in smalllosses. There are still improvements to be made to the implementation, so there arepotentially more benefits to be gained by performing filtering within motionestimation and compensation