HEVC is the up-to-date video compression standard, especially targeted for HD/UHD video contents. HEVC can achieve much higher performance than its previous generation standard AVC, yet at the cost of much higher computational complexity. In order to make sure a hardware based and HEVC compliant encoder capable of running at real time while in an economical fashion, certain encoding tools are needed to be leveraged to fit with the hardware pipeline philosophy. In the same cost scale, the hardware based on our encoding algorithm can achieve higher Rate-Distortion performance than competitors in different GOP configurations.

Video retargeting is a video processing technology that adjusts the original video content into various devices with different display resolution. It is widely used in universal media playback, video transcoding and any area that would exploit the adaptibility. The challenge lies in how to still perserve the visual consistency of the retargeted video while making it track the most distinguishing content of the original effectively. In this project, we designed a video retargeting system addresses the temporal consistency issue holistically and is still capable of preserving high degree of visual interestingness. The experiment results indicate that our proposed video retargeting system outperformed the conventional methods.

Every day we encounter so much video data, but most of the time we are only interested in certain parts of it, not all. Unfortunately, we never actually know where those parts are in the video until we finish watching it. How exhausting! What if you could create a video summary, like a news digest? In this project, we proposed a video summarization method that allows viewers to understand the summary and recover the original plot correctly. Experiments show that our proposed method outperforms the traditional scalable approach by 10% in terms of summary informativeness and enjoyability.

While Dynamic Adaptive Streaming over HTTP (DASH) is a fast-growing trend to deliver broadband videos, we extend its adaptivity with a new dimension --- viewing conditions. Due to human visual acuity limitations, viewers can only perceive certain frequency spectrums depending on the viewing distance and angle. By encoding the video content into multiple presentations with various frequency ranges, the streaming client is able to choose the most efficient version accordingly, enhancing R-D performance significantly.

While streamed video often suffers from packet loss during the transmission. How exactly do these lost packets introduce artifact and degrade the percepual quality of the decoded video? This research project seeks to a computational model in the decoded image domain to measure the glittering artifact introduced by inperfect error concealment.