Real-time no-reference histogram-based freezing artifact detection algorithm for UHD videos

摘要

During video transmission different errors can occur, which can introduce distinct artifacts in video received at the end user side. One of the most common artifacts that appears in such cases is frame freezing and it can significantly reduce the quality of received video, i.e. decrease the level of end user Quality of Experience (QoE). Thus it is necessary to properly detect the occurrence of frame freezing, in order to assure the satisfactory level of end user QoE. In this paper a novel no-reference (NR) objective algorithm for detection of various types of freezing artifacts in video, called Histogram-Based Freezing artifacts Detection Algorithm (HBFDA), is proposed. HBFDA uses method for comparison of consecutive video frames, which consists of splitting frame into regions and comparing regions' histograms of consecutive frames. In order to operate on different types of video contents (including those containing a low level of movement), while achieving high level of accuracy and reliability, HBFDA dynamically adapts its parameters in real-time. HBFDA performance are compared to this of two freely pulbicly available algorithms for frame freezing detection on videos from three different databases: VQEG HDTV, LIVE Mobile and newly created UHD Video Freezing Database (UHD VFD), which we made publicly available to the research community and it can be downloaded at http://www.rt-rk.com/other/UHDVideoDBReadme.html. Experimental results show that HBFDA has lowest number of false freezing detections when compared to other tested algorithms and that it achieves the highest average total accuracy of 99.26% when all databases are taken into account. Additionally, for UHD videos, HBFDA is able to process 75 frames per second, which makes it suitable for real-time video applications.