Observant and alert, excited and concentrated, video game streamers can use a variety of skills to attack the opponents and meanwhile defend themselves in game play. Live games provide a perfect opportunity for numerous gamers to watch master players fight against each other, and also bring a large number of fans to professional players, thus further making video games more attractive.
Video games have gradually transformed from a subculture to a mainstream culture. For lots of gamers, watching major events like the League of Legends tournament at Asian Games, 2018 Fall Arena of Valor tournament, 2018 League of Legends World Championship, etc. has become a new lifestyle. At the same time, however, gamers' pursuit of high-definition picture quality and smooth playing experience has become the biggest pain point for carriers currently. Aiming at this problem, Kingsoft Cloud has recently introduced a product, Kingsoft Smart High Definition (KSHD), which can offer ultra-high-definition picture quality and is suitable for a diversity of video scenes, and will become an efficient assistant for carriers.
KSHD Strikes the Biggest Pain Point of the Video Industry
Nowadays, as the resolution of mobile terminals gets higher and higher, the viewing experience is also getting better and better. And a majority of live broadcast users have raised their requirements for picture quality and playing smoothness of various live streaming including live games. But smoothness and high-definition picture quality mean an increase in the bandwidth cost for live broadcast platforms, which becomes the biggest pain point of the video industry.
To deal with this pain point, Kingsoft Cloud has put forward a solution. Based on self-developed algorithms, KSHD integrates AI, encoding, image processing and other techniques. With deep neural network, KSHD can perceive the video content, optimize subjective experience and conduct intelligent encoding adjustment, thus enhancing the picture quality and visual effects of videos and adapting videos for screens with higher-definition. Suitable for various video scenes, it can not only bring a better visual experience, but can also help customers save 20% - 40% of bandwidth in video transmission.
In terms of the classification of video scenes, KSHD has formed over ten classes of and dozens of groups of video scene model libraries through deep learning. When using the KSHD service, customers are able to analyze live streams in real time, then match them with the corresponding video scene models, and process the picture quality in real time according to their requirements on the video picture quality.
Creating Different Solutions Flexibly for Different Scenes
To begin with, take Panda Kill, a variety show produced by Panda TV, as an example. It is an indoor program, based on which KSHD can match the corresponding scene model. After that, KSHD will select the most suitable coding parameters according to the result of scene classification recognition, combined with code rates, frame rates, resolution, etc., and at the same time, sharpen, denoise the video source and compensate the levels of it. It is worth mentioning that, about the lights and reflection points appear in the program, KSHD can eliminate those lights and reflections invisible to human eyes through AI modeling, and save the texture, helping live platforms save bandwidth more.
Comparison of "Panda Kill" picture quality (left: original, bandwidth: 5M; right: after being processed by KSHD, bandwidth: 3M)
Helping Customers Save Bandwidth While Maintaining the Same Picture Quality
Take the live streaming of League of Legends as an example. Since the scenes in live games are special with a lot of real-time dynamic elements, the picture quality and users’ watching experience will be influenced if the code rates are not adapted. With intelligent code rate control and dynamic optimization of encoding, KSHD can apply AI picture quality to every detail in the videos, tremendously enhancing the picture quality of color areas including the foreground color, background color and UI background color. KSHD's ROI processing technology can protect the visual perception of contents that human eyes feel the most sensitive to and pay the most attention to, and optimize other contents that consume code rates, thereby saving bandwidth. It can save at least 20% of bandwidth while improving the picture quality.
Images of live games
Images after being processed by KSHD
The Three Core Technologies to Improve Picture Quality
From the above comparison, we can see that KSHD can not only help video platform customers save bandwidth while maintaining the same picture quality, but also can enhance the picture quality and make users' subjective feelings better. This is because KSHD owns three core technologies: video scene recognition, visual focus segmentation and video picture quality enhancement.
In terms of video scene recognition, KSHD will set up separate video models for various types of content such as games, news, shows, animations, sports, short videos, etc. to conduct classified real-time recognition based on different video scenes, and configure the encoding parameters appropriate for a particular scene according to the recognition result of different scenes. And, it will select the optimal encoding template parameters according to the texture, the amount of motion variation, etc.
About visual focus segmentation, the ROI processing technology mentioned above is a part of it. When processing videos, KSHD will segment them into different areas and process each area separately. It will protect the visual core areas of human eyes, and sharpen and brighten the macroblock encoding in these areas. On the contrary, it will reduce the code rates of unimportant areas. By segmenting the visual area and processing different areas separately, it can improve the visual experience of users.
Visual focus segmentation
Video picture quality enhancement is to achieve higher-definition picture quality and lower code rates with smaller video files. KSHD can adjust the usage of code rates according to different scenarios. To illustrate, when there are most dynamic elements, a higher code rate is used, otherwise a lower code rate is used. And for the still part of a video image, a lower code rate is used; for the dynamic part, a higher code rate is used. More importantly, KSHD mainly enhances the picture quality of the areas focused by human eyes, and relatively weakens the unfocused areas, which can reduce the code rates and, at the same time, ensure the clarity and smoothness of videos, saving costs for customers.