4K resolution has been becoming commercial rapidly. While 4K video has always been preferred for viewing on large TV displays, the standard that video resolution on their smartphone is being raised as more and more people want 4K video. However, for video providers, the rapid adoption 4K has been a challenge.

High-resolution video is necessary to satisfy audiences, but it often leads to a surge in video data, which in turn increases network usage and storage costs. So, we’re going to introduce a video AI that improves video compression efficiency to solve this problem. The AI model is said to make it possible to streaming 4K video without breaking the bank, How does it work?

AI to optimize video quality.

Optimization means seeking maximum efficiency within a given scope, when it applied to video content that meaning maximizing the quality of your video while minimizing resources or costs, such as time and network. Video optimization AI we’re introducing in this content is an AI model that can be applied to the video compression stage to make video production more efficient. In the industry where video is in high demand can be a noteworthy this AI model to deliver a lot of high-resolution video content more efficient.

Can you see the difference between the two images above? The image is a snapshot of a part of a video original version(left) and the optimization AI-applied version(right). I think with the naked eyes, you can barely tell the difference between the left and right images. To confirm this more objectively, we measured the quality using VMAF(a perceptual video quality assessment algorithm developed by Netflix) and found that the original video snapshot(left) scored 85.5 and the video snapshot with optimization AI(right) scored 87.28, which is about 2 points higher.

A technology to optimize video quality using AI model in the stage of compression.

Video optimization AI which mentioned is a video perceptual quality optimization AI model developed by BLUEDOT called ‘DeepField-PQO(Perceptual Quality Optimization)’. It is an AI model that optimizes the bitrate of the video to reduce the capacity while maintaining perceptual quality, which is the quality of a video as seen by human eyes. As shown in the video snapshots above, there is almost no difference between the images before and after applying AI, and the difference in data between the two videos is 21.5%, of course, the snapshot with PQO is 21.5% lower.

In technical aspect, BLUEDOT’s DeepField-PQO(hereinafter PQO) is an AI-based video pre-processing technology that converts a video for more efficient compression before delivers to the encoder. Therefore, when the video is processed with PQO and then encoded, the compression efficiency is increased, and perceptual quality of the video remains same, but data is lower than the original. Before AI applied, predetermined parameters were applied to pixels in a video frame at once, resulting in encoding without considering the quality and bitrate relationship between before and after compression, so the compression results were often unsatisfactory. However, BLUEDOT has developed a preprocessing algorithm that applies AI to change the pixels in a frame in consideration of the metric and perceptual similarity between the input image and the inferred image(the expected result of image processing), and the bitrate during compression, resulting in more natural compression results.

Features of AI that pre-process video in encoding.

BLUEDOT’s PQO is an AI model to process video before encoding, making it easier to compress while maintaining perceived quality. The great thing about a pre-processing solution like PQO is that it can be applied to any codec. As a video streamer, probably using the codec that works best for your production environment, and if you’re using a specialized codec, likely to run into the limits of what video technology you can apply. BLUEDOT’s PQO is processed a video before encoding, it can be applied to any video that can be encoded.

For example, AV1 codec is a highly compression-efficient codec that can reduce 50% data of h.264 at the same quality level, which is used by many video streaming companies, such as NETFLIX. Although AV1 is highly compression-efficient, it also uses a lot of CPU resources, resulting in slow encoding speeds and high network costs. Due to this cost burden, many video streamers want to use lower resources, but worried about decrease of video quality. However, with PQO, even if the compression efficiency is lower due to resources, you can increase the compression efficiency because it preprocesses the video, and you can improve the compression speed by using fewer resources. By using PQO to hardware encoders with same PPA(Power / Performance / Area) level, you can encode with high compression efficiency even if you don’t implement many features.

With so many videos being uploaded every day and so many viewers consuming them, video optimization AI will be the video technology that can bring satisfaction to both the users who supply and consume video, especially in the era of high-definition video, such as 4K. If you’re interested in learning more about DeepField-PQO, an AI-based video pre-processing technology that can optimize video encoding to achieve maximum efficiency, you can find more information on the site below.