Spring Kind Webinar Preview: Increasing the Value of Existing Content with Machine Learning

Spring Kind Webinar Preview: Increasing the Value of Existing Content with Machine Learning

By Tony Jones, Principal Technologist, MediaKind June 3, 2020 | 3 min read
4K TV, Consumer Experience, GAN, Generative Adversial Neural Networks

Ultra-High Definition (UHD) content is now part of mainstream TV production. 60% of TV sales in major markets are now capable of UHD quality, according to research from IHS Markit, and the availability of UHD TV sets in the home is expected to soar rapidly over the next few years. By 2023, IHS Markit predicts there will be 574 million 4K TV households globally. The burgeoning popularity of UHD has raised expectations for stunning, immersive high-quality TV and video experiences.

The challenge for broadcasters and service providers is to satisfy this consumer demand while finding new ways of monetizing the wealth of high value, archived content. Yet only a limited amount of this library content is available to consumers in a UHD native format. Akin to the early transition from standard to high definition, broadcasters must now fill full-time UHD channels with content that is better than HD, and consistent with consumer expectations. But where can they find this valuable UHD content?

Filling full-time channels with UHD content

One potential solution lies in up-conversion, which was introduced during the introduction of HD channels in the mid-2000s. However, traditional up-conversion techniques can result in end-user experiences that are more ‘HD-like’ than UHD. These techniques can potentially be performed at either the TV/STB or at the broadcast headend prior to transmission.

Although it is theoretically possible to transmit a ‘native-only’ format through switching dynamically within the channel as the content changes, it does not provide a satisfactory solution. Dynamic behavior is not well defined, and there is a high probability of visual and audible transitions being perceived. While modern TVs are often equipped with up-conversion hardware, each TV is likely to process the content differently. This results in an inconsistent viewing experience.

Machine Learning Based UHD Up-Conversion

MediaKind has responded to this challenge by developing innovative research into the use of Machine Learning to perform advanced up-conversion of high value library content, using neural networks in creating up-converted images that are more akin to native UHD images. This enables the delivery of UHD channel viewing experiences that are visibly superior to HD services.

The use of Generative Adversarial Neural Networks (GANs) to synthesize detail in the upconverted image allows the viewer to enjoy a more compelling, higher quality experience. This method offers the possibility and potential to interpolate in a more non-linear manner, which achieves a significantly enhanced result.

With this new innovation in up-conversion technique, MediaKind has unlocked the strategic and monetization potential of machine learning architectures, providing insights into how the training process works and an opportunity to examine up-converted image clips in detail.

To learn more about this pioneering new method, you can download MediaKind’s recently published application paper here.

Register now to learn more in my webinar on June 10

On Wednesday, 10th June (9am CST/3pm BST/4pm CET), I will be presenting the findings of MediaKind’s new application paper on this subject, exploring how the use of up-conversion and leveraging GANs can synthesize detail in the converted image, resulting in an end user experience that is much closer to native UHD.

During the session, I will explain how broadcasters, operators and TV service providers can fill the gap in availability of UHD content through the non-linear capabilities provided by convolutional neural networks (CNNs) using a hybrid GAN architecture. Using parallel clips of up-converted UHD HDR content (from HD), this demonstration will provide an in-depth, side by side comparison through the use of CNNs using hybrid GAN architecture, against that of conventional, bi-cubic up-sampling. Register for this session here.

I look forward to joining you on the 10th June! If you’re unable to attend this session, you can watch on-demand from June 11. Other presentations from MediaKind’s Spring Kind webinar series are available to watch here.

Click here to sign up for the webinar: ‘Increasing the Value of Existing Content: Machine Learning for HD to UHD Up-Conversion’  Register Here

Related Posts