As Internet bandwidth consumption increases, networks struggle to keep up with demand. For example, mobile device users are experiencing more frequent congestion delays as video sharing enables ever more impulsive sharing across mobile networks. As described in the Allot MobileTrends Report H2, 2009, the report “validates video as a mainstream medium and demonstrates that it is the single most influential factor driving the need for increased mobile network capacity.”
To address this demand, Euclid focuses on decreasing the encoding size of video through application of their EuclidVision technology. EuclidVision’s higher level modeling techniques enable a smaller encoding of a video which in turn decreases bandwidth consumption. The smaller encoding also allows video to be viewed in bandwidth constrained environments.
As video is analyzed, macroblock, feature and object models are created. These models are associated across two or more videos and are combined into libraries of models, further reducing the encoding size of each video. Given that the models in these libraries can be reused in more than one video, providing more powerful models for each video and reducing the need to build as many models during the decoding process. The libraries can also enable additional bandwidth reduction for similar videos while distributing model-encoding costs across the set of videos. The leverage of these models is the starting point for the future of video compression.
Key points:
- More Modeling: EuclidVision can model video data as macroblocks, features, or objects.
- New Leverage: Model libraries can be used to consolidate common feature and object models, reducing file sizes. Delivering even smaller file sizes than single file compression alone.
- Computation/Bandwidth Trade-off: The existing and new benefits of using models are greater than the processing cost incurred to create them.
eFLEX is Euclid’s prototype implementation of EuclidVision and utilizes EuclidVision’s modeling capabilities to compress videos as shown in the following example:
Consider watching an episode of a half-hour TV show that has been encoded with the eFLEX codec. Initially, your local eFLEX model library is empty. You choose to play the video and data transfer begins. The content provider may choose to send you a pre-built model to place in your library, or they may forego this step and let your decoder initially build models while decoding the conventionally encoded video.
In the case where the content provider sends you pre-built models, perhaps for example, the ten most bandwidth-intensive objects occurring in the TV show, you incur an additional bandwidth cost to download and save the models into your library. As you watch the show, encoding and decoding of data is performed similar to a conventional streaming scenario with one exception: since ten of the objects within the video are contained in your library, the content provider does not have to send them to you each time they occur in the video. Instead, the decoder recalls the object models from your local model library when necessary, using these models to recreate the actual objects and inserts them into the decoded stream.
Depending on the TV show, these ten objects might occur thousands of times throughout the episode. Considering that they are not transmitted over the network each time they occur, the effective file size of the entire show can be reduced considerably, saving a significant amount of bandwidth. Consider further, a case where the pre-built models contain the top twenty most bandwidth-intensive objects, the effective file size of the video is now even smaller, and even more bandwidth is saved. The benefits derived from adding more models can reach a point of diminishing returns, as any modeling method would. Additionally, while these downloaded models are being utilized, the decoder builds additional models as it decodes the stream, saving further bandwidth.
Consider when the current episode is completed and you decide to watch the next one. Since you already have both the preloaded and generated object models in your library, you don’t incur an upfront bandwidth cost. You can enjoy the bandwidth savings starting with the opening credits. As the second episode streams, your decoder can build more models, further reducing the effective file size. The second episode can have a smaller effective file size than the first. The third episode can have a smaller effective file size than the second, and so on. The benefits can accrue across episodes.
How will you spend your bandwidth savings? Depending on the speed of your broadband connection, eFLEX might be able to save enough bandwidth to let you watch the high definition version.
The benefits delivered by EuclidVision technology and the eFLEX codec accrue to the entire video industry - not just content consumers - as in the previous example. Now that you can model video data, what else can you accomplish besides reducing file sizes and saving bandwidth?