Open source video codec Ogg Theora hot on the heels of H.264

MPEG-4 AVC (H.264), the efficient, open industry standard for video encoding, has made huge strides to become the industry leader in all areas – it plays on mobiles and MP3 players, it's used by HDTV and Blu-ray Discs, and cameras and HD camcorders record in it. H.264, currently the most efficient video compression algorithm, is also, since Adobe integrated the codec into its Flash Player, used for web videos and now Microsoft is also adding H.264 to Silverlight 3. Nevertheless, there is one important snag with H.264 – from 2011, license fees will be required from sites streaming video using this technology.

For this reason, the World Wide Web Consortium (W3C) and members of the Web Hypertext Application Technology Working Group (WhatWG) proposed the open source Ogg codecs Theora (video) and Vorbis (audio), developed by the Xiph Foundation, as the standard for the planned video and audio elements in the "W3C Editor's Draft" for HTML 5. However, in the end concerns expressed by, among others, Nokia over "patent risks" for such "proprietary codecs" resulted in this passage being deleted from the draft. Nonetheless, the forthcoming Firefox 3.5 will support the Ogg codecs, as do experimental versions of Opera 9.6.

The Ogg Theora development team are hoping to achieve substantial efficiency gains from "Thusnelda", a newly developed, backwards-compatible encoder, which is to be deployed in the forthcoming Theora 1.1. The latest enhancements – such as the use of a forward discrete cosine transformation (FDCT) – are showing significant progress. Thusnelda is now, for the first time, in some cases achieving better PSNR (Peak Signal to Noise Ratio) figures than the free H.264 encoder x264. It's worth noting that PSNR is an objective, purely mathematical value which does not necessarily correspond to visual perception; nonetheless, codec developers use the value as a start point for looking at the encoding efficiency of their creations.

A posting on the FFMpeg mailing list noted that the figures for x264 were incorrectly interpreted, leading to a 4db worse assessment of PSNR values. This adjustment would mean that x264 was still better, on average by 2db, than Thusnelda. Despite this, Thusnelda delivers far better results than the current Theora Encode 1.0, which should be sufficient for many internet applications. The quality improvement should primarily benefit Wikipedia and the Wikimedia Commons, who only use royalty-free Ogg codecs for audio/video content. However there is still much work to be done on optimising the quantisation matrices and adaptive quantisation before Theora 1.1, with Thusnelda, sees the light of day.

(djwm)