So Many H.264 Codecs, So Little Time

HD Tests
I ran the HD tests with a new test file that contained a mix of high- and low-motion footage that I created for Streaming Media East last May. I shot all source footage in 1080i HDV, most with Canon’s XH A1 camcorder. I captured and edited the clip in Premiere Pro, then output a scaled, deinterlaced 720p file from After Effects using the AlgoSuite plug-in. Again, all encoding tools simply encoded, with no preprocessing whatsoever.

I produced all HD test files at 1280x720 with a frame rate of 30 fps and a data rate of 800Kbps for video and 128Kbps for audio. These rates are insanely aggressive, but as you’ll see, it definitely helped separate the contenders from a quality perspective. I produced using 2-pass variable bitrate (VBR) encoding, with target/maximum settings of 800/1600, adjusting the target as necessary to produce a file within 5% of the target. I encoded using the High profile when available and maxed out all quality-related encoding parameters.

As mentioned previously, with a target of about 11MB total file size, Compressor would go no lower than 18MB, so I produced a file in Squeeze using the Apple codec that met the target. Unfortunately (and inexplicably), Semaphore couldn’t load the file, so I couldn’t analyze the data rate further. However, after analyzing the Episode file, I boosted the target in Episode to 800/4800 to produce a file that matched Carbon Coder’s 1.34Mbps maximum data rate.

In still-image trials, the numbers don’t really do Dicas justice; while clearly a touch behind MainConcept in all trials, the codec was miles ahead of Apple, which struggled on all but the lowest motion clips. In high-motion clips, the difference was very significant.

I wondered if the problem with the Apple codec related to Compressor, as silly as that sounds given that it’s an Apple product. So I produced two equivalent files in Squeeze—one using the MainConcept codec, the other using the Apple codec. Even on this locked-down tripod, talking-head clip, the MainConcept codec clearly outperformed the Apple codec.

In motion trials, the sheer abundance of still-frame artifacts in the Apple clip ensured a noisy video during real-time playback, pushing Apple to last. The Dicas clips exhibited slightly more background noise than MainConcept, though again, the difference was relatively modest. None of the codecs dropped frames to meet the target data rate, so all shared top scores in this category.

Overall, MainConcept edged Dicas in this category, where most viewers would have a hard time telling the two codecs apart without side-by-side comparisons. On the other hand, if you decide to use the Apple codec, you should count on requiring a substantially higher bitrate than either Dicas or MainConcept to produce equivalent quality.

Tutorial
I’ve been producing lots of screencams lately, and I wanted to see how the H.264 codecs performed encoding these types of files. The test file is a 1024x768, 15 fps screencam captured in Camtasia and edited and output in Premiere Pro. I rendered the file using VBR encoding with a target data rate of 200Kbps for video and 32Kbps for audio, using the highest-quality encoding parameters supported by the encoding tool.

The results mirrored the HD tests, where Apple was out of the running very early, and MainConcept enjoyed a small, probably commercially irrelevant quality advantage in both still and motion tests. Apple clearly isn’t optimized for screencam encoding, and I got nearly identical results when producing from Squeeze using the Apple codec.

Again, much of the difference between MainConcept and Dicas derives from the improved contrast that Carbon Coder or the MainConcept codec itself appears to apply, resulting in clearer, more vivid colors. In motion trials, the MainConcept codec seemed to recover from major screen changes two or three frames faster than the Dicas codec, though most users wouldn’t notice without side-by-side comparisons.

Overall, if you’re in the market for an H.264 encoding tool, I would definitely prioritize any tool with the MainConcept codec, though the advantage over the Dicas codec is very modest. On the other hand, if you’re currently producing H.264 video with the Apple codec, you could probably improve quality, cut bitrate, or both by switching to a tool with either the MainConcept or Dicas codec.

The State of Video Codecs 2015

H.264 still accounts for most video encoding today, but HEVC/H.265 and VP9 are beginning to make noise. What will 2015 bring?

19 Mar 2015