MPEG-2 Encoder Shootout
Byline: Barry Braverman
Look, fellow shooters, let’s talk straight.
Distribution on VHS is dead. DVD is and will be the name of the game for a while. And so we need an MPEG-2 encoder that we can truly hang our waveforms on. It’s really that simple.
Or is it? Here we spend our days and lives tweaking and re-tweaking every last light on the set and camera menu option, applying just the right amount of Black Stretch, Cine-look, and Pre-Knee control to achieve visual perfection. And then what happens after the gifted editors and compositors work their magic? It’s off to some nameless MPEG-2 encoder with a dumbed-down wizard application for compression to DVD-compliant streams.
It’s sickening really when you think about it, that we craftsmen-shooters have allowed ourselves to settle into this sorry state. Up to 90 percent of our drop-dead gorgeous images is about to be tossed, and who is around to make these critical compression decisions? If it isn’t you, it most likely is someone with a whole lot less interest in your images – and career.
Not all encoders are created equal
There are many MPEG-2 DVD encoders on the market, and each encoder (or transcoder, if compressing from a non-tape source) has its own personality, user interface, and respect for the DVD-Video standard. And while the DVD specification mandates certain technical requirements for DVD players, such as the 9.8Mbps maximum bit rate for the multiplexed stream, there is little commonality from one encoder to another in other respects – even among models produced by the same manufacturer.
The Sonic Solutions SD-1000 and SD-2000 hardware encoders, for example, normally produce a GOP (Group of Picture) size of 13 with an IBBP frame structure. The SD-500 encoder offered by the same company produces a default GOP size of 15, also with an IBBP sequence, although in the first two encoders the last P-frame is retained, but in the latter encoder, it is dropped. Oh, man. In the face of such complexity, no wonder many camera folks would sooner duck the DVD encoding issue entirely. Unfortunately, given the direction of the current and increasingly DVD-dominated world, we can no longer afford that luxury if we intend to remain gainfully employed far into the future.
To stay competitive and in control of one’s images, the shooter must understand in a substantial way the capabilities and limitations of encoders. Some encoders are more adept at compressing high-motion scenes than scenes with complex color. In other encoders, the reverse might be true, thus underlying the critical importance of understanding the range of encoding tools available and their relative capabilities.
Such an understanding directly applies to your role as a visual storyteller. Varying GOP sizes and structures (a GOP of 4 to 18 frames is legal in NTSC) may have a profound effect on how your images ultimately look on playback. A short GOP or a structure devoid of B-frames (bidirectional frames) may produce markedly plastic-looking images. Maybe this is what you want. But whether it is or isn’t, it is you who should be making these critical image decisions, not some nameless engineer or third party working in a post house at 3 a.m.
Every MPEG-2 encoder applies its own logic based on a proprietary set of rules or algorithms. Some encoders implement these algorithms in hardware and may cost upwards of $25,000 to $50,000. Other encoders, usually at much lower cost, may implement the compression algorithms entirely in software (often times bundled with self-contained authoring packages like Apple DVD Studio Pro or Adobe Encore DVD).
Shooters who leave critical MPEG-2 compression decisions to others do so at their peril. But with so many encoders and transcoders on the market, what is a shooter to do? Read on.
(For more technical specifics on how compression algorithms work, see Steve Mullen’s tutorial on page 47 of this issue).
During the first week in February, a select team of 10 DVD specialists and I holed up in darkened room at Video Symphony in Burbank to evaluate a range of professional and quasi-professional MPEG-2 encoders. The products included hardware- and software-based encoders from Sonic Solutions, Canopus, Innobits, Discreet, Cinema Craft, and Apple. We wanted to look at both Mac- and PC-based solutions, focusing on the most popular players on both platforms. Here is a list of the competitors included the shootout, the issues we attempted to resolve, price, and platform:
Apple Compressor 1.1: Back-ended into FCP 4 and DVD SP2, it’s certainly inexpensive, easy, and convenient. But would it hold up to the more feature-laden and pricier entries? Price: N/A (included as a standalone application with FCP 4 and DVD SP2). Mac.
Canopus ProCoder 1.5: “Best bang for the buck” is a comment heard often on the street, so naturally we wanted to test this for ourselves. An updated version 2.0 will likely be available by the time you read this. Price: $499. PC.
Canopus MPEG PRO: A hardware-based encoder that handles direct input of composite and S-Video signals. There isn’t a lot of competition among hardware encoders, especially in this price range, so naturally we wanted to see if this product was suitable for the professional DVDers. Price: $499. PC.
Cinema Craft Encoder SP 2.66: The pro version of this software is used at major studios and A-title DVD authoring houses, but we wondered if a comparable level of performance could be expected from the more modestly priced SP version. Price: $1999. PC.
Discreet Cleaner 6: A venerable old friend often thought of as the second most essential software program on a Mac (after Adobe Photoshop). But would version 6 continue to charm us under the scrutiny of our punishing tests? Price: $499. Mac.
Discreet Cleaner XL: Since the product’s release, the company has been rather quiet about its Cleaner product for PC, so hey, we wanted to find out what was up. Price: $499. PC.
Innobits BitVice 1.4: With all its latest improvements and features, the BitVice’s buzz in DVD circles has been considerable of late. We wanted to see how it handled the tough stuff. Price: $297. Mac.
Sonic Solutions Transcoder 2.1.7: It’s old, slow, and runs only under Mac OS 9. But if you’re authoring at spec-level in Sonic DVD Creator or Fusion, it may still be the most practical way to go. Price: N/A (software included with Sonic DVD Creator and Fusion systems). Mac.
Sonic Solutions SD-2000: To many folks, this is the gold standard of encoding hardware. But the encoder is almost four years old, and the low-cost all-software solutions are catching up – or are they? Price: $25,000. Mac, and soon Windows (according to Sonic).
TMPGEnc: The muscular low-cost encoder has many bells and whistles, but two judging teams had trouble creating technically legal files. Hmm. Thus, we could only partially evaluate this product while wondering what went wrong. Price: $48. PC.
Our evaluations were conducted in two parts, the first covering practical and operational concerns, ease of use, the intuitiveness of the user interface, and workflow. The judges also considered the availability of essential features like multi-pass VBR encoding, noise reduction, inverse telecine, and 24p support.
Particular attention was paid to the encoder’s feature set in the 3Mbps to 4Mbps range. Practically speaking, this is critical for any encoder because while virtually any modern MPEG-2 encoder performs adequately at 7Mbps, the same cannot be said at 3.5Mbps. Truth is, we must often encode our shows at or less than 4Mbps in order to accommodate a full two-hour program on a single-layer DVD. As we discovered in our tests, not all of the encoders would allow us to encode at bit rates below 4Mbps in variable bit rate mode, a major shortcoming for those of us who make our living at least partially in commercial DVD development
We scored the entries on a zero-to-five scale in the workflow portion and on a zero-to-10 scale in the performance category, the scoring thus reflecting how we felt about the relative importance of the two arenas. Judges were given a maximum of 45 minutes per encoder to evaluate the interface and features and complete their encoding tasks. The manufacturers’ operating manuals were available for reference at all times, although judges were encouraged not to consult them in order to properly gauge the intuitiveness of the encoder’s user interface.
In the course of their evaluations, the judges compressed three short files, the first being a one-minute clip of the opening title sequence from a popular sci-fi TV drama. (See figure 1, page 30.) This clip featured a plethora of brilliant light flashes, rapid-fire effects, and the textured faces of cast members superimposed over the visual chaos. This clip would be a major challenge for any MPEG-2 encoder.
Test clip #2 (figure 2, page 30) was a 30-second sequence from a National Geographic project that I shot on film in 1982. The telecined footage from Betacam exhibited significant grain noise in the film-to-tape transfer and thus would likely present a particularly tough test. This, combined with long fades and dissolves, would be an arduous task for MPEG-2 encoders at low bit rates.
Test clip #3 (figure 3, page 32) consisted of a simple but punishing test of white letters crawling across a black screen. The temporal shift from one field to the other inside the frame contributes mightily to Artifact City in some encoders. This trial can be tough on the MPEG-2 encoder, especially those that do not permit field-based encoding. Most do not.
Our performance tests were accomplished at three quality levels; the first at 3Mbps in the encoder’s “best-quality” mode. In most cases, this meant a “VBR two-pass” encode, where the first pass is used for analysis. But not all encoders tested support VBR at bit rates less than 4Mbps, so CBR (Constant Bit Rate) encoding in those cases was used instead. Additional encodes were performed at 4Mbps VBR two-pass and 7Mbps CBR from a DV-compressed QuickTime file. (See “CBR vs. VBR,” page 33, for a comparison of CBR and VBR encoding.)
The hardware-based encodes were performed via a Sony DSR-1500 VCR utilizing the same QuickTime files output to DVCAM tape. The Sonic SD-2000 reference encode was accomplished via SDI from the DSR-1500; the Canopus MPEG PRO encoder made use of the S-Video input. The total running time of the three test files was less than two minutes.
Cruel blind justice
To evaluate performance, we employed a blind test in order to grade contrast and resolution, artifacts, integrity of the text test, and general overall quality. The files, assigned a number, were authored in Sonic DVD Creator and subsequently multiplexed and imaged to DVD-R general media. The judges reviewed the disc as a group on a Sony model DVP-NS415 player. Table 1 shows the results from the judges in order of estimated overall quality.
Recommendations and analysis
Based on the blind test, it was obvious: the Cinema Craft CCE-SP encoder offers performance at least one or two orders of magnitude better than competing products. Encodes were exceptionally clean, sharp, and clear of obvious banding in the TV drama opening sequence. Judge Greg Mohr’s comment was typical when he noted that “the high quality of CCE-SP’s output made other encoders’ output look like a third-generation dupe.”
The encoder also offered a range of options that DVD professionals want and need: advanced motion detection, forced I-frames on scene changes, inverse telecine, noise reduction, and multiple-pass VBR encoding (beyond two passes!). The performance edge did not extend to the creeping text test, however, where judges noted combing artifacts at lower bit rates in the most difficult transition areas. The CCE-SP’s workflow and user interface were also judged to be a notch or two below where it ought to be, prompting one team of judges to balk at the markedly “engineers-oriented interface.” The routine import of QuickTime files also proved to be a challenge to judges who resented having to troubleshoot the QuickTime installation on their PC workstations.
The Tale of the Tape
(in order if estimated overall quality)
Still, the quality advantage of the encoder was clear in the side-by-side comparisons. The CCE-SP is not inexpensive at $1,999, but if you must have the best software encoder in the world today, this is it.
The Canopus ProCoder’s user interface may have put off the Mac users in the group, but for the more Windows-inured judges, worn down perhaps from years of experience in other programs like Sonic Scenarist, the ProCoder’s less than cordial UI was offset by the quality of its encodes. Overall contrast and resolution were well above average for the encoders tested, and the ProCoder’s low price ($499) undoubtedly contributes to its current popularity.
Still, the drilling down through multiple cryptic layers and pulldown menus to access the ProCoder’s advanced settings was a needlessly long and arduous journey for many judges. One judge complained after finally arriving at the Promised Land that many of the default values were wrong. “How often is ‘Upper Field Dominant’ the correct choice?” judge Ken Delvo wondered. As most users opt to encode at mastering quality, he pointed out, it was astonishing also to see a dialog box advising that this preferred setting requires “10 to 20 times” longer encoding times. For serious DVDers, the Canopus ProCoder contains another drawback – its inability to encode in master quality at rates less than 4Mbps VBR. Perhaps this will be rectified in a future version of ProCoder.
Canopus plans to introduce an updated ProCoder version at NAB 2004 with several crucial new features. While a reduced-rate VBR capability is not mentioned in the advance announcement, version 2.0 will sport a new Watch Folder that can be placed on a network of Macs and PC, thus greatly facilitating workflow in cross-platform shops. HD support has been added in the updated version as well, offering users the ability to encode 720p and 1080i to and from SD. Very nice.
And here’s the best part for those of us who’ve been frustrated by the lack of closed-caption support in MPEG-2 encoders. Up until now, most encoders and transcoders did not preserve the closed-caption data from the source video, which has led often to a frantic search by clients for the original .cc data files. These files may have been lost or discarded years ago, and reconstituting them meant in some cases many thousands of dollars of unexpected and seemingly needless expense. The beauty of ProCoder 2.0 is its ability to retain this critical data throughout the encoding process. Thank you, Canopus.
“Crisp and clear with superior edge delineation.” That is how judge Mohr described the BitVice’s slow but silky smooth encodes. It’s clear that speed is not what the BitVice is about, so while encode times may be long, especially when using noise reduction, the wait is usually worth it. The quality produced is comparable to that of ProCoder, “without the UI hassles.” Some key features found in the ProCoder, like inverse telecine, are not yet offered in the BitVice, so prospective BitVice users should bear that in mind. Still, the noise-reduction capabilities of the BitVice were by far the most sophisticated of any transcoder tested. This particular feature is extremely important in achieving consistently clean and sharp encodes from challenging sources like VHS or 3/4 in. The noise-reduction capabilities, combined with the simplicity of the user interface, make the BitVice the obvious choice for Mac users who wish to augment their encoding prowess in conjunction with DVD Studio Pro.
With respect to Cleaner XL, the fades and dissolves in the film footage test were not as solid as with the BitVice, but the overall quality of Cleaner XL was excellent at low to moderate bit rates. At least a few judges declared that, based on the TV drama sequence alone, this is the encoder they would buy. “A very solid performer across the board,” judge David Deignan noted. Curiously, the performance of Cleaner XL was quite a bit better than that of the Mac-based Cleaner 6 product, although XL fell short in the ease of use, user interface, and workflow categories.
One might assume from the misinformed chatter around that Compressor cannot compete with the more sophisticated (and pricier) products in this shootout. The blind test, however, demonstrated the contrary. In more than one case, a judge confused the Compressor’s output with that of the Sonic Solutions $25,000 hardware encoder!
Specifically, the judges found that Compressor displayed minimal artifacting at all bit rates tested, albeit with considerable softening. In contrast, the CCE-SP encoder retained much better sharpness overall, that reality underlying the sophistication inherent in the Cinema Craft encoder.
In the title sequence test, Compressor held its own, achieving excellent quality at the 4Mbps VBR two-pass setting. The judges noted in particular an increase in performance when some noise reduction was applied. Compressor users should therefore consider the use of minimal noise reduction (level 1) as de rigeur on most projects. The application of noise reduction significantly increases encoding times, however. The time penalty will depend largely on the speed of the Mac’s CPU.
While Compressor lacks an inverse telecine feature, its integration into Final Cut Pro and DVD Studio Pro should be a major consideration, as convenience is critical to any efficient workflow. Most notable for shooters is Compressor’s native 24p support. Here shooters can originate in 24p, edit, transcode, author, and output to DVD at 24p without ever incurring debilitating NTSC aliasing artifacts. Compressor has this workflow thing down! It’s a great all-around general-purpose transcoder that just happens to be really convenient.
Evaluating encoder performance can, of course, be highly subjective. The judges – all involved daily in DVD at a professional level – certainly had their favorites. Indeed, beyond agreement on the superiority of the Cinema Craft encoder’s performance, there was little agreement over which encoder would best serve their overall needs.
Judge Winnie Cheung preferred Compressor to the ProCoder despite the quality edge in the Canopus product. The ProCoder’s user interface was simply too much for her to overcome. Other judges, like David Deignan, saw sizable advantages in the BitVice, as it delivered high quality while maintaining an all-Mac workflow. The more Windows-centric folks favored the ProCoder and Cleaner XL, both tools being capable of outputting highly professional results. ProCoder’s stitching capability was particularly appealing to some because files of varying types can be concatenated into a single encoded file.
In the end, the judges agreed that the right encoding tool is undoubtedly the one that best integrates into the user’s existing workflow and mindset. Tools that seem too much of a hassle or whose interface is too convoluted will be of little value in one’s daily workflow regardless of the sheer quality of output.
My heartfelt thanks to the talented and astute judges who participated in this shootout: Winnie Cheung, Ken Delvo, David Deignan, Peter Veloz, Greg Mohr, David Harrington, Lance Pierre, Josh Martelli, Ryan McBride, and Jagapunt Rojboonsongsri.
Barry Braverman is director of DVD training at Video Symphony in Burbank, Calif., where he teaches Sonic DVD Creator, Sonic Scenarist, and Apple DVD Studio Pro.
To comment on this article, email the Video Systems editorial staff at email@example.com.
CBR vs. VBR?
IN CONSTANT BIT RATE (CBR) MODE, THE ENCODER assigns a fixed number of bits to scenes regardless of the source video’s complexity. If the incoming video contains high-motion or color-complex scenes, the encoder must increase the relative compression in these scenes and therefore reduce picture quality in order to maintain the constant (specified) target rate.
In variable bit rate (VBR) mode, the available bits on a disc (approximately 36,000 million on a DVD-5) are assigned according to a scene’s complexity. Scenes with greater motion or color nuances receive more bits; scenes with low complexity, like talking heads, receive fewer bits. Owing to this inherent efficiency, VBR encoding produces a much higher overall quality than CBR at the same target rate. The DVD-Video format allows for the encoding and transfer of data at variable bit rates between 0Mbps and 9.8Mbps. – BB
COPYRIGHT 2004 PRIMEDIA Business Magazines & Media Inc. All rights reserved.
COPYRIGHT 2004 Gale Group