I spend a fair amount of time interacting with industry experts on e-mail forums. Recently, one forum, the DVDList, has been abuzz about what most members perceive as the dawn of a new era in high-quality HDTV: the end of the format war between Blu-ray and HD-DVD.
I am not the most popular member of the DVDList at the moment. I have taken the position that Blu-ray will not replace SD DVD as the delivery system of choice for home theater systems. I believe Blu-ray will appeal primarily to the videophile audience and those who like to collect movies rather than renting them, which is how most movies are acquired for home viewing today.
The Blu-ray zealots claim that all TVs benefits from the use of a 1920 × 1080 raster updated progressively (i.e. not interlaced as with 1080i). They may be right, but for the wrong reasons. They claim the improvement in image quality on a 1080p display fed by a Blu-ray player can easily be seen even on the 32in to 42in panels sporting the 1080p raster. In this, they are wrong, as this column will explain.
A recent post from Jerry MacKay, Littleflick Pictures, to the DVDList will help explain what's going on here: “I just rented a movie last night, and it had a commercial on it for Blu-ray. It showed footage from ‘Pirates of the Caribbean,’ ‘Wild Hogs,’ ‘National Treasure,’ ‘Lost,’ etc. Watching the commercial, I thought ‘Man, Blu-ray looks GREAT!!!’ The movie I rented was an SD disc, played from an upconverting DVD player going to my Sony 46in DLP projection HDTV.”
Thanks Jerry! You just revealed the dirty little secret about SD DVD. Because the format is optimized for interlaced displays, it does not typically deliver the full resolution potential of a 480/576 line format. The level of vertical detail is significantly lower than what can be delivered using frame-based 480/576p formats.
SD DVD players could have used the full resolution potential of 480p from day one if they had included a cheap convolution filter to reduce the vertical detail in the interlaced composite and S-video ports of the players. Instead, SD DVD authors are forced to reduce the levels of vertical detail for everyone. Scenes with higher levels of horizontal detail — like car grills — must be prefiltered to prevent severe aliasing on interlaced displays. The average level of vertical detail is reduced throughout because of the filtering required for presentation on interlaced displays.
The commercial MacKay watched via his upconverting SD DVD player looked sharper for good reason. It was produced in HD using the full resolution available and then resampled for encoding as 480p rather than 480i. It has significantly more vertical detail than the rest of the content on that disc. This makes it pop on today's big-screen progressive displays and alias on those old analog interlaced displays. The Blu-ray vendors are not worried about the folks watching on those old displays. They are trying to get viewers to upgrade to Blu-ray on big progressive displays.
Authoring a movie title for SD DVD is a complex process. The average bit rate for the movie must be calculated so that it will fit in 4.7GB. A compressionist typically reworks scenes that exhibit compression artifacts at this average rate. They may allow the bit rate to increase for that scene or low-pass filter it to reduce the stress on the encoder.
This raises an important point. DVDs are one of the few digital video distribution mediums where people are paid to make the content quality as good as possible given the bandwidth restrictions. The other medium where video quality is taken seriously is with some of the Internet video download services. Apple's iTunes store is a prime example.
The SD movies that Apple sells and rents via the iTunes store are encoded with 640 × 360 samples from progressive source. There is no prefiltering for interlaced displays. This resolution is slightly shy of the 720 samples per line for SD DVD, but the vertical resolution is superior because the source is not filtered for interlace, which reduces the effective resolution of 480i to less than 300 lines.
Every TV department of any consumer electronics store today touts the wonders of 1080p. The salespeople say 1080p is the best possible resolution for an HD display. They are less likely to tell you that size matters.
1080p is a good thing on almost any display, especially those 50in or larger. Display oversampling is beneficial. Increasing the pixel density helps to hide the visibility of the raster and can be used by other applications.
The key issue here is the sampling theorem. Shannon and Nyquist learned that when sampling images from the real world, a sampling rate 2X or more than the highest frequency we want to capture is needed. Cameras typically use optical filters ahead of the sensors to limit the level of detail seen by the sensor to half the resolution of the sensor. This allows the camera to see horizontal and vertical lines that are not perfectly aligned with the sensor sites, and it helps to eliminate aliasing on diagonal lines. Increase the sampling rate beyond 2X, i.e., oversampling, and things get even better.
When an image is resampled to a lower resolution for emission encoding, it reduces noise and puts more information under the modulation transfer function (MTF) curve, which improves the contrast level viewed by a human. The picture appears sharper.
In the world of computer applications, the Nyquist limits can be ignored for certain applications. For example, we can draw single pixel lines and have sharp edges on text. We perceive some aliasing, but we get much higher contrast. This forms the basis for the graphical user interface and much of the content on the Web. A computer can also display Nyquist limited imagery like video and digital photos; however, the highest frequency in these images must be reduced by half.
So a 1080p display is a good thing. It makes it possible to see more of a Web page when a browser is put into a big-screen TV. And it allows us to upsample whatever source the TV receives with more samples to present the available information. The question, however, is: Can you see this extra detail? The answer depends on two factors: screen size and viewing distance.
Recently, CNET Labs directly compared two 50in displays, one with 1280 × 720 resolution and one with 1920 × 1080 resolution. The source used to feed both was 1080p from a Blu-ray player. (See “Web links.”) The observers found that on TVs 50in and smaller, the added resolution has only a minor impact on picture quality. Bottom line: It's almost always difficult to see any difference — especially from farther than 8ft away on a 50in TV.
Carlton Bale took exception with this conclusion in a story published on his Web site. He created an interesting diagram that attempts to relate how much resolution is needed for various screen sizes at different viewing distances. (See “Web links.”) While taking issue with the CNET study, his conclusions are nearly the same.
The realities of decoupling
In the world of analog TV, everything is tightly coupled. The cameras, equipment in the studio and the TV all operate synchronously at the same scanning rates. The main reason that NTSC has improved in recent years is related to CCD cameras that oversample in the horizontal axis. These cameras typically use more than 1000 samples per line to produce a 720 × 480 raster.
With DTV, acquisition, emission and display are decoupled. It allows creators to use the highest resolution possible to acquire the image and then resample it to lower resolutions for different applications that range from the big screen in the family room to the 2in handheld display. In bandwidth-constrained channels, such as those that exist today for cable, DBS and ATSC broadcasts, delivering the highest pixel count can reduce the delivered image quality as the EBU found when comparing 720p and 1080p formats for emission. (See “Beyond MPEG-2” in “Web links.”)
The MPEG compression algorithms are low-pass filters. The more the source is quantized, the lower the quality of the delivered samples. Resampling to a lower resolution for emission can improve the delivered image quality in a bit rate constrained channel.
What really matters is that we deliver the highest quality samples possible to the receiver, which must scale everything to the native display resolution. The folks that author DVDs already know this. It's time for broadcasters to understand that less, not more, may be the most productive path to improving the quality of DTV.
Craig Birkmaier is a technology consultant at Pcube Labs.
Send questions and comments to: firstname.lastname@example.org