The resolution/contrast (MTF) model has always bothered me a bit. It seems to me there are at least 3 components that are somewhat independent, say resolution/contrast/precision.
Precision decreases when you go to a lower bit depth and also during quantization during encoding. But except in extreme cases it doesn't directly lower contrast but just makes the image less accurate and less pleasing.
Lowering resolution drives the MTF curve exactly to zero past a certain point. This probably accounts for the discrepancy between, say, the theoretical resolution of film and the observed much lower detail after telecine process since film has a long tailed MTF curve where much of the MTF value is off to the right in discarded areas.
During video encoding you can imagine that in any encoded 8x8 block you can reduce resolution simply by zeroing the highest frequency components, leaving more bits to represent the other lower frequency components. This will in turn likely reduce the contrast of adjacent displayed screen pixels as sharp edges need to be represented high frequencies. I've previously posted images here showing the results of this.
Or you can attempt to represent all frequencies but quantize each value so you have fewer bits of accuracy, as most codecs do now, not counting loop filtering and such. This lowers precision but I'm not sure it lowers contrast as much until you get to the more extreme cases.
I think there is sort of a sweet spot of balance between the two processes for any given material and bit rate but don't know of rigorous studies suggesting how to choose it. Some encoding utilities like Gordian Knot make multiple test passes to calculate resolution recommendations but I think maybe only wavelet encoding can currently attempt to really balance the two factors on the fly.
Anyway, I don't think quantization or a reduction of precision is quite the same as lowering either resolution or contrast. It's a third factor.
- Tom Craig Birkmaier wrote:
At 9:18 PM -0400 5/16/07, Mark Schubin wrote:There are two different things to consider. One is resolution. 20/20 vision is defined as the ability to just make out a high-contrast feature (like the bar of the "E" in an eye chart) that subtends an angle of one arc minute. Some have better than 20/20 vision; many have worse. For 20/20 vision, 480-line detail would just be visible on a 25-inch 4:3 screen at 9 feet.The other thing is sharpness. It is proportional to the square of the area under a curve plotting contrast ratio against resolution. The more samples in the source, the higher the curve at any given resolution. That's visible to anyone who can perceive a picture on any size set. The difference between an HD camera and an SD camera is perceptible even on a 6-hour-mode VHS recording.Mark is finally taking this thread down the correct path. A path that we have traveled on Open DTV many times before.Several participants in this thread have tried to justify the higher potential resolution that can be delivered by a 1080P format based on human visual perception. We can argue over the "finer" points of this subject, but bottom line, what consumers want and need is a sharp image on their display at the viewing distance that works for their installation.What this means is that we must take into consideration the viewing angle (the portion of the human visual field that is covered by the display) at the designed viewing distance. As Mark points out this means that we need to understand the raster requirements to deliver a sharp image AND the contrast that is delivered by the display. We must also understand that there is some variability in the visual acuity of humans viewing the display.The traditional way to measure visual acuity is to view line pairs - you could do this with both horizontal and vertical line pairs, however, human visual acuity is essentially equivalent in the H&V, so the tests are typically done with vertical line pairs. The tests measure perception in cycles per degree of the human visual field. The perception of a sharp picture is associated with resolution in the range of 22 cycles per degree. By design, NTSC achieved this on a 19" display viewed at 7 picture heights. The criteria established by NHK for HDTV was up to 30 cycles per degree, but most human observers cannot use all of this detail.Furthermore, contrast is critically important when viewing fine details. The tests usually use black lines on a white field - i.e. maximum contrast. What happens when the contrast level decreases?The details can no longer be resolved. Mark uses the example of one red pixel on a white clown's nose - The ability to perceive it will depend on both the contrast level and the size of that pixel relative to the viewing distance. At resolutions finer than 25-30 cycles per degree most people will not see the red pixel.I use the example of a blinking red stop light on the horizon at night. In this case, it may be possible for the human visual system to perceive the detail at frequencies as high as 40 CPD. But we are perceiving this because of two factors: 1. high contrast; 2. the temporal changes taking place as it blinks on and off. In full daylight you would not be able to resolve the red blinking light until it was closer to 20 CPD or even less, if the sun is behind the light.The reason MTF is so critical is that fine details are NOT captured at full contrast by a video camera. If we shoot a pair of lines at the equivalent of 22 CPD with a typical video camera, we will capture two grey lines against a white field - i.e. significantly reduced contrast.Bottom line, in order to resolve the extra detail in a 1080P image we need a very high contrast display and ideal ambient lighting conditions. Both of these factors work against large screen front projection systems.We can do the math to determine the level of resolution needed to deliver a sharp image on various sized screens at various viewing distances. In fact we did this in the SMPTE Task Force Report on Digital Image Architecture in the early '90s.http://www.pcube.com/pdf/Report%20of%20the%20SMPTE%20TFDIA.htmlThe math tells us that square pixel rasters with 480/576 lines are sufficient to deliver a sharp image on displays up to about 40 inches at nominal viewing distances of 7-9 feet. If you want to get closer (and most people will not) then you need more detail. The math also tells us that we do not need the equivalent of 1080 lines progressive until the display is nearly 100 inch diagonal at a viewing distance of three picture heights. Obviously there are some overlap areas here where you may need more detail if the viewing distance is reduced, or the display has very high contrast. Unfortunately, most screens that are greater than 70 inch diagonal suffer in their delivered MTF.But they DO NOT suffer in terms of magnifying the image relative to the viewing angle. So the result, as Bert reported, is that they tend to look soft, and you can see all of the garbage that was buried on a smaller display. At the top of the list here is compression artifacts. This is where 1080i on a big screen starts to fall apart, when it is encoded at the bitrates that are available for distribution. It is also worth noting that MPEG-2 introduces higher contrast transition between pixels when we quantize too severly. Thus we may see ringing or mosquito noise around the edges of text and other high contrast artifacts that actually violate sampling theory (it is very common to see a black pixel next to a white pixel in a severely quantized image - something that would not be produced by a camera due to proper filtering.Bert went through some math for a hypothetical 720P display and came very close to the right conclusion. on a 50" display viewed at 3 picture heights 720P is more than adequate. What he did not observe, is that very few viewers will sit that close to a 50" display. The preferred viewing distance for such displays is closer to 5 picture heights - so there is plenty of margin in such a display.None of this suggests that there is anything wrong with an oversampling display, such as a 50" 1080P display. Such a display will hide any potential raster effects (and compression artifacts), and has the advantage of resolving finer details in NON-Nyquest limited applications like viewing a web page.What it does suggest is that a 720P emission encoding is more than adequate for such a display, and will deliver a BETTER picture than a 1080p emission encoded stream with the same or slightly higher bit rate.We often lose sight of the real issues with digital compression, because the artifacts are NOT constant. If the picture looks good most of the time, we are forgiving of the times when it falls apart. To be fair, this is true for cable, DBS and DTV distribution, not for DVD distribution.Any digitally compressed source has two bandwidth requirements: 1. the average bit rate needed to deliver a sharp rendition of the source;2. the peak bit rate needed to retain that sharpness when the information content increases and the encoder is stressed.The movie industry tries to take both into account when doing the compression for a DVD release. They must meet an average bit rate requirement to make the content fit on the DVD, and they need every bit of peak bit rate that is available for the tough scenes. This typically works out to an average of 5-7 Mbps and peaks of up to 11 Mbps. REMEMBER, we are talking about 720 x 480 @ 24P here, not HD sports.With MPEG-2 compression you can eat up 20 Mbps for peaks quite easily, even with SD source. Many of the compressionists I have talked with cite example off source that throws peaks up to 30 Mbps for an SD movie. So the compressionist has to figure out how to make that fit in 11 Mbps. When this happens in most distribution systems we see blocking artifacts - the compressionists do what they can to hide these artifacts.With 720@60P we frequently exceed the bit budget for the ATSC DTV channel. Yes, we can pre filter and do other tricks to reduce the bit rate, but this sacrifices resolution and contrast. With 1080i we always exceed the ATSC bit budget for peaks - in fact we need the entire channel just for the average bit rate.If you doubt this, set up a 100" 1080P display and view it from three picture heights. You will see blocking artifacts and quantization distortions out the wazoo! Distributors are getting away with this crap because few consumers have BIG screen displays, and they are accepting of compression artifacts that ARE NOT constant.What is REALLY IMPORTANT here, however, is that the extra detail that is possible with 1080p is the first thing that gets quantized away when the source is encoded. This is why you can often deliver superior results by encoding the source at a lower resolution (with the inherent benefits of oversampling) then upsample when feeding that big screen 1080P display. This is exactly what Hans did with the EBU demo.Improved compression codes will help the situation. When we eventually achieve the 2:1 improvement in compression efficiency said to be possible with H.264, we will be in much better shape to deal with the peak bit rate requirements. Unfortunately, most distributors want the recovered bits to stuff more programs into the available bandwidth. Even with this, however, 1080i and 1080P will continue to be stressed if the encoded bit rate remains below 18 Mbps with NO HEADROOM for the peak requirements. The net result will be significant reductions in delivered resolution during the peak requirements.Regards Craig ---------------------------------------------------------------------- You can UNSUBSCRIBE from the OpenDTV list in two ways:- Using the UNSUBSCRIBE command in your user configuration settings at FreeLists.org - By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word unsubscribe in the subject line.
-- Tom Barry trbarry@xxxxxxxxxxx ---------------------------------------------------------------------- You can UNSUBSCRIBE from the OpenDTV list in two ways:- Using the UNSUBSCRIBE command in your user configuration settings at FreeLists.org
- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word unsubscribe in the subject line.
