Quote:
Originally Posted by Huge Knuts
Curious...first post and everything, but I honestly want to understand something from the point of view of people who know BD very well.
Can anyone expain to me the science around this statement from Michael Bay?
"My films finished in Blu-Ray are better - more true." - Michael Bay
What does "more true" mean and what specifically from a technical point of view would net a the movie to look "more true"?
|
Assuming Michael Bay saw a 'Blu-Ray' encode of a portion of his movie. All it means is that he has a higher bitrate encode of a segment that he understands well. That he understands each frame of picture of TF is likely NOT in dispute, he not only is reponsible for visualizing and framing the pictures originally, he also spent time in the editing room on it.
All moving pictures is encoded by taking each sequence of pictures and organizing them as group of pictures (GoP). There's three kinds of pictures in a GoP. An I picture which you can think of as JPEG like - an anchor frame where it is encoded spatially using only "intra' modes only, each pictureis broken into macroblocks of 16x16.
Macroblocks in each I picture is encoded either by DCT convertion from spatial to frequency domain and quantizing them or one of the fancy intra modes where you take a diagonal or horizontal or vertical column of pixels and take the delta of the pixels of a macroblock to the column/row and then transforming the residual (left over) with a DCT and then also quantizing them.
With macroblocks P pictures, you can also take the difference between the previous I or P pictures displaced by what is known as a motion vector (ie: to approximate the position of the current macroblock as a macroblock in the previous frame. This difference known as the residual is also then DCT and then quantized.
With B pictures, you gain the ability to choose between a past or future I or P picture. (B pictures in AVC can also be used as a reference picture, but that is rarely done because it becomes too complex).
Note there is also an in loop filter step in here (in VC-1, it's known as overlap smoothing and inloop filter, in AVC it is known as deblocking filter). There are optional steps, but is always turned on from what I can tell in all HDM encodes today. This in loop filter step hides obvious blocky encoding artifacts but introduces other subtle less detectable artifacts. The artifacts typically is difficult to tell from a slightly out of focus image unless it becomes too aggressive in which case it becomes very obviously smudged. The strength of the in loop filter is determined PRIMARILY by the quantization factors in each macroblock, as quantization is applied at multiples of 4x4 of 8x8 subblock boundaries in the macroblock.
Note that quantization is the only part in the compression process that introduces lossiness. When decoding the pictures, the quantization number is multiplied back to the frequency values, the inverse discrete cosine transform is done, the inverse intra mode or the motion compensation (remember the motion vectors) and the original macroblock is rebuilt --- except it is only an approximation of the original as detail is already lost from the quantization and from the resulting in loop filter.
There are several technically sound reasons that could account for Michael Bay's preference for the presumably BD-encoded content over the HD encoded one. The primary note to remember is that he has the original raw uncompressed originals! If he is familiar with it he will likely recognize some of the compression artifacts if he sees them. If he has good memory, he may not even need to do it side by side as he knows what the subtleties the original has and the important segments (to him).
(a) He might have compared an higher peak or average bitrate AVC to a lower bitrate VC-1 encode. Note BD allows a peak bitrate of 39Mbps for video. VC-1 allows a theoretical 28Mbps peak for video (which is seldom hit because bits are stolen by higher bitrte audio or pip tracks).
(b) TF is AVC though, so another possibility is that he is comparing a higher bitrate AVC and a lower bitrate AVC and liking the higher bitrate more. This is the most technically sound one because if each macroblock is encoded the same modes between the two encodes, and the quantization factor is smaller in the higher bitrate one.
Remember the quantization that we talked about? Quantization is allowed to vary or change in each macroblock of a picture. Quantization has to go up mainly if the encoded picture exceeds the allocated budget ( to stay within a target bitrate --- the physical limit of each storage media in this case). Lower quantization means more detail is preserved, and more important less in loop filtering or smudging takes place.
More quantization = less entropy = more zeros in the post-quantize values & smaller residuals = more lost detail = more in loop filtering applied to cover the loss.
I've seen a person familiar with certain source materials look at extremely high bitrate AVC encoded video and point out certain frame numbers and regions that he thinks look weak, and if you go back to the frames and regions and apply a measurement, he's right. IE: because he knows the original well enough, he can spot what most people can't.
With us normal public though, we don't have the original, so when we see soft regions, we don't know if it is because of over-quantization or actual out of focus cameras or narrow depth of field in camera, or even deliberate pre filtering (removing grain?) that happens sometimes.
All countries
Argentina
Austria
Belgium
Brazil
Bulgaria
Chile
China
Colombia
Cyprus
Czech Republic
Estonia
Denmark
Finland
Greece
Hong Kong
Hungary
Iceland
Indonesia
Ireland
Israel
India
Latvia
Lithuania
Malaysia
New Zealand
Norway
Philippines
Poland
Portugal
Romania
Russia
Singapore
South Africa
South Korea
Sweden
Switzerland
Taiwan
Thailand
Turkey
Ukraine
United Arab Emirates
Linear Mode
