This shows you the differences between two versions of the page.
|
h264-resource-wip:avchd [2012/02/15 03:59] driftwood |
h264-resource-wip:avchd [2012/03/05 21:41] (current) driftwood |
||
|---|---|---|---|
| Line 1: | Line 1: | ||
| - | Note from Driftwood (14th Feb 2012): This is a work in progress of AVCHD resource and info. Things we know need to be rewritten and simplified and put in here. Below is not the final order of the jigsaw - some of the sections may end up in other headings. I'm going to finish everything first, and then you guys can edit it down more. Everything is written in' my way of interpretation' ie nothing is copied from other sources - if its incorrect, correct it! | + | Note from Driftwood (March 2012): Being edited in Word & 'pdf professional' before being made into wiki.This is a work in progress of AVCHD resource and info. Things we know need to be rewritten and simplified and put in here. Below is not the final order of the jigsaw - some of the sections may end up in other headings. I'm going to finish everything first, and then you guys can edit it down more. Everything is written in' my way of interpretation' ie nothing is copied from other sources - if its incorrect, correct it! |
| ====== What is AVCHD? ====== | ====== What is AVCHD? ====== | ||
| Line 12: | Line 12: | ||
| ====== AVCHD Prediction ====== | ====== AVCHD Prediction ====== | ||
| - | A good analogy of AVCHD Prediction is how sampling works in music - you take musical sections and sample them so that you can copy and paste them later on (or before) in the song to make identical chorus or verse backing tracks. Similarly, in video the individual pictures can be broken up and sampled to form zones in the picture from previous (and/or future picture zones) to make up the current picture in the video. This process is called Prediction and it helps maximise compression to save data/memory space whilst giving good reproduction of the original part of the picture. To achieve near perfect prediction the pictures are broken down into smaller blocks of pixels for even more accurate reproduction. But they each have a special meaning and way of working. | + | A good analogy of AVCHD Prediction is how sampling works in music - you take musical sections and sample them so that you can copy and paste them later on (or before) in the song to make identical chorus or verse backing tracks. Similarly, in video the individual pictures can be broken up and sampled to form zones in the picture from previous (and/or future picture zones) to make up the current picture in the video - ie if the original sampled zones haven't changed much it forms a good prediction for the use in the current frame. This process is called Prediction and it helps maximise compression to save data/memory space whilst giving good reproduction of the original part of the picture. To achieve near perfect prediction the pictures are broken down into smaller blocks of pixels for even more accurate reproduction. But they each have a special meaning and way of working. |
| * **Prediction zones are chopped up into Macroblocks** | * **Prediction zones are chopped up into Macroblocks** | ||
| Line 42: | Line 42: | ||
| ===== Inter prediction ===== | ===== Inter prediction ===== | ||
| - | * **Inter prediction**; is the method of foretelling a block of luma or chroma samples from a picture that has previously been coded and stored (which becomes a reference picture) in the decoded picture buffer (DPB). The DPB is a repository of pictures coded from before or after the current picture in the display order. | + | * **Inter prediction**; is the method of foretelling a block of luma or chroma samples from a picture that has previously been coded and stored (which becomes a reference picture) in the decoded picture buffer (DPB). |
| + | * ** Decoded Picture Buffer in GH2 ** The DPB is a repository of pictures coded from before or after the current picture in the display order. The dpb holds the number of previous frames each P-frame can use as references. Note that the H.264 spec limits DPB size for each profile level. The GH2 holds 2 reference frames maximum it seems, even though the H264 standard at level 4.0 profile allows 1,280×720 (9 ref frames max) and at 1,920×1,080 (4 ref frames max). A good reference about how the dpb works along with 2 reference frames is found here in a TI chip h264 implementation: 'Understanding H.264 Decoder Buffer Mechanism' [[http://www.ti.com/lit/an/sprab88/sprab88.pdf]] A very good reference about maximum size of dpbs exists at x264's link: [[http://en.wikipedia.org/wiki/H.264/MPEG-4_AVC#Levels]] | ||
| Note: Everything in grey boxes to be written... | Note: Everything in grey boxes to be written... | ||
| Line 59: | Line 60: | ||
| ====== AVCHD Transform & Quantization ====== | ====== AVCHD Transform & Quantization ====== | ||
| - | Definition, | + | Definition, |
| Transform & Quantization | Transform & Quantization | ||
| Process | Process | ||
| Transform & quantization of 4x4 blocks | Transform & quantization of 4x4 blocks | ||
| *Transform & quantization of 8x8 blocks is not used in GH2 | *Transform & quantization of 8x8 blocks is not used in GH2 | ||
| - | Quajntization Scaling Tables | + | ===== Quantization Scaling Tables ===== |
| + | **Achieving the balance in Quantiser Matrix Design** | ||
| + | Higher QM values will lead to bigger rounding errors, nullifying subtle luminance variations in Keyframes. Using high values for high frequencies in B/P-frame QM could improve compressability. However, the tradeoff for using too high values could cause ringing. | ||
| + | |||
| + | Lowering the high frequency QM values in I-frames could be used to add bits to fine detail in keyframes, possibly creating better values for the B/P-frame process to work with. | ||
| + | |||
| + | Using the same QM for both Inter and Intra frames is less efficient than using different QM's, because of the difference in method used. | ||
| + | |||
| + | Finally, the standard Panasonic Lumix GXX matrices are already quite well made. The only reason I could see anyone using a custom QM is for increasing the amount of detail/size of a video (since for higher compression you could just as well use a higher quant, no need for a custom matrix) or for making things look flatter or softer and less sharpe. | ||
| + | |||
| + | <b>What is colour space - YUV or YCbCr?</b> | ||
| + | |||
| + | In simplistic terms, component video's color space is also known as YUV or YCbCr, where Y encodes luminance, U or Cb (blue) the difference between the blue primary and luminance, and V or Cr (red) the difference between the red primary and luminance. The low chromatic acuity (clearness of vision) allows a percentage data reduction of the color difference signals. In digital video, this is achieved by chroma subsampling. | ||
| + | In AVC colour space profiles the notation used is as follows: | ||
| + | 4:4:4 = no chroma subsampling. | ||
| + | 4:2:2 = chroma subsampling by a factor of 2 horizontally; this sampling format is used in the standard for studio-quality component digital video as defined by ITU-R Rec. BT.601-5 (1995), for example. | ||
| + | 4:2:0 = chroma subsampling by a factor of 2 both horizontally and vertically; it is probably the closest approximation of human visual color | ||
| + | acuity achievable by chroma subsampling alone. This sampling format is the most common in JPEG or MPEG and in the GH2's AVCHD implemntation. | ||
| + | |||
| + | |||
| DC transforms | DC transforms | ||
| Coding (CAVLC/CABAC) | Coding (CAVLC/CABAC) | ||
| + | transform coefficients | ||
| + | definition; coefficients - A number used to multiply a variable. | ||