Bigger factor for blockiness/compression artifcats

[pannayar]

I understand the basics of half D1 (352x480) having more (is it almost exactly double?) bits available per pixel compared to full D1... and hence at lower bitrates halfD1 would reproduce a "softer" but less blocky image...

The part that seems counter-intuitive to me is the less blockiness with halfD1, I mean the halfD1 pixel should be twice wide horizontally and wouldnt that contribute to a "blocky" result more than fullD1 at the same bitrate where the problem seems less severe to me (you have many pixels, but only limited bits to represent them).

Btw, why is halfD1 352 and not 360?

[Cornucopia]

Think of it this way...

When a scene is "blocky" due to lower rez (as in 352x480), it's transition capability is on the order of what would normally (full rez) be every OTHER pixel, right?
But when a scene is blocky due to compression artifacts, it's transition capability is on the order of every 16 pixels, with SHARPER contrast between one side of a transition and the other side.

Think of D1 with no overscan as 704, NOT 720. 720 is with overscan.
1/2 of 704 = 352.

Scott

[edDV]

A good overview tutorial for MPeg2.
http://www.fh-friedberg.de/fachbereiche/e2/telekom-labor/zinke/mk/mpeg2beg/beginnzi.htm

"Blockiness" occurs at the block or macroblock boundaries. Blocks are 8x8 pixels. Macroblocks contain 2x2 blocks or 16x16 pixels. Half the horizontal pixels doubles the bitrate allocation to remaining macroblocks.

CCIR-601 (D1) used 704x480 to represent a 4:3 frame. Half D1 is thus is 352x480. 720x480 results from 8 guard pixels added to left and right to buffer horizontal timing issues avoiding side crop. When you cap from an analog tuner to 720x480 you will see 8 pixel side stripes. DTV tuners (ATSC or DVB) capture to 704x480 (4:3 or 16:9) without side stripes.

MPeg2 allowed added picture information to fill out the 720 width but note that aspect ratio specs are identical for 704x480 and 720x480.

Example of analog capture side pixels.

[davideck]

I understand the basics of half D1 (352x480) having more (is it almost exactly double?) bits available per pixel compared to full D1... and hence at lower bitrates halfD1 would reproduce a "softer" but less blocky image...

The part that seems counter-intuitive to me is the less blockiness with halfD1, I mean the halfD1 pixel should be twice wide horizontally and wouldnt that contribute to a "blocky" result more than fullD1 at the same bitrate where the problem seems less severe to me (you have many pixels, but only limited bits to represent them).

Half D1 sampling will reproduce a "softer" image at any bitrate. The reduction in horizontal resolution caused by downsampling is independent of the blocking artifacts caused by compression.

The width of the pixels determines the limit of horizontal detail that can be preserved. If done correctly, downsampling to 352 must also include lowpass filtering the source image to reduce its horizontal resolution to what can be preserved and reproduced at 352 samples per line. Filtering out higher frequencies reduces detail and softens edges, but halving the number of samples reduces the required bitrate for a given level of blocking artifacts.

Downsampling video to 352 is like turning down the treble control on an audio signal. Some sources will suffer, some will not. Practical filtering for 352 samples per line reduces the horizontal resolution to below analog broadcast quality. Even good VHS sources lose information.

[jagabo]

The more pixels you have (the larger the frame size) the more information you have to throw away to achieve the same bitrate. This is what causes macroblock artifacts. Half D1 resolution has half as many pixels to compress, so at a particular bitrate will generate fewer macroblocks artifacts. If the artifacts aren't there you won't see them.

By using half D1 you are trading away resolution to get fewer macroblocks.