CompuServe Messages

#Wish list

    07-Jan-95 16:46:06
Sb: #145380-#Wish list
Fm: Marion K. Marks 70700,2777
To: John Ellis 72440,3046
Not so. When JPEG performs subsampling, it looks at "u" and "v" pixels of chrominance information for every "y" pixels of luminance info and considers them ONE pixel of chrominance (averaged from the "u" and "v" values of those pixels). The format for specifying subsampling is "y:u:v:" so "4:4:4 subsampling" basically says, "For each 4 pixels of luminance, take the corresponding 4 pixels of chrominance and treat them as 4 pixels." In other words, don't subsample at all. The chrominance values of the four pixels are treated as individual pixels. But the human eye is more sensitive to abrupt changes in luminance than in chrominance, so long ago the broadcast experts who designed the NTSC system figured out that they could save bandwith by treating every two pixels' worth of chrominance as if it were only one. (Actually, they didn't call them "pixels" back then — being analog, it was more along the lines of cycles of the 3.58MHz color burst reference sine wave attached to the front porch of NTSC scan lines.) Thus we have 4:2:2 subsampling in NTSC broadcast signals. This basically means that the luminance can change every pixel, but the hue and saturation can only change every other pixel. Due to human visual perception limits, this is an acceptable compromise for the bandwidth gained. (Actually, it's not hue and saturation per se, but "redness" and "blueness," but that's a LOT more complicated and it's much easier to think in terms of hue and saturation. The basic point I'm trying to make is the same either way.) So if JPEG subsamples at 4:2:2, the chrominance averaging is no more than that performed by NTSC anyway. Nothing is really lost (except some possible Nyquist artifacting, but that's not likely to be noticed at 30 frames per second), and quite a bit of extra compressibility at a given quality level can be gained. But when JPEG is set to 4:1:1, then we have FOUR pixels of chroma info averaged into one, over the space of four pixels of luminance info! So hue and saturation are basically held to the same value over FOUR pixels of changable luminance! This is TWICE the subsampling that NTSC itself does, so the chrominance detail of the image IS noticeably reduced when brought to NTSC. The three various values of subsampling for JPEG exist because there are different things people do with JPEG. 4:1:1 may be just fine for CD-ROM animations for games, for instance. It would save more room and reduce luminance artifacting from JPEG at the expense of less-noticeable chrominance detail, since CD-ROM animations have to be compressed quite heavily to fit in the 300kB/sec. bit rate that MPC2-spec double-speed CD-ROMs require. For output to prepress for glossy magazines, 4:4:4 may be the only acceptable value, since NTSC isn't involved here and there is no maximum data rate or file size, and JPEG would be used at all only to save a little disk space over other lossless compression schemes. But for true broadcast-quality NTSC, 4:1:1 is unacceptable. 4:2:2 is acceptable but may result in Nyquist artifacting in the chrominance data. 4:4:4 may be overkill for most situations, but would not permit Nyquist artifacting to occur in the chrominance data (not that any but the most highly-trained eye would notice if it did).