Graphics packages & RAW converters offer different output formats with different colour sub-sampling options.
Colour sub-sampling is specified as a ratio. For example, 4:4:4 is just the ratio of the colour components, Y:Cb:Cr, where Y = luminance, Cb = blue difference, Cr = red difference. In this article I'll use the alternate term for YCbCr, YCC. The colour components are not sub-sampled in 4:4:4:, but are sub-sampled (a form of compression that groups colour components into block of pixels) in the more common 4:2:2, 4:2:0 or 4:1:1 varieties. The eye is not very sensitive to the loss of colour resolution (it's more sensitive to the loss of luminance resolution - black & white) so you can usually sub-sample the colour stuff to reduce the filesize, without much visible effect.
The example below shows the same crop processed using the two different colour sub-sampling methods. You can see in the "colour compressed" version on the left how, at the areas indicated, the colour variations have been simplified compared to the same areas on the right.
Here is an example of a full-sized (14.6 Mpixels - Pentax K20D) PEF saved with 2 different TIFF formats & also as 100% quality JPEGs with 3 different colour sub-sampling schemes in SilkyPix:
Format | Filesize | Comp. to 8-bit TIFF | Comments |
TIFF-16 bit | 85,071 KB | 200% | Only needed if more PP is to be performed. |
TIFF-8 bit | 42,585 KB | 100% | Good for final output. |
YCC444 | 12,977 KB | 30% | aka 1x1 (H x V) chroma sub-sampling, but no chroma sub-sampling applied. 132%. Not EXIF compliant. |
YCC422 | 9,818 KB | 23% | aka 2x1 (H x V) chroma sub-sampling. Taken as 100% JPEG size ref. Common for DSLRs. |
YCC420 | 7,611 KB | 18% | aka 2x2 (H x V) chroma sub-sampling. 78%. Common for Web use. |
YCC411 would be the same size as YCC420, but is not EXIF 2.21 complaint.
TIFF is usually RGB444, but can also use YCC444
From
EXIF 2.2 spec, page 5: 4.4.3 Pixel Composition and Sampling
The combination of pixel composition and pixel sampling for image data shall be RGB 4:4:4 and either Y:Cb:Cr = 4:2:2 or Y:Cb:Cr = 4:2:0.
Which indicates that JPEG YCC422 & JPEG YCC420 should be used for maximum comparability.
Outside the field of photography, Y'CbCr411 is also used in NTSC SD DV encoding, while Y'CbCr420 is used for both PAL SD DV & for HDV encoding.
**********************************************
TIFF can have optional lossless compression (usually LZW). As an example of TIFF compression, I've processed this album shot of a Kookaburra, but used the uncompressed TIFF version converted from RAW, not from the JPEG version shown here. (Click on it to see larger version):
Program | TIFF Compression | Relative Filesize | Comments |
Silkypix Pro 4.1 | None | 100% | |
SilkyPix Pro 4.1 | LZW low | 114.9% | |
Silkypix Pro 4.1 | LZW high | 74.6% | |
Infranview 4.25 | LZW | 112% | EXIF was stripped off |
FastStone 4.0 | LZW | 67.8% | |
XnView 1.97 | Binary: LZW, Colour: No Compression | 98.9% | |
XnView 1.97 | Binary: LZW, Colour: LZW | 112% | |
XnView 1.97 | Binary: LZW, Colour: LZW+Prediction | 67% | |
XnView 1.97 | Binary: No Compression, Colour: LZW+Prediction | 67% | |
The high compression programs appear to use LZW+Prediction (XnView & Silkypix mention this explicitly - Silkypix uses it for "LZW high"). Some programs can't handle LZW+Prediction.
LZW compression by itself was unsuccessful with this particular photo and increased the size of the resultant file. But, with simpler photos, TIFF LZW-only compression may be of benefit:
Here are the results with this photo outputted from the Silkypix Pro RAW Converter:
TIFF Compression | Relative Filesize | |
None | 100% | |
LZW low (LZW) | 88.5% | |
LZW high (LZW+Prediction) | 40.0% | |
Infranview strips off EXIF when it saves TIFF files & does not offer LZW+Prediction.
**********************************************
I'll now explain the relationship between Mpixels & the TIFF (8-bit) filesize.
Consider a "14.6 Mpixel" camera (Pentax K20D). The size of the original image in the RAW format is 4688(w) x 3124(h) pixels. Multiply W * H = 14,645,312 = 14.6 Mpixels (camera manufacturers use 1 million pixels = 1 Mpixel).
Now not all these pixels contain info that ends up in the final photo. There is a strip of black-level data included on the outside. The actual size of the output of the RAW converter is 4672 x 3104 = 14,501,888 = 14.5 Mpixels.
Ok, that's Mpixels. To get bytes, multiply by 3 (8 bits = 1 byte. There are 3 bytes per pixel used to indicate the RGB colour of each pixel - 24 bits, hence "24-bit colour")
= 4672 x 3104 x 3 = 43,505,664.
The value mentioned in the table for an 8-bit TIFF is 42,585 KB. Why the difference? Well in KB (the size shown in Windows Explorer under "Properties"), we typically use 2^10 = 1,024 instead of 1,000. So 43,505,664/1,024 = 42,486 KB. (That's the case with general computer. However, HDD manufacturers use 1,000,000 instead of 2^20 - 1,048,576 - because they can quote a 4.8% larger "MB" value that way. And by using 1,000,000,000 for a GB, instead of 2^30, they get a 7.4% bigger value. Finally, with 1,000,000,000,000 for a TB, instead of 2^40, they can quote a 10% bigger figure.)
There is also a slight overhead from the TIFF file-format header, hence the final filesize of 42,585 KB. Divide by 1,024 again to get the MB value, 41.6 MB.
Dan.