theSuede Offline Upload & Sell: Off
|
p.1 #13 · What sharpening do you use B4 web | |
Use a good resize algorithm like Lanczos or SinC in stead of BiCubic. I know this is not a popular choice because of the inability of Photoshop to use any other algorithm except the three included ones natively, but they all stink outside their intended uses...!
Nearest Neighbour is for squarely pixelized integral value upsampling, BiLinear is for smooth and geometrically good (from a pure mathematical point of view) but soft downsampling, and Bicubic.... Well I don't really know what it's for. It is from some points of view slightly better than BiLinear, but still not a good algorithm for two-dimensional matrix resampling. It seems to be purely (machine-time) performance oriented in the Adobe version of it.
If real detail retention is as important as apparent "sharpness" (edge acuity) then you have to go outside the PS "box". The only way to make Photoshop do a geometrically and mathematically "good" resize is to do it in LAB colour-space, or set the "mix RGB with Gamma:" to 1.00 in the "Colour Settings" menu. Otherwise Photoshop will attempt to do linear computations on your logarithmic RGB-space, and this really kills detail... Unfortunately resetting the "mix RGB with Gamma" will cause problems in most other areas as PS is "after-corrected" to do most of its transformations and filter-work in a Gamma-corrected space when working in RGB. This is really poor programming.
The examples that has been given to you above are all good, so if you like the result of any of them, use it! I just wish that you wouldn't have to resort various work-arounds to "cheat the program" when doing a simple task like this.
Personally I've only found one way to make Photoshop do a visually pleasing and detailed downsize - oversharpen the large original (like - LOTS and LOTS! The normal "sharpen" or "sharpen more" will normally do just fine - depending on the resampling rate you will use of course, and then downsample with BiCubic Normal. This will counteract the strange behaviour that the BiCubic shows (in Adobe's interpretation of it)...
Sorry for the rant, but I really think it's kind of strange (and irritating) that Adobe still after 13 complete version updates or so of this program still can't do a simple thing like a resampling right...
Below I included an example of the differences between algorithms that I originally made for a local forum. Upper left is the "original" 100%, the others are downsampled to an uneven value to make the differences more apparent (an even 50% or 33% is easier on the computation). I believe the resampling was set to 42-44% or something. I then upsampled the results 2:1 with "nearest neighbour" to make the difference at pixel level easier to see. The BiCubic downsampling was done in Photoshop (BiCubic "sharper"), and the Laczos was made externally to PS.
NO sharpening has been applied to any of the samples!
|