Resize Transform

[Marpel]

A couple questions about the Resize Transform and, specifically, sharpening.

1. Why is Bicubic the only method to which a sharpen slider is attached?

2. And, if using Bicubic along with the sharpen slider (say at the default 16.7%) does this mean the sharpen and smart sharpen transforms are not necessary (or the sharpen slider for this method only, is preferred)?

Thanks

Marv

[Dieter Mayr]

Hello Marv

My understanding:
As Bicubic is the probably most used resizing method (because of it's good quality at fast speed) the additional sharpen is added for convinience.
I never would say that additional sharpening is unneccessary as I use sharpening very conservative and make the final sharpening just before the output (screen or printing), just to the amount the desired output size needs.

You might want have a look in the "Image Resampling" white paper.

[jsachs]

The sharpen slider in bicubic resampling has a very subtle effect. Sharpening is something of a personal preference and also depends on the subject matter and how you plan to view the image (on-screen, print, how big, etc.). I generally do some mild sharping after resampling (unless using the nearest neighbor method).

[Dieter Mayr]

A good article (beyond others) about sharpening can be found on the Cambridge in Colours site:
http://www.cambridgeincolour.com/tutori ... pening.htm

[Marpel]

Thanks for the replies (and link).

Although I have an understanding of the sharpening philosophy (although am not an expert by any stretch of the imagination), I was curious as to why only one of the resizing methods had a sharpen slider. I have only used Lanczos and also wondered if it was so much better than Bicubic that it did not need (or less needed) the additional sharpening that Bicubic did.

Marv

[jsachs]

The answer has to do with the way bicubic and higher order downsampling is implemented, namely by smoothing the image prior to reducing its size. This has the effect of computing each pixel in the result image as a weighted sum of pixels in the input image.

In the bicubic case there is a way to alter the smoothing kernel so as to produce a sharper result image and this is what the sharpen slider does. There is probably a way to do something similar with Lanczos sampling, but if so I have never seen it written up. Personally I find the differences between different resampling methods nearly impossible to distinguish in the end result when applied to real world images. The method I like best is bilinear followed by one pass of Sharpen.

[Marpel]

Thanks for the further reply Jonathan.

I find your chosen method interesting (for some reason, I thought bilinear was inferior to the others). I obviously need to try each method on the same image just so I can see for myself.

Thanks again,

Marv