Digital Light & Color

I realize this is one of those things that have little real world significance, but…

I needed to re-size an image to below 10% and found that when using any of the 3 Lanczos methods, the result was a weird coloured (almost like a colour negative transformation) version of the original. The other, non-Lanczos methods seemed to work fine, so nothing lost, however, just curious if the Lanczos methods are restricted to 10% and above (worked for 10% but not for any number less).

Marv

Marv, I couldn't say I have seen something similar you described, so i runned a little test, images in the range of about 3500 x 2500 pixel, downsized to 100 x 75 pixel.
Tried with all 3 Lanczos variants, but i could not find anything unusual.
Could you post a image that produces that effects ?

Dieter,

The image that I had been having issues with was a photo of a red heart against a white background, which I do not have any longer. However, I was able to recreate the same result by generating a new white image and putting a red square into it (by using a square mask to composite a red image and a white image). Not sure the colours necessarily matter.

While doing this, though, I discovered that the problem only occurred if the image to be down-sized was 48 bit colour. Perhaps you used a 24 bit image in your test(s)? If so, could you retry it with a 48 bit image and see what happens? My initial image size was 6048 X 4032 but I have also tried it with smaller 48 bit images and reached the same result.

If your results still do not match mine, I expect the issue is at my end rather than the image itself.

Thanks,

Marv

Not seeing it here either. What version of PWP are you using -- I vaguely remember fixing something like this a long while back.

You are correct, Jonathan. I thought I had the latest release, however I had 7.0.10 and not .11. Updated and problem solved.

Thanks,

Marv

I'm pretty sure I didn't fix any resize problem that recently so probably just the re-install fixed it.

This is not really related to the original comment, but may be useful in extreme downsizing anyway: using the nearest neighbor to severely downsample an image with high frequency details. Those details tend to get lost with other kernel weighted averaging downsampling methods.

http://www.dl-c.com/cgi-bin/discus/show ... #POST11650

p.s. look at the date there... Man!