Which is better for Downsizing images to go on Second Life - Bicubic Sampling or Bilinear Sampling?

 

I shared the results of the Graphics Resolution Test with the avatar Wayanoru Megadon and he gave me a link to an article suggesting that the resampling technique used before uploading might make a difference in the quality of the image as viewed in SL. The article suggested that bilinear sampling might be superior to bicubic sampling.


http://beqsother.blogspot.com/2019/02/compression-depression-tales-of.html

 

In the images below, I show very small 88 x 57 px samples taken from a image using both techniques. I then show enlargements of the pixel sizes to 880 x 570 px so you can see the difference in the pixels more clearly. Lastly, I uploaded the 88 x 57 px images to SL and took screen captures of the results.

Return to Home Page

88 x 57 px sample downsized with bicubic sampling

 

 

 

88 x 57 px sample downsized with bicubic sampling

 

 

 

 

Image above with pixels magnified 10x for visibility

 

 

 

 

Image above with pixels magnified 10x for visibility

 

 

 

 

88 x 57 bicubic sample as it appears in SL when applied to a prim (screen capture)

 

88 x 57 bilinear sample as it appears in SL when applied to a prim (screen capture)

 

 

Conclusions

 

First, given the very small amount of data that SL had available in the 88 x 57 px images, I think it did a good job of upscaling to provide a reasonable image when I pasted the small graphic onto a large object in SL. Whatever algorithms SL is using when they downsize or upsize images to be viewed inworld, they do present good quality images for most people using the program.

With respect to the use of bicubic or bilinear resampling, to my eye, I think the edge goes to bi-cubic resampling, as evidenced by the greater definition and higher contrast in both the enlarged samples and the images displayed in SL.

The difference is most noticeable in the top eyelashes towards the outer corner of the eye.

 

Of course, this is a matter of taste, but generally more pronounced detail at the pixel level leads to a better overall image when viewed in SL.

 

- Rob Fossett

2023 08 15


 

 

 

Return to Home Page

 

© Copyright. All contents are copyrighted by the individual AI Graphic Artists, 2023

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Note: Images are copyable and available for non-commercial use only. All images are copyright (c) 2023 by the graphic artists who produced them.