Luminous Landscape Forum

Raw & Post Processing, Printing => Digital Image Processing => Topic started by: texshooter on August 13, 2013, 08:46:47 pm

Title: 1.50 or 1.67?
Post by: texshooter on August 13, 2013, 08:46:47 pm
Jeff Schewe says to downsample for the web in increments of 50%, while Sean Bagshaw says 67%. Who is right?
Title: Re: 1.50 or 1.67?
Post by: Schewe on August 13, 2013, 11:25:03 pm
Actually, I said in increments not smaller than 50%....any flavor of BiCubic will toss pixels away indiscriminately...so if you downsize in a single step of more than 50% you loose more pixels than you keep...downsampling in increments can allow you do image processing while downsampling. YMMV

BTW, who is Sean Bagshaw? Written any books?
Title: Re: 1.50 or 1.67?
Post by: texshooter on August 14, 2013, 12:31:01 am
He's Tony Kuyper's protege. In his video tutorial he recommends downsizing the original to 167% the size of the final (1st pass) then downsize that to the final size (2nd pass). But he's not the only one saying this. I'm reading this multiplier, two-thirds (1.67), all over the web as though it were a magic number, but I can't find any techy stuff to convince me it's true (or at least convince me the opiner knows what he's talking about).

If I understand your book Image Sharpening correctly, I should downsize a 5600 pixel wide image to a 1280 wide like this:

5600 x 50% = 2800 then
2800 x 50% = 1400 then
1400 down to 1280 [rounding off on the last pass instead of the first pass]

and most importantly, the multiplier above should never be lower than 50% because that would entail too many downsizing passes, which would throw too many pixels away.

Am I hot or cold?
Title: Re: 1.50 or 1.67?
Post by: Schewe on August 14, 2013, 01:23:53 am
Correct...but the last pass using BiCubic Sharper while the previous steps are normal BiCubic...
Title: Re: 1.50 or 1.67?
Post by: ablankertz on August 15, 2013, 07:04:51 pm
>>Actually, I said in increments not smaller than 50%....any flavor of BiCubic will toss pixels away indiscriminately...so if you downsize in a single step of more than 50% you loose more pixels than you keep

You seem to be saying that downsizing in increments using bicubic preserves more information than downsizing in one step. Could you give us the math that explains it? Also, what about sinc based algorithms?
Title: Re: 1.50 or 1.67?
Post by: Bryan Conner on August 16, 2013, 12:14:23 am
Correct...but the last pass using BiCubic Sharper while the previous steps are normal BiCubic...

Is there an advantage to doing the downsizing in this manner in Photoshop compared to allowing Lightroom 5 (or 5.2RC) to do this?  Thanks for sharing in this forum, Jeff.
Title: Re: 1.50 or 1.67?
Post by: Schewe on August 16, 2013, 12:37:18 am
Since CC's Image Size is new, I can't tell you definitively that PS CC is "better" than LR's resampling (I think it might be). But it all depends on the original and the amount of resizing involved...upsample 50%? Prolly not much difference, upsample 100%, maybe a tilt to CC's Image Resize. 200-400%? Prolly CC's image resize with the correct reduce noise settings. Just understand that CC's Preserve Detail with reduce noise is new. So it will take a while for things to shake out.