Shrinking and Enlarging Images

[Rich54]

In a recent thread about fitting baby pictures into a 4x6 canvas, both @IamSam and @thebestcpu suggest using Edit>Transform to shrink the baby photos to size (after converting to smart objects). That method should work just fine, but I’m wondering if there's a difference in image quality when you shrink with Transform vs. shrinking with Image Size — either nearest-neighbor or the various bicubic sampling methods.

When you shrink an image, pixels are discarded (and when you expand, pixels are created). Adobe provided several sophisticated sampling options in the Image Size command to address this. My guess is that the sampling technique used in Transform is probably more primitive, although I don’t know this for a fact. Personally, when I need to shrink or enlarge an image “a little”, I use Edit>Transform. But when I need to shrink or enlarge “a lot”, I put the images in separate files, resize them in Image Size, and then bring them back into the composite file.

Any thoughts?

 

[IamSam]

Great question as it relates to the differences between free transform and bicubic sampling in image size. I would think the latter to be the better choice. But I'm also not sure about that. I will have to do some investigating.

Smart objects are unique in that they are containers for the original. The container is resized up or down but he original image retains all of it's authentic content. There's no loss when and no matter how many times the image is resized up or down.

The techniques described in the above mentioned thread where using embedded (SO) images either clipped to a panel or attached to a panel via smart object. The images where not actually being resized to the panels actual size, rather they were resized, up or down, to fill/fit within the panels on which they are to appear. In both techniques, the images were smart objects themselves and only containers for the original. You are essentially cropping the image.

If I were to be limited to resizing the actual image to a given size, provided it would mathematically scale to a prescribed size, I would use "Image Size" with a bicubic sampling method. If it wouldn't scale, it would have to be cropped after resizing. I would also do this in a separate file (perhaps one at a time or batch) and bring them into my document as needed.

 

[thebestcpu]

Hi @Rich54
I will continue on the topic as well.
First, one can control which interpolation method is used when scaling Smart Objects. The default method is set in the Preferences/Settings panel in the General panel section.
If you have just a single transformation, the quality will be the same if you use the same interpolation. So if you take individual images, you bring in and a) set the PPI the same without resampling, b) crop the way you want, and c) do a single resizing of the size/pixel dimensions you want, and then do the placement, there is no difference in quality.
One of the advantages of using a Smart Object is that I can ignore all those individual steps and just resize it as needed. As @IamSam mentioned, the last change (or cumulative changes) are applied to the original pixels just once, so there is no cumulative degradation, and I don't have to be as careful in all of the steps to get it right when working directly with the pixels.

Just a simple timesave for me.
I hope the incremental explanation compares the options fairly.
John Wheeler

 

[Rich54]

Smart objects are unique in that they are containers for the original. The container is resized up or down but he original image retains all of it's authentic content. There's no loss when and no matter how many times the image is resized up or down.

Agree. The SO itself always retains its original quality, regardless of how it is later manipulated in the composite. But simply the act of shrinking a smart object means that pixels need to be temporarily discarded in the view itself. I didn't know how Photoshop did that, but John points out that scaling transformations are controlled in Preferences and that they do, in fact, employ the same sampling techniques as Image Size. I never knew that until now, so it makes the whole issue somewhat moot.

First, one can control which interpolation method is used when scaling Smart Objects. The default method is set in the Preferences/Settings panel in the General panel section.

I never knew that!
Here's the preference panel in my CS5 version. On my very first day of Photoshop, I must have read in a book to set the preference to Bicubic, and I blindly obeyed without understanding the issue. Live and learn.

 

[IamSam]

I never knew that!

I knew about this setting but I wasn't sure it applied to a free transform function. I always assumed that it did but I was not positively sure. Mine is set to Bicubic auto. Again I will assume that Ps applies the best setting for either enlarging or reducing the content. The downside I see to using smart objects in actual sizing and not fitting, is that you are limited to Ps's basic interpolations and you have to go to prefs to change. Image size does this in one interface. I think if I were having to seriously upscale images, I would opt for one of the plugins.