Return to Digital Photography Articles
Photoshop Upsizing: All at once or incremental?
It has been widely stated that a good way to increase the size of a photo in Photoshop is to incrementally upsize it, rather than simply doing a single one-shot resize. In fact, I am writing this because I had suggested this as a better method to upsize in Photoshop, on the basis of what I had heard, without actually confirming the results for myself. After another poster, hollis_f, in DPreview countered my suggestion, I went ahead to check the results myself.
So much has been said to this point that it seems to have been generally accepted as true. Several publications (including even one of Scott Kelby's Photoshop tips guides) state this as a tip. It doesn't seem to make a lot of sense, though, that you'd get better results by incremental upsizing than the algorithm used for a single upsize.
The following shows a comparison between the two methods in Photoshop CS. I took a photo that was shot with a 6 megapixel digital camera, in JPEG super-fine mode (which should be relatively low JPEG compression), and compared the two methods of upsizing a portion of the photo. One was upsized 14 times in a row by 110%, giving a total increase in size of 379.7%. The other was resized by a single upsize of 379.7%. In both cases, I used Resample Image: Bicubic to reduce any extra sharpening artifacts.
Conclusions?
It is hard to observe a significant difference between the two, although there certainly is sone (as is shown by the layer subtraction result). If anything, it appears that there could be slightly better reproduction of detail in the direct upsizing method. Have a look at the texture on the wooden vase. Of course, note that this is one sample experiment from a single image. Other image content might behave very differently. Also note that by saving these images in the JPEG format (with moderate quality), I am also introducing some additional JPEG artifacts. Visual observation indicated that the visual effects of resaving were almost non-existent, partly due to the large block size (after 400% upsizing).
I wonder if perhaps this method is no longer necessary with Photoshop CS. Perhaps in prior versions of Photoshop one did indeed get better performance out of repeated incremental upsizings.
100% Crop Comparison of Upsizing Methods
The first image shows the standard method: a single upsize operation with the Image > Image Size method, with a percentage increase of 379.7%.
This next method shows the effect of 14 Image Size operations in a row, each with 110% upsizing, giving a total multiplicative effect of 379.7% upsizing.
The final image shows the mathematical difference between the two techniques. You'll note that the only points of difference are generally in the edges, as would be expected. The following is a composite of the two resulting images with the layer mode set to difference. Finally, an auto-levels greatly exaggerated the resulting differences.
Reader's Comments:
Please leave your comments or suggestions below!One typically uses a different algorithm for enlarging than they do for decreasing an image size. In enlarging an image, where one must create new pixels, one can either user:
- Nearest neighbor -- This is the simplest method of all. It simply copies pixels that were nearby and replicates them. For example, if one were increasing an image by 400%, then one inserts 3 pixels of the same color and intensity as the original pixel. Note that the one-step process doesn't generally do this because of the poor quality that results.
- Bilinear - A process based on interpolation, it tries to determine what might have lied between neighboring pixels. The simplest method of all would be to average the four surrounding pixels and insert this averaged color and intensity into the new space that is created. For example, let's consider an image that is 100 pixels wide and we want to upsize it to 200 pixels wide. Well, we'd need to insert one extra pixel between every pair of pixels. Assume pixel #32 had an intensity of 20 and pixel #33 had an intensity of 28 (ie. it was brighter). The new pixel (#32.5) would be 24 (ie. a brightness between the two). With real bilinear interpolation, four pixels are averaged (X & Y), not just two.
- Bicubic - A better refinement on Bilinear is to use the surrounding 16 pixels instead of just the surrounding 4 pixels. The average can then be based on a more accurate slope of intensity and color (cubic function).
You'll find that bicubic interpolation is the most common method (for best results), and Photoshop allows one to further define how it works by adding in smoothing or sharpening.Either way, the layer study is an artistic composition.
You're right! It is quite the work of art!