At some point in time, most all photographers capture an image they will want to output as a large print. As digital photographers, we sometimes desire a final print size that is larger than our base file or native-sized image will comfortably allow, so we have to somehow manufacture more image information for the file in order to print it larger. The process we use to accomplish this is called “interpolation.” Interpolation is also referred to as a “resample,” “upsize,” “upscale” or “uprez.”

Digital cameras (and scanners) capture images at some maximum “native” level of resolution. This means there are a fixed number of pixels available in an original file to be used in reproducing a print. Since we need some minimum number of pixels in every inch of a print to render detail, the maximum output size we can obtain from any given native file is limited by the total number of pixels in that original file. Creating the additional information necessary to accomplish is where the interpolation process comes into play. The good news is there are several methods of accomplishing this task; the bad news is there is little agreement as to which is the best method…

Whenever the topic of interpolating digital files for the purpose of large print output is mentioned, you usually hear arguments break out over which method is the “best” one. The most popular methods of interpolation mentioned are usually some form of stepped bi-cubic interpolation in imaging programs such as Photoshop, or the use of programs specifically dedicated to accomplishing this task like the Genuine Fractals interpolation product or Q-Image’s printing program. While most of these traditional methods work fairly well, they all have their shortcomings. Over time I have tested virtually every method available, and generally found myself preferring a carefully run, stepped Bicubic interpolation routine, with selective sharpening at various intervals between the upsizing steps. This is a time-intensive process and requires above-average Photoshop skills to achieve a respectable final result, yet it often runs faster than the other options. However, with the introduction of Photoshop CS, it became possible to streamline this process even further. I feel the method developed and shared below generates a more intricately detailed image than other methods do, it runs faster than most of the other options, it is very easy to use, and best of all a method that does not cost you anything if you already own the latest version of Photoshop, Photoshop CS!

So what is different in CS that allows it to accomplish this task so well? Really, only two things:

1) CS allows us to do more work in 16-bit mode than earlier versions did, and

2) it gave us two new interpolation routines, Bicubic-Smoother and Bicubic-Sharper.

In previous versions of Photoshop, we usually had to convert our image to 8-bit mode fairly early in the workflow, long before we got to the end of processing our base image. This left us with an 8-bit file to interpolate. With CS, we can pretty much stay in 16-bit mode to the very end of the base processing, at least most of the time, meaning we will have a full 16-bit file to interpolate. More color data in the file means less color rounding errors during the uprez, which in turn means a smoother-toned final interpolation. However as beneficial as 16-bit is, it is a relatively minor improvement for interpolation purposes if the original file has been processed carefully, and fortunately it is not a necessary step to achieve superior interpolation results.

By contrast, the new interpolation routines are the real improvements that make this new process possible. With normal Bicubic interpolations in earlier version of Photoshop, if we upscaled to our target image size in one giant step, we would often get interpolation artifacts, which usually presented themselves as jagged edges or blotches in the final image. Using several smaller steps would help alleviate these artifacts. The new Bicubic-Smoother option in CS has been designed to avoid that tendency, and moreover you can now apply it in one giant step to get your image to its new target size. Similarly, during a downscale using one giant step with normal Bicubic usually left you with a very soft image. We again used steps, and sharpened in-between specific incremental downrezzes to avoid losing the original sharpness we worked so heard to obtain. Bicubic-Sharper also overcomes this problem and allows the original image sharpness to be maintained during the downscale, even if it is done in one giant step. In fact, if you step downsize using steps with Bicubic-Sharper, you can end up with a grossly over-sharpened image when you get to your desired final size – so it is advisable to use it only in one step.

Before we get into the process itself, I should mention my experience hovers mainly around upscaling Canon digital camera files. I have only had limited experience working with Nikon NEF raw files, and it was not all that positive. About the best I could do was get to 16x24 before the images just fell apart. By contrast, I have successfully uprezzed 4MP 1D images to 24x36 and they looked very good. I assume that given more time with the NEF, I could tweak my procedure so they uprezzed better, but since I shoot primarily with Canon digital cameras I have not devoted any time to doing so.

First, I generally use Bibble or ACR to convert my raw images, but whatever conversion software you prefer is fine. During the conversion step I do apply some initial, slight sharpening in the converter. Sharpening affects local contrast, which in turn can affect overall contrast, so I don’t over do it – and I also tend to keep my contrast setting on the low side during raw conversion. This allows for maximum latitude in the converted image for later processing in Photoshop. There I will adjust contrast towards the end of the process via curves. The most important thing during conversion is not to have so much initial sharpening that you impart halos in the converted image – halos will be disastrous in the final interpolated image, as they will create serious detrimental artifacts during the interpolation process.

Next I get my image fully "print ready," including sharpening, at its NATIVE RESOLUTION. Sharpening at this stage is contrary to what most folks do, but I believe they are stuck on their workflow based on past experiences with older interpolation routines. I used to sharpen after the interpolation as well, but with this new procedure I have done the comparisons for myself and found that sharpening at this stage of interpolation generates a better final print – and generating the best final print is my ultimate goal. My image adjustments include in this approximate order, PTLens for optical corrections, noise reduction if needed, cropping to desired proportions and composition without rescaling*, cleaning/cloning, any local adjustments, curves, color and finally, sharpening**. I sharpen to the same level I normally would to print the image so I now have my image fully print-ready at whatever its native file resolution is.

Rescaling Note: When you use the crop tool, set it to your desired proportions only, and be sure to leave the “resolution” box empty. This will perform your crop at the native resolution without any resample. Depending on your camera, and assuming you maybe cropped your original file to 16x24 proportions, the native resolution may now only be 100 PPI – not enough for a print. That's okay, it's at your final desired proportion and we’ll adjust the resolution later during the upscale.

Sharpening Note: For the above sharpening step I use EasyS, a sharpening plug-in available from this site. I find that EasyS does a superb job of sharpening and does NOT impart halos, though Photoshop’s USM is adequate if you know how to use it properly and apply it carefully.

Now I uprez the image in CS via “Bicubic Smoother” to roughly 20% OVER what I intend for my final print size. I perform this uprez in one large step, not smaller incremental steps. We can use the measured dimension or the pixel dimension boxes to scale the uprez in the Image Size dialog box. I also set my desired image resolution at this time. So, if my final desired print size is 16”x24” at 300PPI resolution, I set the resolution box in Image>Image Size to 300PPI. I then set the image size to 20% over my desired target. Since I desire 16x24 as my final print size in this example, I would up the width in this step from 24” to 29”. (Note: Make sure "Constrain Proportions" is checked when you do this – this may mean that when you set the 29” dimension in the width box, height may change to 19.1 or 19.5 inches. That’s okay, it’s what it’s supposed to be so don't try to reset it.) Finally, make sure "Resample" is checked and the method selected is "Bicubic Smoother". Click okay.

After my computer has finished this resample, I sharpen again, paying particular attention to edges. I sharpen them to the point where they clean up visibly, but as always, I avoid adding any halos. To sharpen at this step I use USM on the “Luminosity” blend mode and usually fade the result to 80% opacity before merging. (You can accomplish this without reverting to layers in Photoshop by going to Edit>Fade USM immediately after applying the USM.) Note that on a large uprez, the amounts of USM required may be much stronger than you are used to applying for normal sharpening. Also note that the image may not look very good at "actual pixel" view at this point, so it is a good idea to go to View>Print size to preview your result – it should look pretty good here and will get significantly better in the following step.

Now I am ready to downsample to my final desired print size. I do the same thing here I did before, but this time I set the width to my desired final size (24” in our current example), keeping resolution at the same setting used in the uprez (300PPI in this example). However, this time I resample via “Bicubic Sharper”, and again do this in one step.

The combination of interpolating up to 20% beyond what my final desired size is, sharpening, then downsizing to my actual final image size is a simplified form of fractal sharpening. I have found this step very effective in generating a highly-detailed, but smooth-toned final print.

Once this last downrez is complete, I am done – and I should have a 16x24 at 300PPI that looks very, very good when printed.

Example Interpolation

Here is an image from my 1DMKII. I will crop a portion to show it enlarged for the different steps in the process. Relevant gear/exposure data for those interested: camera was the 1DMKII, lens was the Canon 200/1.8 with 2xII converter attached for an effective 400mm focal length; ISO 250; exposure was 1/2500th @ f4; raw capture mode; AWB. Note that I did all of the processing on a 16-bit tiff file in Adobe RGB color space, then after the processing, the final images were converted to the web jpegs you see here.

1) Here is the full frame original image, with the 500x500 pixel portion we are going to interpolate outlined in red:

2) Here is the actual 500x500 pixel crop we are going to uprez, shown at “actual pixel” resolution, fully post-processed and sharpened with EasyS, but still at its native resolution:

3) Here is a screenshot for the 16x uprez that follows. To increase the area of the image by 16x, we increase each dimension by 400% or 4x. Since the original file is 500x500 pixels, and since we are interpolating it up to 20% over our desired target size, we set the target size to 2400x2400 pixels. (Here is the math: 4x500=2000, 2000+20% = 2400):

4) The file is now sharpened with USM targeting the edges, and then faded to 80% in the luminosity mode – this is the fractal sharpening step.

5) Now it’s downrezzed to our final desired 2000x2000 image size:

6) Here is the final image, uprezzed 16x (400% in each dimension), processed as described above, but now viewed at “print size” resolution. Please note that what you are now looking at on your screen represents an approximate 61/2”x61/2” section from the final print printed at 300PPI resolution, and would translate to a final print sized 32x48 inches if we were viewing the entire image (!):

(You may detect what appears to be a slight color shift in the above image relative to the original and 4x uprezzed versions. This is due to saving a screen capture of the working tiff as a jpeg directly and not from the interpolation process itself.)

7) Here the crop has been uprezzed 4x (200% in each dimension) with roughly the center half cropped out to fit on the web page, processed as described above. It is being viewed at “actual pixel” resolution so you can see an actual pixel representation after a typical uprez. The full image after this interpolation would print out to almost 16x24 at 300PPI:

Addendum: Initial uprezzing during raw conversion.

Many times you will see a recommendation to do a first uprez during the raw conversion process. With the process described above, I made it clear I process to the camera’s native file size. First, this keeps my files smaller during post-processing which in turn means the operations I perform in post run faster. Second, while uprezzing during conversion will give you a larger initial image to work with, it can also impart undesirable artifacts in certain areas of the image. I have tested it with ACR on my Canon files and can detect virtually no difference between doing it there and doing it later in the main CS program in MOST images. However, I have found certain situations where it generates additional undesirable artifacts, hence the reason I do not use it myself as a general practice. These artifacts will grow worse if you plan to make the image even larger later in the process using any interpolation routine.

Regardless, you may want to experiment with your own raw files. You can usually get to about 200% of native size during raw conversion. Results may vary by file as well as equipment, so it's worth testing for your camera and image processing combinations, especially if you only need a slight upscale for your image.

Here is a sample of a “bad” artifact obtained while doing a 200% uprez in ACR on a 1Ds file. First the full-frame image, downsized for web. This image was taken with my 1Ds and 400DO lens, 1/800th @ f4.5, ISO 100:

And here are miniscule crops from that image, showing a specular highlight in the pheasant’s eyeball. The first was converted to 2x (142% in each dimension) during raw conversion in ACR, and then enlarged via Bicubic-smoother for critical web view. The second was converted to native size in ACR, and then similarly enlarged via Bicubic-smoother to match the first image’s scale. The highlight artifact is easily visible in the image that was interpolated during raw conversion in ACR: