INTERPOLATION REVISITED and Photography
Back in 2005, I wrote a fairly extensive article series on interpolation. The series looked at several different methods of interpolation. The conclusion of the series was that, at 200% linear interpolation, there was not much difference between the interpolation methods. At larger interpolations, the best interpolation method often depended on the image.
A lot has happened with digital photography since that time. Camera sensors have increased in size and quality. Photo editing software has substantially improved. Printer resolution and gamuts have increased. Of course, interpolation methods and software have improved as well.
I began to wonder if the results of the testing that I had conducted in the previous interpolation article series were still valid. In short, it was time to do some new testing. Consequently, the purpose of this article is to review current interpolation methods by presenting the results of my testing.
This article is more constrained than the earlier article series for two reasons. First, the theoretical background for interpolation has already been covered in the prior article series. If you wish to understand the theory, please read the previous series, Interpolation. Second, the prior article series looked at 200%, 300%, and 400% linear interpolations. This article evaluates only 200% linear interpolations. The reason for this is that sensor sizes have increased to the point where it is unnecessary for most photographers to interpolate much more than 200% linearly. An image from an eight megapixel camera interpolated linearly 200% and printed at 300 PPI will create approximately a 16" x 24" inch print. Similarly, an image from a twenty-one megapixel camera will produce a 24" x 36" print. Most photographers do not need to print larger than this.
I began to wonder if the results of the testing that I had conducted in the previous interpolation article series were still valid. In short, it was time to do some new testing. Consequently, the purpose of this article is to review current interpolation methods by presenting the results of my testing.
This article is more constrained than the earlier article series for two reasons. First, the theoretical background for interpolation has already been covered in the prior article series. If you wish to understand the theory, please read the previous series, Interpolation. Second, the prior article series looked at 200%, 300%, and 400% linear interpolations. This article evaluates only 200% linear interpolations. The reason for this is that sensor sizes have increased to the point where it is unnecessary for most photographers to interpolate much more than 200% linearly. An image from an eight megapixel camera interpolated linearly 200% and printed at 300 PPI will create approximately a 16" x 24" inch print. Similarly, an image from a twenty-one megapixel camera will produce a 24" x 36" print. Most photographers do not need to print larger than this.
INTERPOLATION METHODS EVALUATED
Six interpolation methods were evaluated:- Bicubic
- Bicubic Smoother
- Stairstep
- Genuine Fractals
- Blow Up
- PhotoZoom
INTERPOLATION RESULTS
All of the interpolation methods were applied to three different images. Since the results were similar for each image, the results for only one image are presented here. Figure 1 shows the image that was used for the interpolations show here. Figures 2 -- 7 show a section from the image, at 100% view on the monitor, after it was interpolated and a moderate amount of sharpening was applied (the last image has a watermark since I used a trial version of PhotoZoom).Figure 1: Rattlesnake Image
Figure 2: Bicubic
Figure 3: Bicubic Smoother
Figure 4: Stairstep
Figure 5: Genuine Fractals
Figure 6: Blow Up
Figure 7: PhotoZoom
CONCLUSION
If you have been staring at these images and thinking that you don't see much of a difference, you have already reached the conclusion of this evaluation. At 200% linear interpolation, with the exception of Bicubic which tended to be of lower quality, I did not see much of a difference between these various interpolation methods.Now, please keep in mind that these images have lost some of their quality when they were prepared for the web. This reduces the differences between the images somewhat. The original, interpolated images do show small differences between the interpolation methods. However, in my opinion, the differences are minor (with the exception of Bicubic as previously mentioned), and it is often debatable as to which one looks the best. Furthermore, these differences are unlikely to be noticeable in a print. Basically, my conclusion is that, depending on the image, Stairstep or Bicubic Smoother interpolation did as well as the interpolation software packages at 200% linear interpolation. Notice that I indicated that this depended on the image. For some images, Stairstep did better than Bicubic Smoother. For other images, it was the other way around.
Now, does this mean that one should not purchase one of the interpolation software packages? Actually, no. These conclusions only hold at 200% linear interpolation. For larger interpolations, Genuine Fractals, Blow Up, and PhotoZoom (as well as other interpolation packages) may very well produce superior results. Furthermore, these packages have additional functionality beyond interpolation. If they meet your needs, they may be a good idea.