Recently, I have done quite a bit of Core Graphics programming and discovered that one key point in ensuring that an app performs well is choosing the right CGContext interpolation level. Interpolation, according to the Wikipedia, “is a method of constructing new data points within the range of a discrete set of known data points.” Interpolating is what is done when you scale up an image, but also when you scale it down. In both cases, interpolation has a tremendous impact on the result of the scaling.
Core Graphics CGContext allows 4 different levels of interpolation: kCGInterpolationLow, kCGInterpolationMedium, kCGInterpolationHigh, and kCGInterpolationNone. Their effect is pretty trivial to describe: kCGInterpolationNone will give you the most jagged result of all; kCGInterpolationHigh will give you the smoothest result of all. What is less clear from the outside is which impact interpolation will have on your app.
So, I put up a benchmark test and here are the results.
In my benchmark test, I draw a set of 6 different 649×649 b&w bitmap images to build a set of 12 different jigsaw puzzle tiles. The bitmap images are scaled down by a factor of 5.0 (i.e., to 129.8×129.8) to make them fit on a 320×480 display (at retina display scale).
Run times were measured using Instruments and only considering the time spent inside of the CGContextDrawImage function, so to remove all effects not related to the interpolation itself.
The device used was a 4th gen iPod touch.
– kCGInterpolationLow: 969 msec
– kCGInterpolationMedium: 1690 msec
– kCGInterpolationHigh: 2694 msec
– kCGInterpolationNone: 545 msec
As you can see, there is a big factor between no interpolation and high interpolation (almost 500% increase). The question arises whether this time is spent for some good sake or not, so let’s take into account the visual result.
As to the visual outcome, if you compare High or Medium to Low or None, the difference is staggering. Even noticeable, IMO, the difference between None and Low; while the difference between High and Medium is not particularly noticeable in this particular test.
What is clear is the difference in run times, so that in my particular case, I ended up using `kCGInterpolationMedium`.