Chapter 8. Countering artifacts

Artifact: Technology an object, oservation, phenomenon, or result arising from hidden or unexpected causes extraneous to the subject of a study, and therefore spurious and having potential to lead one to an erroneous conclusion, or to invalidate the study. In experimental science, artifacts may arise due to inadvertant contamination of equipment, faulty experimental design or faulty analysis, or unexpected effects of agencies not known to affect the system under study.

--The Collaborative International Dictionary of English v.0.48 [gcide]

Static sensor noise

Some sensors have permanently stuck pixels, or pixels that have a different sensitivity than others. This is a problem that can be fixed by having a "blank" exposure that records the difference between pixels. The pattern found by taking this blank shot can be used to correct both the intensity of recorded values, as well as replacing known to be bad pixels. This is usually called "dark frame" subtraction. This is mostly applicable to very well controlled exposures with stable sensors. Some forms of scientific imaging like astro-photography with sensors that are kept at a constant temperature can make use of dark frame removal.

Some digital cameras keep a track of bad pixels themselves, when getting a RAW file, the bad pixels are usually not retouched.

Film grain

Analog film effectivly has many small light sensitive particles that act as sensors. With fast films, that need little light, the density of these particles is low, which lead to a grainy texture on the images.

Film grain is countered by various lowpass filters, like blurring and median. Other more advanced filters reconstructs vector fields of "flow" in the image.

Note that sometimes the goal is to create noise to match original film grain, this can be achieved by analysis of blank frames of video to extract patterns with statistical properties similar to the film-grain.

CCD sampling noise

CCD's have the same problems as physical film except the noise appears to be even more random. Various noise reduction techniques like bilateral filtering that preserves larges changes while smoothing out small variations.


The quantification process can introducate artifacts. Increase in contrast and colorspace conversions can amplify quantification artifacts. The old adage, GIGO, Garbage In, Garbage Out. there isn't very much to do with data that is bad from the outset.


Old analog TV equipment did a neat trick to allow for both a high temporal rate as well as a high number of lines. Only every second line were transmitted, first the odd lines, then the even lines. This way of displaying images only work well for CRT monitors. For LCDs an interlaced image must be resampled to avoid combing artifacts.

Line doubling

Line doubling is the simplest form of deinterlacing, it replaces the odd lines with the even lines. This gets rid of the interlacing effect, but reduces the vertical resolution in half.

Linear blend

"Linear Blend operates by taking a line, then averaging the pixel values in it with the line below, effectively blurring the frame. This almost completely eliminates the effects of interlacing - at times you may notice slight ghosting (instant transition from one thing to another), the image appears to persist for a fraction of a second. This is absolutely not a serious issue, as far as I am concerned - in fact it's really only noticeable if you are me, and a perfectionist." [quote from]

Temporal resampling

By reconstructing motion vectors for the video frames, it is possible to predict the position of object between frames, thus providing a volume video object where any position can be sampled. Such an advanced approach can be used for both deinterlacing and high quality time stretching.


Aliasing artifacts, result from a spatial sampling frequency that is too low. A common example is moire effect showing up on textiles, like tweed, with detailed repeating patterns like tweed.

One way to counter aliasing is blurring the image. If we have control over the creation of the problematic image, we can also render a high resolution version that is downsampled to the target resolution using decimation.

Color casts

Color casts occur when a the white balance isn't correctly calibrated, usually white balance compensation is done on the raw, higher bitdepth, data from the CCD, to avoid quantization loss in the progress. For more detail on color casts see the previous chapter on automatic algorithms.