Monday, March 23, 2009

Artifacts of modern information society

Artifact, i.e. an error or misrepresentation introduced by a technique and/or technology. Most obvious are artifacts in lossy image compression techniques such as, e.g. the jpeg compression algorithm. For jpeg, the entire image is divided into 8x8 pixel squares (of course for all three color channels... but not RGB. For jpeg the RGB picture is first transposed into the YCrCb color space, i.e. Y for luminance and Cr, Cb for chrominance being subdivided into red and blue. BTW, there is also subsampling, i.e. luminance is sampled with higher accuracy than chrominance w.r.t the human sensory perceiption). Then these 8x8 squares of intensity values are transformed into the frequency domain via a discrete cosine transformation. Up to now, no artefacts, no data loss.....

But, next comes a quantification algorithm that rounds the frequency values within the 8x8 squares. If this is done with high accuracy, often one doesn't even notice it within the picture. But, increasing quantification also results in higher data compression, which also results in more data loss, which creates ... artifacts. You can see the typical jpeg raster effect, whenever using high jpeg compression.Here's a cool animation of an image being jpeg compressed 600 times (in a 20 second short movie). The simple algorithm goes:
Open the last saved jpeg image
Save it as a new jpeg image with slightly more compression
Repeat 600 times.

Generation Loss from hadto on Vimeo.