Lossy compression exploiting self-similar patterns in image data — obsolete VFX method. High compression ratios caused quality degradation and reconstruction artifacts. Replaced by modern codecs.
In the 1990s, fractal compression promised a revolution for digital image processing—theoretically elegant, practically problematic. The method utilized mathematical self-similarity: an image was divided into blocks, and the algorithm searched for repeated patterns at different scales. Instead of storing pixels, it stored the mathematical description of these self-similar structures. Sounds efficient? It was—on paper.
On set or in the edit suite, it quickly became apparent: at compression rates that truly saved storage space (1:100 or higher), the image material disintegrated into characteristic block artifacts. These reconstruction artifacts were particularly disruptive with moving subjects—exactly the opposite of what VFX supervisors needed. The method worked reasonably well for static backgrounds or landscapes with repetitive structures, but failed with details, facial expressions, or fine textures. Creating a Digital Intermediate with fractally compressed material was a nightmare: every grading attempt, every keyframe effect amplified the artifacts. The intermediate process itself, in any case, demanded uncompressed or minimally compressed data—fractal compression was in direct contradiction to this.
In practice, the technology was used for archiving and hard drive optimization, but never for productive VFX pipelines. The codec also required high CPU load during decompression, which was significant on the hardware of the time. Modern methods such as ProRes or ARRIRAW-based solutions quickly took over—they offered better quality with only marginal storage disadvantages.
Today, fractal compression is practically forgotten in the film industry. It appears at most in historical discussions when discussing failed digital approaches. Anyone who occasionally encounters archived material from this era must expect that recompression or up-conversion will lead to unavoidable quality losses. The lesson learned remains: mathematical elegance does not automatically translate into productive practicality.