The Geometry of Meaning: SIOX Twenty Years Later and the Invariants of Semantic Computing

This anniversary article re-presents and reinterprets the 2007 International Journal of Semantic Computing article on SIOX, Simple Interactive Object Extraction [9]. The historical contribution was a practical foreground-extraction tool for images and videos: a user supplied a small amount of semantic supervision in the form of foreground and background examples; the system mapped pixels into a perceptually meaningful color space; compressed the examples into color signatures; classified the remaining pixels by distance to those signatures; and refined the resulting confidence matrix. Twenty years later, this pipeline reads less like an isolated image-processing trick than like a transparent miniature of a broader pattern in artificial intelligence and semantic computing: representation, compression, separation, and action. We therefore preserve much of the original article, revise the related-work discussion as a 2026 retrospective, and argue that twenty years of semantic-computing applications now call for a more explicit unifying theory.