Sensitivity of ALIAS to Small Variations in the Dimension of Fractal Images

Based on collective learning systems theory, a Transputer-based parallel processing imageprocessing engine, known as ALIAS (Adaptive Learning Image Analysis System) has been applied to a difficult image processing problem: the detection of anomalies in otherwise normal images. This system was designed and developed at the Research Institute for Applied Knowledge Processing (FAW) in Ulm, Germany, in cooperation with Robert Bosch GmbH. To test its ability to detect small differences in the complexity of similar images, ALIAS was trained on a set of non-deterministic self-affine fractal images of dimension 2.10, and then tested with five unique sets of fractal images of dimension 2.12, 2.14, 2.16, 2.18, and 2.20. Formal experimental results reveal that ALIAS easily detects the difference between control images of fractal dimension 2.10 and test images of fractal dimension greater than 2.16. Informal observations suggest that this difference cannot be easily detected by the human eye. Overview of Collective Learning Systems Theory Based on seminal work in learning automata theory by Lakshimivarahan and Narendra [1, 2], the collective learning paradigm was first proposed by Bock in 1976 [3]. In this paradigm a group of collective learning automata (CLA) embedded in an external evaluating environment is known as a collective learning system (CLS), as diagramed in Figure 1.