Pictorial analysis: a multi-resolution data visualization approach for monitoring and diagnosis of complex systems

This paper aims to describe the prospective methodology (Pictorial Analysis) of a human-computer interaction capable to select and adjust visual representations for better feature and pattern selection. As an adequate tool for on-line monitoring and diagnosis of very complex systems which require human supervision in addition to the computerized one, our approach exploits human capabilities of pattern recognition in doing data analysis and selecting appropriate representations, features and class descriptions by means of interactive human-computer learning . It is based on the analysis of multidimensional relations through transforming the initial data (heterogeneous arrays, signals and fields) into artificial pictures . Practical applications of this approach prove it extremely useful in critical areas of safety, such as flight control, power plant monitoring, etc.