Nowe klasy inteligentnych systemów interpretacji danych obrazowych. Systemy E-UBIAS

Systemy UBIAS, opisywane we wcześniejszych pracach, zapoczątkowały powstanie nowej generacji inteligentnych systemów rozumienia danych obrazowych, w szczególności medycznych. Systemy te są również przydatne przy operowaniu w multimedialnych bazach danych ilekroć wykonywane operacje (wyszukiwania, grupowania, raportowania) muszą być oparte na merytorycznym sensie zawartości obrazu, a nie na formie informacji obrazowej. Systemy E-UBIAS przedstawiane w niniejszej pracy stanowią nową kategorię kognitywnych systemów przystosowanych do semantycznej interpretacji obrazu. Innowacją wprowadzoną w systemach E-UBIAS i poszerzającą możliwości wcześniejszych systemów UBIAS jest wprowadzenie nowego elementu zdobywania wiedzy podczas normalnej eksploatacji systemu (czyli swoistego procesu uczenia). Wskazano na możliwości zastosowania tych systemów, a także na nowe możliwości ich wykorzystania do pozyskiwania danych obrazowych traktowanych jako charakterystyki biometryczne.

In previous papers given by authors the special class of computer vision systems was introduced and discussed. Such system named UBIAS was dedicated to cognitive analysis of images - especially medical ones. Replacing in UBIAS systems the image analysis and pattern recognition processes by cognitive resonance and automatic understanding we in fact introduce new class of vision systems which possibilities and advantages are not fully recognized until yet. Nevertheless the UBIAS image understanding systems have limitations related to "hand made" knowledge acquisition and representation process, which is necessary before starting the UBIAS system functioning. Sometimes collection of rules given by expert is too narrow, sometimes presentation of such rules in linguistic form used in UBIAS system is difficult, sometimes there are also another sources of problem - result is the same: The efficiency and quality of understanding of the images under consideration is not satisfactory. In all cases, when automatic understanding of the image using UBIAS methods is not good enough the solution can be E-UBIAS (Extended Understand-ing-Based Image Analysis System), which is presented in this article. The E-UBIAS systems set a new category of intelligent vision systems, which extend the possibilities of UBIAS towards acquiring knowledge and learning processes. In E-UBIAS systems two types of knowledge are used as a support of cognitive resonance procedures, leading to automatic understanding of the image semantic content. First type knowledge is acquired from experts ad is represented in linguistic form in graph-grammar structures used in typical cognitive resonance process. This is deductive part of the system. Second type knowledge is registered during normal work of the system. Every reasoning process performed by the system for some particular image give us new piece of information, which can be concatenated with other similar pieces coming from previous experiences. Such part of system is inductive one. Moreover systems belonging to the E-UBIAS class can be also used for collection of biometric characteristics for every person, which image was analyzed by the system and which semantic description was processed for automatic understanding purposes. It can be very useful in many security applications.