Cognitive Techniques in Visual Data Interpretation

The extremely rapid progress of science dealing with the design of new computer systems and the development of intelligent algorithmic solutions for solving complex problems has become apparent also in the field of computational intelligence and cognitive informatics methods. The progress of these new branches of informatics has only started a few years ago, but they are already making a very significant contribution to the development of modern technologies, and also forming the foundations for future research on building an artificial brain and systems imitating human thought processes. We are already able to build robots with basic machine intelligence, which can sometimes perform complex actions and also operate by adapting to changing conditions of their surroundings. This very impressive development of intelligent systems is manifested in the creation of robotic devices which use artificial intelligence algorithms in their operations, movements, when solving difficult problems or communicating with humans. It is also evidenced by the introduction of new methods of reasoning about and interpreting objects or events surrounding the system. One of the fields in which the need to deploy such modern solutions is obvious are cognitive vision systems used both in mobile robots and in computer systems which recognise or interpret the meaning of recorded signals or patterns. Years ago, such applications inspired the authors of this book to undertake research on the design of computer systems, later called cognitive, which would allow the semantic meaning of registered patterns to be determined. This subject turned out to be extremely interesting, scientifically inspiring and appealing to different groups of scientists because, as became apparent later, it contributes to broad groups of varied, open scientific problems in which algorithms from this field can be used. Examples of such problems include: determining the ontology and meanings of text phrases in web searching, supporting strategic decisions (in business) and diagnostics (e.g. in medical practice), or modelling the operation of brain structures which in the future will allow the secrets of the human psyche to be revealed and might also allow computer models of the brain to be designed.

This chapter briefly describes aspects of the traditional approach to pattern recognition and classification. Among the number of known methods developed over the years, we can distinguish structural analysis and pattern classification algorithms which the authors have used in their research work for many years to study selected classes of medical images [69]. This chapter will provide a brief summary of key information about traditional approaches to image analysis, and present the methods, developed in recent years, of semantically classifying selected patterns [69]. Structuring information about these subjects will help us, in the following chapters, to raise the possibility of using cognitive and associational techniques for a deeper, meaning-based classification of images.

Since time immemorial man has been fascinated by the structure of his mind and has striven to learn the secrets of its operation. The human image of the world, all that we believe in and perceive at a given moment, the feeling of happiness or its lack, are the result of the operation of the nervous system, and in particular the brain in which our mind functions.

Cognitive informatics (CI) is a discipline which combines the subjects of both the cognitive science and informatics based on information mechanisms and processes taking place in the human brain. So cognitive informatics uses natural intelligence merged with computer applications broadly applied in interdisciplinary research and science. Cognitive informatics thus covers not only the aspects of using mathematical theories to describe and analyse data and information presented in large knowledge bases but also engineering fields with informatics, cognitive sciences, neuropsychology, system sciences, cybernetics, computer engineering, knowledge engineering and computational engineering.

It became possible to develop cognitive information systems on the foundation of intelligent information systems whose purpose was not just the simple analysis of data consisting in recording, processing and interpreting it, but primarily an analysis by understanding and reasoning about the semantic contents of the processed data (e.g. images, biometric patterns, economic data etc.).

The purpose of the subject discussed in this book, namely systems for the cognitive interpretation of image data, is to present new classes of information and vision systems called UBIAS (Understanding Based Image Analysis Systems). These systems were developed to support analysing, interpreting and reasoning about the semantic meaning of images.

The UBIAS systems described in the previous chapter, designed for analysing medical images, are developed quite rapidly as research has been conducted on analysing and cognitively categorising images of various lesions occuring within the central nervous system [41, 46].

This book presents selected aspects of the development and implementation of cognitive vision systems. The introduction of such systems is the natural consequence of the development of artificial intelligence methods and such scientific disciplines as cognitive informatics. These disciplines now make it possible to design sophisticated, intelligent systems whose job is not just to acquire or collate various categories of data, but also to facilitate their in-depth semantic interpretation. Attempts to define semantics are also frequently made for data in the form of texts. However, if we only consider image data in the form of various types of photographs or visualisations, there has been no wide scale research on their semantic interpretation yet, so one can hardly expect the limited work that has to have already yielded concrete, satisfactory results. The authors of this book are filling this scientific gap by trying to show how cognitive computer systems capable of interpreting images will develop in the near future. Obviously such considerations are presented only using selected image classes, in particular chosen types of medical diagnostic images. However, it seems that the presented approach will have much broader applications and in the future will allow other, no less important patterns to be interpreted.