Strengths and Weaknesses of the Image Understanding Technology Compared to Previously Known Approaches

Abstract

The techniques of automatic image understanding, presented in this book, utilising the linguistic approach have several major advantages over classical image recognition algorithms. It is readily apparent that for many types of images, in particular medical images, it is difficult to interpret and define the representative vector of numerical features required in the classical approach applied in theoretical decision-based methods. This means that a certain type of images containing structural information can be extremely difficult or even impossible to classify on the basis of selected features represented in numerical form. This is so because the structures ought to be described in such a manner that some relationships and constituent elements of the structure are first defined while the structure itself can be described in general terms with a use of a model or strictly specified. The presence of semantic information requires therefore that analysis be made, both of the classification and description (meaning) sense. The classification task is based chiefly on operations of seeking similarities (usually referred to as the grammar derivation path); yet this operation has some generalisation properties and together with semantic information obtained in the course of analysis, it allows us to recognise a practically unlimited number of classes and objects.