Seminár Ústavu informatiky



In the talk we discuss a methodology of image processing and a recognition of shape of graphic objects. At the beginning we briefly introduce an approach for preliminary image processing that allows to obtain characteristic attributes of the studied objects and utilise them for the determination of some structures of objects. Further the philosophy of a recognition of the shape of objects will be presented. In particular we describe the group of methods (18 items) that we use for an identification of the shape of objects. These methods are based on so-called geometrical correlation which utilizes the signature analysis. This approach allows to distinguish the shape of objects invariant to affine transformations with any beforehand set accuracy. For a determination of quality of the offered methods the technique of an estimation of quality of methods of identification of the shape of objects will be stated. Moreover, some methods based on the geometrical correlation without use of the signature analysis will be presented. In one of them we will analyse a metrics allowing to identify the shape of objects invariant to projective transformations (!).  Finally, we will demonstrate methods for increasing quality of characteristic attributes for pattern recognition.