An image spatial accessing methodology based on a formal language (SCAN) is
presented in this paper. SCAN is a special purpose context-free language d
evoted to describe and generate a wide range of 2-D array accessing algorit
hms from a short set of simple ones. These algorithms may represent sequent
ial scanning techniques used for image processing, such as generation of im
age data structures (pyramids, trees), encryption, compression, of images,
etc., but at the same time they stand as generic spatial data accessing str
ategies. The SCAN language provides a method of composition of 2-D accessin
g patterns. The method is motivated by the principle of recursive decomposi
tion of an image array into hierarchical levels for efficient local and glo
bal processing. The words of the SCAN language are simple linear forms whic
h convey information for both type of decomposition to be applied, and the
specific accessing patterns to be composed.
In this paper, we provide the formal definition of the SCAN language and de
scribe the underlying method for spatial access. Properties of the accessin
g patterns generated by the language are also investigated, and the underly
ing mathematical model is discussed. Finally, a scheme for the parallel imp
lementation of the SCAN language is presented. (C) 1999 Pattern Recognition
Society. Published by Elsevier Science Ltd. All rights reserved.