Page(ID:7346/pag003)Pattern recognition definition language proposed by Page at IEEE PR conference in 1975, based on the work of Nilsson References: INTRODUCTION This paper deals with a framework for a theory of what should be saved after access to a large data base and subsequent computation and what should be redone. As Sammet has pointed out, this problem goes back to the days when the classical decision was made to compute rather than store trigonometric functions [ 13 ] . The problem area which the issue of whether to save a computation for future reference or to do it over if needed again is of course most important if the computation is very expensive and likely to be needed again, perhaps by another user. Although there are many types of large data bases to which the ideas to be presented will apply, in this chapter we will emphasize for purposes of concreteness, data bases which possess the following properties: (1) A large volume of geographically organized data. (2) A large group of users, generally unsophisticated in the use of computers and scattered around the country and the world. (3) Many different reasons for the users to access the data which prevents the development of a few standard application programs which serve all users. An example of such a data base might be remote sensing data accumulated from earth satellites and used by the agencies of states in one region for such purposes as land~use, highway planning, etc. We will be primarily concerned with the strategy of repeated data access rather than the tactics of how individual files are organized. Although file organization is important, we will be concerned with delineating an environment of data access in which an automatic data manager can exist and gather information which can lead to improved system performance at a higher level. Extract: Properties of a Pattern Definition Language Properties of a Pattern Definition Language In the last section we have seen the desirable side effect of providing a higher level language for pattern access definition. Our main purpose in this paper is to study some proper ties of preprocessors of such data. The preprocessor and pat tern language are intimately connected. However specification of a detailed pattern definition language for our concrete example of geographically organized data is outside the scope c this paper. Hence we will describe general properties of such languages which seem sufficient to allow the design of preprocessors. The basic assumptions which will be made concern the structure of the question or pattern definition language are as follows: (1) Pattern searches are defined from predefined patterns us system provided relations and functions in a syntactic manner (2) Top down pattern recognition seems more appropriate than bottom up because noise can be dealt with more easily. (It i easier to see a zebra in the shadows if you know you are looking for one than to group the shadows (bottom up) to discern the form of a zebra.) The pattern definitions control a top down algorithm for pattern recognition. (3) Patterns are two dimensional and occupy a region of bounded size. (4) Gross simplification or summaries of the original data (with geographic relationships preserved) will suffice for many pattern searches. Such summary data can restrict the extent of search if organized in a hierarchical manner and selected according to user needs. Now that some general properties of language for definition of patterns which facilitate the design of a pattern file preprocessor have been considered, let us present an example pattern language. Extract: An Example of a Pattern Definition Language An Example of a Pattern Definition Language Let us proceed with a detailed example to provide the flavor of what such a pattern definition language might be like. The example is supposed to be suggestive but in no way claims to be practical in any sense. Suppose a community of users includes two individuals, one of whom is interested in finding plan sites and the other who is interested in finding camp sites. Both define their requirements in a pattern definition language which is represented graphically in Figure 1. The notation is based on the AND/OR trees found in Nilsson [ 12 ] . A slight generalization is the representation of arbitrary relations holding between the edges such as "next to" and "near". (The actual meaning of these relations could depend on the user and be determined by the preprocessor in dialog with the user.) In Figure 1, for instance, we note that the class of "dry woods" is to be "near" "shoreline" and "next to" "transportation". Likewise a "shoreline" here consists only of either a "river shore" or a "lakeshore" and so on. The pattern definition language need not be graphical in its presentation but could consist of context free productions which describe the branches of the tree. For example, such a production is Representation of the elements of the necessary classes such as Definition 1: Any pattern class defined in the pattern definition language or pattern class combined with features available at lower levels in its definition will be called a data summary. i The task of preprocessing is to construct hierarchical data summaries relevant to user needs. Definition 2: A pattern class will be called a subpattern or subcomputation of any node which is higher in some pattern definition tree. Hence a data summary consists of a pattern and some of its subpatterns, depending on our viewpoint, a pattern will be identified with either its root node or the whole subtree which defines it. in Klinger, A.; Fu, K. S.; Kunii, T. L. "Data Structures, Computer Graphics, and Pattern Recognition" (Largely based on IEEE Computer Society conference held in Los Angeles, May 1975) Academic Press, NY 1977 view details |