AN EEE) OO NEE rT (i THC EN, ES CE 
arguments and data types into frames. 
Histogram Leplacian 
Common Para, ‘Private Pera. 
Iterative Mothod Robert 
"Common Para. Private Para. 
Enhancement 
Common Noise Efimi. Sobel 
ope Para {Common Para. Private Para, 
Prewitt 
Analyse Different "Private Pera. 
Sronmon Para, Krach 
Tent hte Private Para, 
ing 
De . Robinson 
E | um. 
Common x Frei & Chan 
Fera Private Para. 
Hough Trans. 
‘Common Para 
Figure 2. the hierarchy of subroutines in SPIDER 
Frame Title: ETOC1 
Frame Usage: ETOC1 (IP,TH,JRT,ISX,ISY,ITH,IRO) 
Parent: Edge Detection 
Comment: " Detecting straight lines in an image" 
/* private parameters slots */ 
Slot Name: |... JRT 
Data Type: Dimension 
Horizontal: ITH 
Vertical: IRO 
Comment: "2-dimension histogram of (p,6)" 
Slot Name: ITH 
Data Type: Integer 
Upper Range: 32768 
Lower Range: 1 
Default: 180 
If-Needed: Manual 
Auto 
Comment: "Number of quantization for 0" 
Slot Name: IRO 
Data Type: Integer 
Upper Range: sqrt(ISX+ISY) * 1.414 
Lower Range: 0 
Default: 260 
If-Needed: Manual 
Auto 
Comment: "Number of quantization for p" 
Table 1: Structure of the Hough Frame 
The frame concept , Marvin Minsky[4] , consists 
of dividing knowledge up into specified categories. Frames 
function like forms. They are often implemented as form- 
like data structures in which the information in a given 
category is hierarchical. 
Like the entries in a form, frames can have numerous 
slots or places where information can:be stored. Another 
important feature provided with frames is the fact that the 
slots can have default values or procedures. This means 
that it is not necessary to describe in detail all of the facts 
about a given object. 
A frame consists of a slot for storing general 
information about the frame itself, such as titles and 
usage. The "parent" field contains the name of the frame 
511 
that references this one. The level field indicates the level 
of hierarchy of this frame. In addition, each frame has 
slots. The specialization of slots is used to establish a 
property inheritance hierarchy among the frames, which in 
turn allows information about the parent frame to be 
inherited by its children. After a particular frame has been 
selected to represent the current situation, the primary 
process in a frame-based reasoning system is often just 
filling in the details called for by its slots, and the data type 
Will be checked. Some parameters are directly inherited. If 
there are no specifications, the default value can be used, 
or the attached If-Needed procedure can be used to 
decide. Table 1. shows thestructure of a Hough frame. 
There is no limit to the number of slots that a frame can 
have. Slots can also have support fields or attributes. 
These fields help define and describe the value of the slot. 
For example, the limit for numeric slot values is given by 
the upper and lower range values. The default value is the 
value used if no other explicit information is available. To 
each frame there are some specific slots. Default and 
inherited values are relatively inexpensive methods of 
filling in slots; they do not require powerful reasoning 
processes. Thesemethods account for a large part for the 
power of frames. When the needed information must be 
derived, attached procedures provide a means of 
specifying appropriate methods. This representation allows 
for more flexibility and greater accessibility to the 
knowledge. 
(2) The knowledge of plan generation. 
Although there is much knowledge about image 
processing, by way of image analysis strategies, for 
example to detect the edge in an image, there is an 
optimal solution in SPIDER: Sobel (differential), Kirsch 
(template matching type), Frei & Chen and Hueckel. From 
our experiment, Sobel , Kirsch, Frei & Chen may give 
basically the same type of image. Kirsch consumes more 
time, Hueckel's algorithm has the advantage of providing 
an equation for the edge-line pattern detected inside the 
area of analysis; this equation can define the location of 
the edges and lines within a subpixel resolution, which 
could be used for registration purposes for processing a 
scene. lt is better to use a line-following algorithm in 
combination with the edge-line detection program in order 
to obtain a continuous type. We incorporate the above 
knowledge as gained from the experiment into our system 
as heuristic information to guide our search. 
Many analysis strategies have been proposed to 
increase the performance of image analysis. Here we use 
such a scheme to represent the image analysis strategies. 
We describe an image processing procedure by a function. 
The result of applying procedures O to image D is denoted 
by O(D), thus the sequential composition of procedures 
can be described as 
O,(0,..(...O,(O,(D)))) 
where D denotes an input image and 
01,02,...0n-1,0n functions represent image processing 
procedures. These functions are successively applied in 
this order: the innermost function is applied first to produce 
the data for the second function and so on. We omit 
arguments of the functions representing parameters of 
procedures.