bands, in this case four bands. 
All the following graphics from the next figures were 
printed using electrostatic protter. Figure 3 shows the 
total gradient of 30. 
8 —®— 5 ^©—4 ^©^ 3—©— 2 
t ® ,®'P© ©'l*© ©'p© ®'\ 
©X®X©V@V® 
Fig . 1 Values of pixels, the smallest number are the relative 
gradients using eight directions. 
1 ■ 3 Continuous representation of the relative gra d i e n t . 
The relative gradient can be represented in a continuous 
form, using a small square of four resolution elements of 
2 by 2 like a mask. It is interesting for analysis that the 
shape classifier just uses two lines or two records in all the 
proc essing. 
Using this method of continuous representation, it is 
possible to represent the relative gradient as a continuous 
form. Figure 4 shows relative gradients of three directions 
in a continuous form. 
2. SHAPE ANALYSIS 
2.1. Introduction 
The study ot shapes is an important part of image processing. 
This section of this work gives a procedure to measure (i.e. 
to assign a number to) the resemblance between any two 
shapes (Bribiesca and Guzman 1980). 
With the help of procedures like this, a quantitative 
study of shape may be possible (Bribiesca 1981). 
2.1 1 . - Previous work on shape 
Shape extraction Ts and active Field. Sequencial extraction 
of shape features (Agrawala and Kulkarni 1977) can be 
performed making only one pass over the image. For global 
shape analysis, several authors have used Freeman Chains, 
medial axis transforms, decomposition into primary convex 
subsets, polar co-ordinates (Perkins 1978) decomposition at 
concave vertices; decomposition by clustering, mirroring 
axes and stroke detectors. These and other methods are 
reviewed by Pavlidis (Pavlidis 1 978).