Scanline intersections and weights

 

Figure 2.7: A coding of two characters by measuring scan line intersections

Some thought on how to represent the characters in a way which will not be too badly affected by such things as translations, the deck transforms that are produced by italicisation and small angle rotations, may lead to taking horizontal and vertical scan lines across the rectangle, and measuring the number of distinct intersections. Or we can take the quantity of black pixels along a scan line, and list the resulting numbers in some fixed order.

For example, the characters in Fig.2.7 have been coded by listing the number of intersections along two horizontal and two vertical scan lines, as indicated, making it easy to distinguish them as vectors in ${\fam11\tenbbb R}^4$.

The method is not likely to be satisfactory with only two horizontal and two vertical scanlines, but increasing the number can give more information. A little ingenuity can readily suggest variants of the method which may even give some degree of font invariance.