How Accurate Is the Logic-Based Filter? - Logic-based Nonlinear Image Processing - page 25

Figure 3.9 Results of filtering edge noise.
3.2.2 Simple optical character recognition
The techniques described may not only be used for removal of different types of
noise from images, but also for recognition. This means that they may be trained to
admit certain structures found in the images and to reject others. To demonstrate
this property, a filter implemented within a 5 × 5 window was trained on the page of
text shown in Fig. 3.10(a). In this case, the “ideal” output corresponds to the image
shown in Fig. 3.10(b) containing only the letter “e”s from the original text. The out-
put from this filter is shown in Fig. 3.10(c). The results of repeating the process for
letters “a” and “e” are shown in Fig. 3.10(d). It can be seen that the recognition of
the characters is quite accurate. It should be remembered that the algorithm has no