Microscopic Slice Image Processing and Automatic Labeling - Biomedical Signal Analysis: Contemporary Methods and Applications - page 365

(a) directional normalization

(b) training data set (c) classification result

Figure 13.7
Directional neural networks. (a) shows five rotated As together with their
normalized image patches (below).

Figures 13.7(b) and 13.7(c) briefly demonstrate the successful ap-
plication of the ZANE classification scheme based on directional neural
networks to character detection. A character classifier ζ is trained using
a directional network based on a generalized regression neural network
[279], a specialized radial basis function network allowing for more com-
plicated nonlinear function approximations. The classifier is trained us-
ing the seven noisy “A” training samples from 13.7(b) together with 70
randomly selected non-A patches. After directional normalization, PCA