Extracting features from fuzzy cognitive parameter spaces

Underlying our proposed model is the perspective of handwriting recognition as a simple visual task involving the recognition of a finite set of simple, monochrome, stationary, 2D objects. We address the arbitrary nature of feature selection by adopting the rationale that the features chosen must each be cognitively plausible. This excludes many features commonly used in character recognition. The choice of individual features is based on empirical evidence for their significance in visual cognition. The three stages of our model of feature extraction are i) the preprocessing of raw data ii) the derivation of fuzzy-cognitive parameter spaces and iii) the identification of points of interest in these parameter spaces. These stages are discussed in turn in the following sections. We illustrate the description of the stages with examples of feature extraction from online data for handwritten alphanumeric characters.