A combined crisp and fuzzy approach for handwriting analysis

Abstract

This paper presents an off-fine writer-independent handwriting analysis system which utilizes both classical crisp and fuzzy methodologies to output possible personality traits of the writer. The design deploys an analytical handwriting analysis approach based on two primitives, the baseline and the slant angle of the characters. The objective of the design strategy is to present a group of parameters for handwriting analysis based on the text. These parameters allow for the classification of writing into different categories which could be used as a preliminary step for outputting the personality traits of the writer. Two parameters, the baseline and the slant-angle, are the inputs to a rule-base which outputs the personality trait category. The evaluation of the baseline is non-fuzzy (crisp) whereas the evaluation of the slant-angle utilizes the fuzzy paradigm. The approach is based on a combination of classical geometric arithmetic evaluation and fuzzy control designs. For determination of the base line angle two methodologies are explored: the geometric-features based segmentation method and a method based on biologically inspired generation theories or the low pass filtering method. We utilize the geometric features evaluation for the baseline extraction since it proves more robust with respect to the variations of the handwriting in an off-line environment. For determination of the slant type a fuzzy technique is adopted to determine the contributions of the slant-type angle to each of the five variations of the slant-type categories. The uncertainties in the system model are expressed by fuzzy-valued model parameters with their membership functions derived from experimental data. In total five variations of slant type are considered. These include extreme left, controlled left, vertical, controlled right and extreme right. Fifteen personality traits PT1 - PT15 were identified and sets of rules formulation were created, (e.g., If Input1 is "level" and "Input2" is "Controlled Left" then Output is PTx.) The proposed approach takes advantage of two differing methodologies that have clear outputs to evaluate two attributes of handwriting. The outputs are utilized to determine a personality trait. The system can be further enhanced by including more parameters such as size of letters, spacing between letters and other attributes of handwriting as part of the inputs for trait determination.