SVM Combination for an Enhanced Prediction of Writers' Soft Biometrics
Nesrine Bouadjenek, Hassiba Nemmour, Youcef Chibani · 2017
This chapter deals with automatic prediction of writers' soft biometrics regarding gender, handedness, and age range. For this purpose, we propose the use of Sugeno's fuzzy integral and its modified form, namely, fuzzy Min-Max as a strategy to combine different support vector machines (SVM) classifiers paired to different features. Presently, we target a local calculation of histogram-based features: rotation-invariant uniform local binary patterns, a histogram of oriented gradients, and gradient local binary patterns. These features characterize different kinds of handwriting traits to verify SVM complementarity for the combination setup. Experiments are conducted on standard Arabic and English handwritten sentences. The obtained findings proved the effectiveness of the proposed combination system.