Geometrical feature based ranking using grey relational analysis (GRA) for writer identification

The author's unique characteristic is determined by the variation of generated features from feature extraction process. Different sets of features produced are based on different feature extraction methods (local or global). Thus, the process has led to the production of high dimensional datasets that contributing to many irrelevant or redundant features. The main problem however is to find a way to identify the most significant features. The features ranking method using Grey Relational Analysis (GRA) is proposed to find the significance of each feature and give ranking to the features. This study presents the Higher-Order United Moment Invariant (HUMI) as the global feature extraction methods. The combinations of features with the higher ranking are discretized and used as the subsets of features to identify the writer. The result demonstrates that the average classification accuracy of five classifiers by using just the combination of two most significant features have yielded a better performance than using all features.