Pairwise coupling support vector machine and its application on handwritten digital recognition

In this paper, a hierarchical structure combining a linear classifier based on the Mahalanobis distance and pairwise coupling (PWC) is proposed to effectively tackle a multi-class classification problem. Given a testing pattern, the conventional PWC needs to evaluate K(K-1)/2 binary classifiers and treats them as the same. In fact, different binary classifiers have different impacts on the final decision. A weight matrix is introduced through a coarse classifier, reflecting which binary classifiers are more relevant to the given sample. Due to taking advantage of the distribution information of the dataset, the recognition rate can be improved. Experimental results on handwritten digit recognition demonstrate our method is effective and efficient.