Style-conscious quadratic field classifier
Sriharsha Veeramachaneni, Hiromichi Fujisawa, Cheng‐Lin Liu, George Nagy · 2003
When patterns occur in the form of groups generated by the same source, distinctions between sources can be exploited to improve accuracy. We present a method for exploiting such 'style' consistency using a quadratic discriminant. We show that under reasonable assumptions on the feature and class distributions, the estimation of style parameters is simple and accurate, therefore, style constrained classification deserves further attention.