Pattern Field Classification with Style Normalized Transformation
Xu-Yao Zhang, Kaizhu Huang and Cheng-Lin Liu
In field classification, patterns occur as groups (fields) of homogeneous styles, implying statistical dependencies among patterns in each field. This is distinct from the i.i.d. assumption in the traditional classification framework. In contrast to previous approaches that usually have limited performance due to either their computational inefficiency or inherent incapability of style transfer among fields, we extend the Bayes decision theory under two reasonable assumptions and develop the Field Bayesian Model (FBM) for field classification. Specifically, we propose to learn a Style Normalized Transformation (SNT) for each field. Via the SNTs, the data of different fields are transformed to a uniform style space (i.i.d. space). The proposed model is a general and systematic framework, under which many probabilistic models can be easily extended for field classification. Another desirable property of the proposed model is the ability to transfer to unseen styles. We conducted extensive experiments on face and speech data as well as a large-scale handwriting dataset (with 3755 classes and around 495K patterns). The experimental results are highly encouraging: we got significant error rate reduction compared to the state-of-the-art methods.