A study of the adapting schematics for text-independent online writer identification

In the Gaussian mixture model based online writer identification system, the writer specific models are usually learned by adapting the universal background model. However, among all the possible adapting plans, which one performs best is still an unsolved problem, as well as the underlying principles. Towards finding the answer, this paper analyses all the combinations of the parameter adaptation. The conclusion is that the local and global adapting plan make use of different information for discrimination. Adapting both the local and global aspect is capable of use the both information for discrimination. In the experiments, a study of 5 specific adapting plans is performed in different available test data, relevance factor, number of mixtures and feature set, which shows the advantages of the plan adapting both aspects. On the point based feature set, the weight adapting plan performs better than the mean adapting plan, which is contradict with the current research. On the all point based feature set, the plan adapting both the mean and weight acquired competitive performance with the state-of-the-art.