Iterative samples labeling for sketch recognition

Sketch recognition is an important issue in human-computer interaction, especially in sketch-based interface. To provide a scalable and flexible tool for user-centered sketch recognition, this paper proposes an iterative sketch collection annotation method for classifier-training by interleaving online metric learning, semi-supervised clustering and user intervention. It can discover the categories of the collections iteratively by combing online metric learning with semi-supervised clustering, and put the user intervention into the loop of each iteration. The features of our methods lie in three aspects. Firstly, the unlabeled collections are annotated with less effort in a group by group form. Secondly, users can annotate the collections flexibly and freely to define the sketch recognition personally for different applications. Finally, the scalable collection can be annotated efficiently by combining the dynamically processing and online learning. The extensive experimental results prove the effectiveness of our proposed method.