Sketch recognition using auto-completion
Tırkaz, Çağlar and Yanıkoğlu, Berrin and Sezgin, Metin Sketch recognition using auto-completion. In: IEEE Computer Vision and Pattern Recognition (CVPR) 2011, (Submitted)
Full text not available from this repository.
Sketching is a natural and intuitive form of human interaction that is employed in a variety of areas, such as engineering drawings or classroom teaching. Real time recognition of hand drawn sketches is a challenging problem due to the variability in hand drawing, as well as the variability in the drawing order. We propose an auto-completion system that is capable of classifying a partially drawn sketch if there is sufficient information to identify it as a domain object, thus increase sketching throughput. In cases where the label of the partial drawing can not be predicted unambiguously, our system delays the classification decision until more information becomes available. The system is a hierarchical classifier that consists of a semi-supervised clustering followed by supervised classification, so as to deal with the ambiguity involved in classifying partial sketches that may belong to more than one category. We report experiments using the Course of Action Diagrams database, showing an average recognition rate of 90% for partial sketches at human ground-truth levels.
Repository Staff Only: item control page