Citation
Tom Y. Ouyang and Randall Davis. 2009. A visual approach to sketched symbol recognition. In Proceedings of the 21st international jont conference on Artifical intelligence (IJCAI'09), Hiroaki Kitano (Ed.). Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 1463-1468.
Summary
The paper presents a sketched symbol recognition that focuses on the visual appearance of the symbols. It uses a new symbol classifier that is computationally efficient and invariant to rotation and local deformations. It evaluates performance on following three domains: handwritten digits, PowerPoint shapes, and electrical circuit symbols.
The paper represents symbols as feature images rather than as geometric primitives. It also proposes a set of visual features which captures on-line stroke properties (eg. orientation, endpoint location). The paper introduces a computationally efficient classification technique which is robust to rotation and local deformations. The paper then talks about it exceeding performance on all of the three datasets evaluated.
Four reference angles are used as orientation features correspond. The endpoint feature identifies stroke endpoints. The stroke is then sent for smoothening and Downsampling.
IDM Distance between two symbols I1 and I2 is represented as:
D2 = X x,y min dx,dy ||I1(x + dx, y + dy) − I2(x, y)||2
Discussion
Getting good accuracy in three different domains is appreciable. There still is some scope of improvement as SVM-linear seems to perform better. Probably it can be combined with SVM-linear and used as a collaborative system.
No comments:
Post a Comment