Direct Curvature Scale Space: Theory and Corner Detection
IEEE Transactions on Pattern Analysis and Machine Intelligence
Direct curvature scale space in corner detection
SSPR'06/SPR'06 Proceedings of the 2006 joint IAPR international conference on Structural, Syntactic, and Statistical Pattern Recognition
| Hi-index | 0.00 |
The curvature scale space (CSS) technique is one of thekey techniques of the MPEG-7 international standard inComputer Vision and Image Processing. It was selected as acontour shape descriptor for MPEG-7 after substantial andcomprehensive testing. However, to compute a CSS image ingeneral needs to wait a long time. This is verydisadvantageous when the CSS technique is applied to anobject recognition system to perform real-time recognition.In order to solve this bottleneck problem, a hybrid method forfast computing the CSS image is proposed in the presentpaper. In the method, firstly the curve is evolved in low scalespace, and after image noise is suppressed then the curvatureis evolved directly. Numerical experiments show that thehybrid method can perform equally well as the existingmethod. It is suitable for recognizing a noisy curve ofarbitrary shape at any scale or orientation. On the otherhand, the hybrid method only requires 1/3~1/5 CPU timeof the existing one. As a result, the CSS technique isimproved significantly for real-time recognition.