International Conference on Instrumentation, Measurement, Circuits and Systems (ICIMCS 2011)
41 An Approach to Optimal Filter for Edge Detection Based on LS-SVR
Download citation file:
- Ris (Zotero)
- Reference Manager
A design approach to edge detection filter via utilizing Least Squares Support Vector Regression (LS-SVR) based on Canny's criteria was presented. LS-SVR is a discrete smooth filter. Aimed at the inaccurate localization of 1-D discrete localization criterion by Demigny, we proposed the new 1-D expression of criterion for “good Localization” in order to solve the optimal value. In the paper, we introduced the construction ethod of first and second derivative operators. According to the new criteria, the optimal parameters of Canny edge detection operators were determined by numerical experiment. The approach of 1-D optimal edge detection operators were also realized by adjusting LS-SVR parameters. The corresponding LS-SVR optimal parameters can be calculated. Experimental results showed that utilizing the edge detection operators designed by LS-SVR had better noise resistance ability.