Abstract— A new descriptor is proposed for sketched symbol recognition. It incorporates local direction information of sampling points into the BSM (Blurred Shape Model) descriptor, which represent only the spatial distribution of a symbol. In the proposed directional BSM (DBSM), a symbol is firstly decomposed into four directional sub-symbols. Then in each sub-symbol, the nrBSM (Non-Rigid BSM) descriptor is extracted. The four nrBSMs form the final DBSM descriptor. It inherits from the traditional BSM the invariance to stroke order and number. Moreover, the new descriptor results in better performances in sketched symbol recognition.
Index Terms— Sketched symbol recognition, blurred shape model (BSM), directional information, classification.
The authors are all with the Equipment Academy, Beijing 101416, China (e-mail: dengwei@whu.edu.cn, wld@nudt.edu.cn, visworker07@163.com, freelancer_lys@163.com).
[PDF]
Cite: Wei Deng, Lingda Wu, Zhonghua Yao, and Yashuai Lü, " Directional Blurred Shape Model Descriptor for Sketched Symbol Recognition," International Journal of Information and Electronics Engineering vol. 5, no. 6, pp. 469-472, 2015.