具有连续反馈的在线草图符号识别算法

当前多数在线草图符号识别方法是在用户绘制完图形后,通过点击按钮或等待一定时间来反馈识别结果。为提供连续反馈,提出了一种基于图元的部分草图识别算法:对输入笔画进行笔画分割得到图元表示,提取表示两个图元空间关系的新特征,依据指派问题模型,通过该特征的相似度识别符号。算法简单,无须训练,而且与笔顺、笔画数和图形大小无关。实验证明了方法的有效性。

A primitive-based recognizer for online sketched symbols with continuous feedback

Most of current research on online sketched symbol recognition concerns the fully completed symbols. To recognize the partially drawn symbols and reduce input stroke number with continuous feedback, a primitive-based recognizer is proposed. It is a template matching algorithm. In the stage of feature representation, the input strokes are segmented as primitives. We also present a new space feature between a pair of primitives. In the stage of template matching, we use the model of assignment problem to calculate the matching cost of two symbols. The proposed method is invariant with respect to stroke order and number, as well as the scale. Experimental results demonstrate that the proposed method is effective and efficient for recognizing incomplete sketches.