摘要
针对目前二维草图识别技术存在的误识别率高、智能性差等缺点,提出一种通过捕捉设计意图修正贝叶斯推理结果的草图识别技术.首先研究绘制过程中笔触的速率、压力等变化规律与设计意图的关系,建立草图绘制规律与设计意图关系模型,实现实时设计意图捕捉;其次建立二维草图识别的贝叶斯网络,根据绘制几何特征分配条件概率表;再利用捕捉的设计意图修正各个约束特征节点的条件概率,使贝叶斯网络能够推理出符合设计意图的几何元素及约束关系;最后建立草图识别系统,并引入评价反馈机制增强识别的自适应性和准确性.实验结果表明,该技术使二维草图的误识别率降低约30%,有效地减少了漏识别发生,对无设计意图绘制扰动具有区分能力.
The high misrecognition rate and poor intelligence are the central flaws of current 2D sketch recognition methods. This paper presents a design intent capture approach to identify 2D sketches more accurately by improving the Bayesian network. Through analyzing the sketching behavior, we built the relationship between the sketching regularity such as the velocity and pressure from drawing and design intent. Then the Bayesian network for recognition reasoning is constructed, and a conditional probability table （CPT） is assigned to each node according to the sketch geometric feature. By the design intent probability, the CPTs of constraint nodes are modified to improve recognition precision. Finally, the recognition process is proposed, and the weights are automatically adjusted by the user＇s feedback. Compared to the work without intent capture, this technology can figure out the design elements and constraints during sketching, decrease the misrecognition by 30%, and distinguish the drawing disturbance.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2012年第10期1337-1345,共9页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(51175287)
关键词
设计意图
草图识别
贝叶斯网络
绘制规律
条件概率表
design intent
sketch recognition
Bayesian network
sketching regularity
conditional probability table
