在线草图识别中用户手绘习惯建模方法  

User Modeling for on-Line Sketchy Graphic Recognition

作　　者：张斌[1] 孙正兴[1] 孙建勇[1] 彭彬彬[1] 

机构地区：[1]南京大学计算机软件新技术国家重点实验室计算机科学与技术系,南京210093

出　　处：《计算机科学》2004年第6期194-198,共5页Computer Science

基　　金：国家自然科学基金项目(6990300;60373065)资助

摘　　要：手绘草图是概念设计和思路外化的一种高效的表达方式。用户绘制草图时存在的多种形式,及其随意性和模糊性使得用户适应性问题逐渐成为草图识别的核心课题。本文提出了一种在线草图识别的用户建模方法来捕捉绘制草图时的用户习惯,主要包括两个方面的内容:一是基于SVM的主动式增量学习方法,二是基于动态用户建模的手绘复杂图形的识别方法。前者与传统的增量式学习方法相比,在识别精度相同的情况下所需的训练时间和训练数据集要少得多。后者则是基于笔划信息以及笔划间的顺序和空间关系信息,采用增量式决策树捕捉用户的输入习惯和过程信息。实验证明了本文方法在在线草图识别中的有效性和高效性。Freehand sketching is an efficient way for idea externalizing and concept design. User adaptation to be the center of the one of sketch recognition, because of multiformity, arbitrary and fuzziness during users' sketching processes. This paper presents a user modeling method of online sketch recognition to capture users' sketching habits, including two aspects: Strokes recognition based on SVM-based active incremental learning and composite sketchy graphics recognition based on dynamic user model. The former, compared to the normal incrementl learning method, needs much less learning time and small training dataset under the same recognition precision. The latter, based on the stroke's information and the special relation and sequence of strokes, adopts the incremental decision tree to capture users' input habits and process information. Experiments show it both effective and efficient in online sketchy graphic recognition.

关 键 词：在线草图识别 用户适应性 用户建模 增量式主动学习 支撑向量机 SVM 模糊预测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]