手绘草图识别中的相关反馈方法

Relevance Feedback in Freehand Sketchy Shape Recognition

手绘草图是人类思维外化和表达设计意图的有效工具之一,手绘草图的模糊性和用户适应性问题是革图识别中的关键问题。本文提出了将相关反馈机制引入到手绘草图识别中以捕捉用户意图的方法,该方法以抽取手绘草图的向量化特征为基础,首先利用基于图形特征的相似度计算,给出手绘草图候选识别结果集,然后借助用户对识别结果的相关性评价,通过逐渐调整图形构成特征的权重来捕捉用户输入意图,并提高识别效果。实验表明本文所提出的方法具有很好的效果。

Freehand sketching is an efficient way for recording and conveying ideas, especially in the early stages of design and idea externalizing. Main challenge in this area comes from the ambiguity of sketch and the user adaptation of sketch recognition. In this paper, we adapt the relevance feedback for sketchy shapes recognition to capturing user's intents. Firstly, the candidate sketchy objects are extracted by means of similarity calculation, based on the es- tablishment of feature-based vector model of freehand sketches. Secondly, the relevance feedback are used to capture the users' input intends and refine the recognition results incrementally by re-weighting the feature-based vector mod- el alone with the user's relevance judgement. Experiments prove the proposed method both effective and efficient.