混合点状和非规则军标的在线手绘识别

当前对在线手绘军标图符识别的研究只针对单一类型的手绘点状军标或非规则军标,分别使用不同方法进行识别.但在特殊应用中二者常混合输入,当待识别军标图符的类型未知时,如何识别是一个重要问题.提出一种基于最小生成树(MST)覆盖模型的混合识别方法,训练阶段,分别对点状和非规则军标样本建立MST覆盖模型,并训练一个二分类支持向量机(SVM)分类器;识别阶段,先通过几何和结构信息粗判断军标类型,再通过置信度估计和融合的方法确定未知军标的类型.在113类点状军标和36类非规则军标的数据集中实验,军标类型区分准确率为94.7%,最终识别率为91.6%,且能满足实时要求.

Online sketch recognition for mixed point and irregular military symbols

Most of current research on online sketched military symbols recognition concerns only one type of symbols, point symbols or irregular symbols, using different methods to recognize separately. But in practical applications the two types of symbols are mixed. It becomes a major issue to find a way to recognize a type-unknown military symbol. A minimum spanning tree (MST) covering model-based mixed recognition method was proposed. In the training phase, two MST-based covering models were built for point and irregular symbols respectively. And then a two-class support vector machine (SVM) classifier was trained. In the recognition phase, the coarse type identification was accomplished by using the geometrical and structural information firstly. Then the confidence estimations were calculated and integrated to identify the type of the unknown symbol. Different types of symbols were classified by two existing modules. The algorithm was tested on 113 classes of point symbols and 36 classes of irregular symbols. The accuracy rate of symbol type identification was 94.7%, and the final recognition rate was 91.6% in real time.