基于背景-前景分离模型的图表手绘信息识别

在各类文档、资料、报表中,许多数据使用图表进行表示,对图表进行自动识别和信息提取成为图形 识别和文档图像处理中一个重要的研究分支.本文提出了一种基于背景-前景分离的方法,针对司机日志图表中 的手绘折线信息,实现准确的信息提取与高精度识别.这种方法使用启发式规则分析和连通体分析将印刷体图 表的背景信息与手绘折线的前景信息相互分离,利用印刷体背景提供的刻度信息对手绘折线进行精确识别.同 时提供了良好的容错机制,降低手绘随意性造成的识别误差,并精确定位各种手绘错误造成的识别冲突,保证了 识别结果的有效性.通过4组805页测试样本集测试,本文方法的识别精度达到了90％以上.

Driver Schedule Chart Recognition Based on Background-foreground Separtion Model

In various kinds of document, paper and report, chart and diagram are widely used for data representation. Automatic recognizing and extracting information from chart and diagram becomes an important research branch of graphic recognition and document image analysis in recent years. This paper presents a novel method based background-foreground separation model for driver schedule chart recognition. Illuminative rules and connector -component analysis method are used to separate background (printed chart) from foreground (hand-drawn curve), background is analyzed to acquire scale information, foreground is analyzed to extract hand-drawn curve for recognition. Using scale information provided by background, hand-drawn curve can be precisely recognized. Error-toleration mechanism is also designed to locate hand-drawn error caused by filler and resolve recognition conflict automatically. This method is tested through four test bases contain total 805 pages of real driver log, more than 90% of test cases can be recognized correctly.