Although handwritten flowchart recognition has been studied for many years, it is still a tough problem. Most traditional methods are based on grouping strokes, which is known as online handwritten flowcharts recognition. However, these methods cannot be applied directly on the offline handwritten flowcharts which are lack of the stroke information. Hence it is in a great demand to design an effective algorithm to recognize the offline handwritten flowcharts. Inspired by Region Proposal Network in deep learning, a framework is proposed to detect different symbols in offline handwritten flowcharts. With the predicted localization, a pixel searching algorithm is presented to refine the results of symbols. Finally, all the symbols in flowcharts are connected, which turns offline flowcharts into images and structural XML. Experiments are conducted on several datasets. The evaluation results show the accuracy of nodes recognition, text recognition and arrow recognition hits 99%, 96.5% and 84% respectively, and the results of all types of symbols are better than those of other online and offline methods.
Volume 11 | 03-Special Issue
Pages: 1788-1798