A System for Performance Evaluation of Arc Segmentation Algorithms
Wenyin Liu, Jian Zhai, Dov Dori, Long Tang · 2001
Accurate segmentation of circular arcs from line drawings is essential for higher level processing in document analysis and recognition systems. In spite of the prevalence of arc segmentation methods, robust algorithms that perform well in complex graphic environments are scarce, and methods to evaluate such algorithms are even more rare. Extending our previous work, we propose a comprehensive system for evaluating arc segmentation. The system models three types of noise that can be found in real life drawing images: pixel, vector and context. To test the system, we applied it to evaluate an arc segmentation algorithm and obtained accurate numeric values for the various performance metrics of the algorithm. It has also been effectively used as the evaluation system in the IAPR Arc Segmentation Contest.