A geometric approach for accurate and efficient performance evaluation of layout analysis methods

A major component of performance evaluation of layout analysis methods is the comparison of ground truth regions with regions resulting from segmentation methods. The description of document regions must be both accurate in describing complex layouts and efficient in view of the large number of region comparisons that must be performed. Previous approaches favour either accuracy or efficiency, resulting in an impractical compromise. This paper presents an improved approach that uses polygons to accurately describe both segmentation and ground truth regions. Polygonal descriptions are efficiently compared using a rectangular interval based decomposition. This approach has been validated using data from the ICDAR page segmentation competitions.