Performance evaluation methodology for document image dewarping techniques
The performance evaluation of dewarping techniques is currently addressed by concentrating in visual pleasing impressions or by using optical character recognition (OCR) as a means for indirect evaluation. In this study, the authors present a performance evaluation methodology that calculates a comprehensive evaluation measure which reflects the entire performance of a dewarping technique in a concise quantitative manner. The proposed evaluation measure takes into account the deviation of the dewarped text lines from a horizontal straight reference which is considered to be the optimal result. This measure is expressed by the integral over the dewarped text line curves. To reduce the manual effort for identifying the text lines in the dewarped image, the authors propose a point-to-point matching procedure that finds the correspondence between the manually marked warped document image and the dewarping counterpart. This enables an evaluation for unlimited number of methodologies addressing a marking procedure which is applied only once. The validity of the proposed performance evaluation methodology is demonstrated by a concise experimental work that comprises four state-of-the-art dewarping techniques as well as the involvement of different users in the interactive part of the procedure.
