Fast and Lightweight Text Line Detection on Historical Documents

Robust Text Line Detection in Historical Documents: Learning and Evaluation Methods

“…This problem has been little studied in the literature. For example, when confronted to touching bounding boxes, Melnikov and Zagaynov [31] suggested to remove the ascenders and descenders by reducing the height of the annotated boxes by 30% at the top and the bottom. Then, they downscaled the rasterized polygons to the input resolution to train their model.…”

“…Object-level IoU P/R F1 P/R R@.85/.95 mAP@.65 mAP Barakat [23] Mechi [24] Renton [25] Doc-UFCN [6] dhSegment [7] Yang [27] Tarride [2] Soullard [30] Melnikov [31] To tackle this problem, metrics originally designed in the Information Retrieval community [34] have been adapted to images and used during the PASCAL VOC Challenge 2012 to compute the Precision at objectlevel. During this competition, the detection task was evaluated based on the Precision-Recall curve at objectlevel, where the detections are considered as true or false positives according to their area of overlap with the ground-truth objects.…”

Page Layout Analysis System for Unconstrained Historic Documents

“…These approaches, however, make assumptions about the document, such as constant text orientation, and rely on only processing a single text column at time. Other authors rely on segmenting parts the text body, such as baseline or x-height area (see Figure 1), using CNNs first [13,12]. To convert the detected text x-height areas to text line polygons, Pastor-Pellicer et al [17] use local extreme points of binarized text to assign ascender and descender lines to each detected text line.…”