ab, e Keywords: Text detection

Text region classification Histogram of oriented gradients for text n o f s ojec regions are then validated by a text/non-text classifier using a HOG-based descriptor specifically tuned algorit s of u to extract any legible textual information, such as street and traffic signs, store names, and building numbers. The extracted strings are then stored in a geo-referenced database, which is used to answer textual queries by users—for example, to locate the addresses of stores with a specified name or selling a specified product. The user itions. OCR algo-very poorly on are obtai text detec signed specifically for such images, as illustrated in Fig. 3(b The SNOOPERTEXT detector initially locates candidate cha on the images by using image segmentation and shape-base acter/non-character binary classification. The candidate characters found in this step, represented by their bounding boxes, are grouped by simple geometric criteria to form either candidate words or candidate text lines. These candidate text regions are then validated by a binary text/non-text region classifier that re-jects any candidate region that does not appear to contain a single line of text. This classifier uses the T-HOG descriptor [4], which is based on the multi-cell histogram of oriented gradients (HOG) of q This paper has been recommended for acceptance by Daniel Lopresti, Ph.D. ⇑ Corresponding author. Computer Vision and Image Understanding xxx (2013) xxx–xxx Contents lists availa