公开(公告)号：GB2495370A

公开(公告)日：2013-04-10

申请号：GB201215718

申请日：2012-09-04

Applicant: IBM

Inventor： GLEICHMAN SIVAN ,  GEVA AMIR ,  MARDER MATHIAS ,  OPHIR BOAZ

IPC: G06K9/46 ,  G06T7/00 ,  H04N1/409 ,  H04N1/58

Abstract: A method of detecting anti-aliased text in digital images includes identifying 42 a region of a digital image containing first pixels and second pixels. An image is initially converted 40 into a grayscale image. The first pixels are identified if they lie on a horizontal line with a positive intensity gradient, wherein for a given pixel, the left neighbouring pixel is darker (i.e. has a lower intensity value) and the right neighbouring pixel is lighter (i.e. has a higher intensity value). Second pixels (within a predefined proximity/distance from the first pixels) are identified if they lie on a horizontal line with a negative intensity gradient, wherein the left neighbouring pixel is lighter (i.e. has a higher intensity value) and the right neighbouring pixel is darker (i.e. has a lower intensity value). Once the pixels have been identified, a histogram of the pixel intently distribution is calculated 48 for the identified image region. The histogram/pixel intensity distribution is analysed 50 to determine whether the region contained anti-aliased text. The analysis identifies anti-aliased text if three criteria are satisfied, namely (i) pixels in the distribution are distributed into a specific number (e.g. 5-10) of groups, (ii) each group includes a number of intensities within a specified range (e.g., between 1-30 grey-levels), and (iii) the group containing the most pixels is either the group having the lowest intensity or the group having the highest intensity. If anti-aliased text is detected, the region is selected for optical character recognition (OCR) processing to extract the characters in the identified region. An associated apparatus and computer program product for carrying out the method are also claimed. The method is intended to enable higher accuracy optical character recognition of anti-aliased text in digital images.

公开(公告)号：GB2495370B

公开(公告)日：2015-07-01

申请号：GB201215718

申请日：2012-09-04

Applicant: IBM

Inventor： GLEICHMAN SIVAN ,  MARDER MATHIAS ,  OPHIR BOAZ ,  GEVA AMIR

IPC: G06K9/46 ,  G06K9/20 ,  G06T7/00 ,  H04N1/409 ,  H04N1/58

Abstract: A method, including automatically identifying, by a processor, a region of a digital image containing first pixels, each situated on a positive horizontal gradient, and second pixels in proximity to the first pixels, each situated on a negative horizontal gradient. A distribution of intensities of a color channel is then calculated for the pixels in the region, and the distribution is analyzed in order to detect whether the region contains anti-aliased text.