There are tons of. And I believe, no ready recipe can be used universally, this is very task-specific, especially in photographic images. Also I believe, to do good text detection your algo should in some extent mimic human behavior so it probably should be multi-stage, gradually refining results at every stage. Don't account on getting a working code snippet from the internet, most likely you'd have to write the code yourself.
Some articles I had picked out when I was self-studying this field of document image processing. For the moment, there might be newer ones, but these can provide you with the basis. Apologies, I've no time to provide you with direct references and author names - I only listed my file system directory on this topic. You can Google for exact article titles to find links. 1990 Scale-Space and Edge Detection Using Anisotropic Diffusion.pdf 1998 Edge detection and ridge detection with automatic scale selection.pdf 2001 Edge-Based Method for Text Detection from Complex Document Images.pdf 2001 TEXT EXTRACTION FROM GREY SCALE PAGE IMAGES BY SIMPLE EDGE DETECTORS.pdf 2002 Gaussian-Based Edge-Detection Methods - A Survey.pdf 2003 Fast Computation of Scale Normalised Gaussian Receptive Fields.pdf 2003 Real-time scale selection in hybrid multi-scale representations.pdf 2003 Recognition of text in 3-D scenes.pdf 2004 A method for ridge extraction.pdf 2004 A Review of Vessel Extraction Techniques and Algorithms.pdf 2004 Distinctive Image Features from Scale-Invariant Keypoints.pdf 2004 Scene Text Extraction in Natural Scene Images using Hierarchical Feature Combining and Verification.PDF 2004 Text Detection from Natural Scene Images - Towards a System for Visually Impaired Persons.PDF 2005 A novel approach for text detection in images using structural features.pdf 2005 Color Text Extraction from Camera-based Images - the Impact of the Choice of the Clustering Distance.PDF 2005 Improved Text-Detection Methods for a Camera-based Text Reading System for Blind Persons.PDF 2005 Text Extraction from Gray Scale Historical Document Images Using Adaptive Local Connectivity Map.pdf 2006 Multiscale Edge-Based Text Extraction from Complex Images.PDF 2006 Spatial and Color Spaces Combination for Natural Scene Text Extraction.PDF 2008 A double-threshold image binarization method based on edge detector.PDF HTH Warm regards, Dmitry Silaev On Sat, Mar 5, 2011 at 8:56 AM, Saurabh Gandhi <saurabh...@gmail.com> wrote: > Hey, > Any algorithm / whitepaper suggestions for text extraction, especially if > the text is not over-lay text but a part of the image itself. Most > algorithms I saw on the internet are compute intensive. > > -- > Regards, > Saurabh Gandhi > > > > > On Sat, Mar 5, 2011 at 11:20 AM, Dmitry Silaev <daemons2...@gmail.com> > wrote: >> >> Zdravko, >> >> You should do text-detection before passing images to Tesseract. >> Text-detection is a process of determining of image regions containing >> text. Even if an image contains no text, Tesseract anyways will treat >> it as an image of text. >> >> Before recognition Tess applies a so-called binarization algorithm, >> which converts an RGB image to monochrome one (black for text and >> white for background). For your sample image the Otsu binarization >> used in Tesseract (http://en.wikipedia.org/wiki/Otsu%27s_method) would >> certainly give a number of skewed vertical lines resembling >> backslashes and further recognition classifies them as such. >> >> "textord_heavy_nr" and some other variables control size-based noise >> removal but work satisfactory only in case when there's a significant >> body of good text surrounded but some amount of noise. In your image >> everything is noise, so it won't work. >> >> Therefore you need to extend your pre-processing in order to feed Tess >> with images indeed containing text. Decisions can be made based on >> contrast estimation, distinctive color distribution, etc. >> >> HTH >> >> Warm regards, >> Dmitry Silaev >> >> >> >> >> >> On Fri, Mar 4, 2011 at 5:25 PM, zdravco <zdra...@gmail.com> wrote: >> > Hello, >> > >> > I am using tesseract in my project after some image pre-processing. >> > There are some false negatives I was hoping tesseract would eliminate >> > by producing no output. However, sometimes there is a strange output >> > that I get from almost blank images. >> > Here is the sample image: >> > https://picasaweb.google.com/zdravco/TesseractTest#5580227257541654274 >> > >> > When I run it with tesseract rev. 552 using English language I get: >> > " \\\\ R \." >> > >> > Does anyone know if there are some options in tesseract that could >> > eliminate this noise? Or maybe if I could improve my input image with >> > some further pre-processing. I have also tried to recompile tesseract >> > with "textord_heavy_nr" set to TRUE, but then the output is: >> > "an \\“ R \". >> > >> > Thanks, >> > Zdravko >> > >> > -- >> > You received this message because you are subscribed to the Google >> > Groups "tesseract-ocr" group. >> > To post to this group, send email to tesseract-ocr@googlegroups.com. >> > To unsubscribe from this group, send email to >> > tesseract-ocr+unsubscr...@googlegroups.com. >> > For more options, visit this group at >> > http://groups.google.com/group/tesseract-ocr?hl=en. >> > >> > >> >> -- >> You received this message because you are subscribed to the Google Groups >> "tesseract-ocr" group. >> To post to this group, send email to tesseract-ocr@googlegroups.com. >> To unsubscribe from this group, send email to >> tesseract-ocr+unsubscr...@googlegroups.com. >> For more options, visit this group at >> http://groups.google.com/group/tesseract-ocr?hl=en. >> > > -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To post to this group, send email to tesseract-ocr@googlegroups.com. To unsubscribe from this group, send email to tesseract-ocr+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/tesseract-ocr?hl=en.
