Don't waste your time with Tesseract here, I tell ya. You'd only get all sorts of unnecessary hassle. And what's most important, you'll be frustrated by accuracy.
By "pixel-to-pixel" I mean what is described e.g. here, section "Naive Template Matching": http://docs.adaptive-vision.com/current/studio/machine_vision_guide/TemplateMatching.html But in your case that wouldn't be dumb iteration over the entire image, but a single check in a fixed location whether the template image has exactly same pixels as the input image. You can arrange it like this: - Crop out samples of all digits (each sized 85x60) -> digit0.png .. digit9.png - Crop out the same sized rectangle from a fixed location of your source image - e.g. score digit #0 -> score0.png - Do file compare score0.png to digit0.png - If no match - try digit1.png ... - Match found - this is your score digit #0 - Take next score digit ... - Proceed to time digits ... - Done Simple! Above approach probably would adapt for other games, and you'd manage to use same digit samples. File compare might be replaced by XOR and then calculating the mean of all pixels (should be 0 if match). There can be other methods of comparison. You get the point. You'd better invest your time into accumulating a collection of score digit coordinates in each game, than into a struggle with quirky OCR results. Well, unless you're eager to. Best regards, Dmitri Silaev www.CustomOCR.com On Thu, Apr 23, 2015 at 10:51 PM, Leah Siddall < [email protected]> wrote: > Also, I wanted to show the output from the lower third: > > > <https://lh3.googleusercontent.com/-c527dWolCTI/VTlMFuVJG2I/AAAAAAAAATg/i7ELL6UpHFY/s1600/mario_lower.png> > > E53333 I-I-I-I-I-|--E.|- $5 a]. > ICED}: E]- EIEIEIEEHEIEJ GEE-'3 > > > as you can see, i'm not even getting numbers. :/ > > > On Thursday, April 23, 2015 at 12:28:14 PM UTC-7, Leah Siddall wrote: >> >> thanks for your feedback! >> >> I was hoping to kinda not lock into one video game, so precision of where >> the high score may not be the same place will rule out cropping. I planned >> on doing a regex against whatever came back from tesseract. I was already >> counting on garbage information so there was going to be some light >> scripting wrapping this. >> >> But, when i cropped only to the "lower third" section of the mario >> screenshot, i was still not getting anything close to the score or time. >> Why is it struggling wit this font? it seems incredibly straight forward >> except that the "scores" are not a solid color with a border and sometimes >> they are touching. >> >> Since this is a new arena to me, can you point me in the right direction >> of researching how to do the "pixel-to-pixel" matching? >> And, I am new to the idea of training tesseract. can I train it to >> understand this font? >> >> This is more exploratory and fun for me, so I am very willing to learn >> the "correct way" of doing this. I just want to be pointed in the right >> direction. >> >> thanks again!! >> >> >> On Thursday, April 23, 2015 at 2:51:55 AM UTC-7, Dmitri Silaev wrote: >>> >>> Hmmm, fixed image size, fixed region, constant colors, monospace raster >>> font... >>> >>> Do you really want to engage a whole algorithmic monster to handle a >>> problem like this? Not to mention poor performance, training, >>> preprocessing, coping with all sorts of recognition problems is guaranteed. >>> >>> Pixel-to-pixel matching is the way to go! >>> 100% accuracy. >>> >>> Even if you not willing to resort to full fledged programming - just >>> crop out 10 digit samples and match them to your input image using a shell >>> script loop. Give your ImageMagick-fu a chance. Or, you can even use file >>> compare! )) >>> >>> HTH >>> >>> Best regards, >>> Dmitri Silaev >>> www.CustomOCR.com >>> >>> >>> >>> >>> >>> On Thu, Apr 23, 2015 at 9:05 AM, Leah Siddall < >>> [email protected]> wrote: >>> >>>> Hi all! >>>> >>>> I am not having luck with tesseract and the fonts used in NES games >>>> like Super Mario Bros. 3. ( i've attached an example screenshot ). >>>> My goal is scrape a screenshot for the "score" and "time remaining". >>>> The idea is to feed that into a database during a competition to minimize >>>> cheating. >>>> >>>> I've tried cropping, resizing, grayscale, and negating with PNG, TIF, >>>> JPG, and PNM formats then going through every PSM mode on each with poor >>>> results. >>>> The original screenshot is PNG 4800 × 3600 pixels at 144 pixels/inch >>>> straight from the emulator which is like the best possible situation. >>>> >>>> Just trying to get a baseline, I tried against the "Punch Out" >>>> screenshot ( attached ) where the fonts are clearly spaced and lots of >>>> empty space. It would get "CDHTIHUE" and "Nintendo", but totally missing >>>> the word "new" between the boxing gloves and and jumbling the year numbers. >>>> >>>> To rule out user error, I did run against other images with more >>>> standard fonts and had no problems. >>>> >>>> I'm quite comfortable with imagemagick but very new to tesseract. >>>> I am using tesseract version from "brew install tesseract -HEAD" on >>>> OSX 10.10.2 >>>> tesseract 3.04.00 >>>> leptonica-1.71 >>>> libjpeg 8d : libpng 1.6.16 : libtiff 4.0.3 : zlib 1.2.5 >>>> >>>> This would be really really cool to pull off if possible. any >>>> suggestions are greatly appreciated. >>>> thanks!! -leah >>>> >>>> -- >>>> You received this message because you are subscribed to the Google >>>> Groups "tesseract-ocr" group. >>>> To unsubscribe from this group and stop receiving emails from it, send >>>> an email to [email protected]. >>>> To post to this group, send email to [email protected]. >>>> Visit this group at http://groups.google.com/group/tesseract-ocr. >>>> To view this discussion on the web visit >>>> https://groups.google.com/d/msgid/tesseract-ocr/2088977c-529b-45bd-8059-b6906fb666ce%40googlegroups.com >>>> <https://groups.google.com/d/msgid/tesseract-ocr/2088977c-529b-45bd-8059-b6906fb666ce%40googlegroups.com?utm_medium=email&utm_source=footer> >>>> . >>>> For more options, visit https://groups.google.com/d/optout. >>>> >>> >>> -- > You received this message because you are subscribed to the Google Groups > "tesseract-ocr" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at http://groups.google.com/group/tesseract-ocr. > To view this discussion on the web visit > https://groups.google.com/d/msgid/tesseract-ocr/f84a83cc-0ac0-45f9-b487-0c98537848ac%40googlegroups.com > <https://groups.google.com/d/msgid/tesseract-ocr/f84a83cc-0ac0-45f9-b487-0c98537848ac%40googlegroups.com?utm_medium=email&utm_source=footer> > . > > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at http://groups.google.com/group/tesseract-ocr. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAKzLxFMVLxcupHujQim9iZRETdXKo%2BE%2BBGr6w3w0LoSLzxuprQ%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
