1. What version are you using?? (tesseract -v) Tess 3 and Tess 4 are VASTLY different. ALWAYS include this information! 2. The text is too small, only a few pixels. Tess was trained for photos of printed text. It also looks like you upscaled the images, which confuses Tess even more. 3. The text is antialiased. Before Tess processes an image, it uses Leptonica to binarize (convert to black and white with no shades of gray) the image so it can handle it. That means those "in-between" pixels in the image are going to turn to either black or white. Try running tesseract test.png out -l hrv --oem 3 --psm 6 -c tessedit_write_images=1 then attach the file named "tessinput.tif".
On Wednesday, October 25, 2017 at 4:37:47 AM UTC-4, Adrian Pop wrote: > > > I've tried using the ocr to get the texts from these kind of images (even > tried rescaling and the results were not better). I uploaded two examples. > The code I ran is: > tesseract FILE_NAME OUTPUT -l hrv --oem 3 --psm 6 > The language is Croatian (so hrv) > > > For test6.png the result is: > KELEMINDVIĆ STJEPAN > Keleminovići 2 > 10363 Adamovec > The 'O' was made into a 'D' > > For test8.png the result is: > 012 MALKS, BISTRO. VL. BEDNARŠEK KREŠIMIR > Antuna Nemčića 3 > 10430 E:;amubnr > 'Samobor' was turned into 'E:;amubnr' > > For test10.png it's a mess: > ou. PIZZERIANL. BLECIC mmm > > Any ideas how to tweak these more, or why it failed this bad? > > -- 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 https://groups.google.com/group/tesseract-ocr. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/9c619e71-9925-4008-a932-980de29454fa%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
