Issue when using UIImagePickerControllerOriginalImage (large images about tesseract-ocr-ios HOT 5 CLOSED

I've tested even bigger images, panorama images, and everything worked well.
Do you do any image preprocessing before recognition?
It seems like there is just a lot of noise in the bigger image, which is blurred during scaling.
You could simply check it, scaling down an image and scaling up it back before recognition.

from tesseract-ocr-ios.

Thanks for the reply Cyril - The only preprocessing I was doing was just a black and white conversion. But you’re right, the photo must have been too noisy, scaling down and back to the original size seems to do the trick, I get a better recognition rate as well on strings that have fractions (such as 1/4). anyway let me know if there is (or if you have) a special list of special image transforms I should apply to enhance the recognition. thanks again for the help!
On Sep 1, 2014, at 6:43 AM, Cyril [email protected] wrote:

I've tested even bigger images, panorama images, and everything worked well.
Do you do any image preprocessing before recognition?
It seems like there is just a lot of noise in the bigger image, which is blurred during scaling.
You could simply check it, scaling down an image and scaling up it back before recognition.

—
Reply to this email directly or view it on GitHub.

from tesseract-ocr-ios.

Take a look here: #42.
In my case, I try to filter images to remove as much noise as possible. Simple combination of Adaptive Threshold Filter and Median filter removes the small noise and makes an images in binary format (this step is also named as binarization).
The second step, I would like to implement, is to find a text areas in the images to remove shadows and other quiete big noise around the text, which could produce some comma and dot output. It could be possible (perhaps, at least I've read some papers about it) using combinations of Opening and Closening morphological operations. But so far I didn't have a luck to implement the second phase, which could operation on any image.

from tesseract-ocr-ios.

ah! excellent! Doing an adaptive Threshold Filter did the trick. Lower res photos have issues (I have a couple that are 600px wide) but changing the contrast, some adaptive threasholding and most of the camera taken photos get great recognition rate now. Thanks for the help!
On Sep 3, 2014, at 5:14 AM, Cyril [email protected] wrote:

Take a look here: #42.
In my case, I try to filter images to remove as much noise as possible. Simple combination of Adaptive Threshold Filter and Median filter removes the small noise and makes an images in binary format (this step is also named as binarization).
The second step, I would like to implement, is to find a text areas in the images to remove shadows and other quiete big noise around the text, which could produce some comma and dot output. It could be possible (perhaps, at least I've read some papers about it) using combinations of Opening and Closening morphological operations. But so far I didn't have a luck to implement the second phase, which could operation on any image.

—
Reply to this email directly or view it on GitHub.

from tesseract-ocr-ios.

I guess, it could be closed.

from tesseract-ocr-ios.