App take 40M memory when recognize the image using Tesseract about tesseract-ocr-ios HOT 18 CLOSED

Which version of the library are you using? If you're using the 2.2 CocoaPod, there is a known memory leak in that version, but it has since been fixed. The CocoaPod has just not been updated, for some unknown reason (see #49). You can use the latest version of the repo via CocoaPods by using this line in your Podfile:

pod 'TesseractOCRiOS', :git => 'https://github.com/gali8/Tesseract-OCR-iOS.git'

from tesseract-ocr-ios.

I use the latest version 3.03

from tesseract-ocr-ios.

Can you post a code snippet or link to your code so I can try reproducing the high memory usage?

from tesseract-ocr-ios.

I download the source code as Zip on the gitHub, and then copy the TesseractOCR.framework in the product folder to my project.

from tesseract-ocr-ios.

Tesseract* tesseract = [[Tesseract alloc] initWithLanguage:@"eng"];
tesseract.delegate = self;
[tesseract setVariableValue:@"0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijkhlmnopqrstuvwxyz" forKey:@"tessedit_char_whitelist"]; //limit search

                 UIImage* bwImage = [image blackAndWhite];

                 [tesseract setImage:bwImage]; //image to check
                 BOOL result = [tesseract recognize];
                 if (!result) {
                     [self performSelectorOnMainThread:@selector(uiShowMessage:) withObject:NSLocalizedString(@"OCRFailed", @"OCR Failed.") waitUntilDone:YES];
                     return;
                 }
                 if (text == nil || [text compare:@""] == NSOrderedSame) {
                     return;
                 }

from tesseract-ocr-ios.

That looks like it should work okay. I guess as a santiy check you could try using CocoaPods in your project and see if you still have the high memory usage (using the Podfile entry I mentioned above).

I can look at the memory usage for your project if you are okay with emailing me your code (kcon AT stanford DOT edu), or you can use this guide to diagnose it yourself: http://www.raywenderlich.com/23037/how-to-use-instruments-in-xcode

from tesseract-ocr-ios.

ok, thank you.

from tesseract-ocr-ios.

by the way, what is the size of your image?

from tesseract-ocr-ios.

The images I typically recognize in Tesseract are quite small because I require the user to crop what they're interested in to a small rectangle, so my image dimensions are 90 x 70 (width x height). But the Template Framework Project in this repo recognizes on this much larger image (https://raw.githubusercontent.com/gali8/Tesseract-OCR-iOS/master/Template%20Framework%20Project/Template%20Framework%20Project/image_sample.jpg), during which I don't see the kind of memory usage you are reporting.

from tesseract-ocr-ios.

But i got the same result when i launch the TemplateFramework on my iPod,after finishing recognize, the memory usage will kept at 43M.

from tesseract-ocr-ios.

from tesseract-ocr-ios.

i use the image captured by camera

from tesseract-ocr-ios.

Oh I see, sorry I misread your first post because I thought you were saying the memory was growing by that amount (uncontrollably), but you're just pointing out that it's the memory in use after Tesseract recognizes any image.

I was able to reproduce your result, and although it's unfortunate, I'm afraid this is an artifact of Tesseract in general, so there's nothing we can do about it for this wrapper library. See this issue on the main Tesseract project, where someone reports a similar static memory usage from Valgrind and also where one person comments:

"Some of the dawgs are held statically to minimize the time consumed by deleting and re-creating
apis, and memory consumed running them in parallel from multiple threads. This isn't a real leak, 
and memory actually used will not grow over time as a result."

This matches the results of a profile I ran of the Template Framework project memory usage. See how the function that reads the DAWG data uses 17.78 MB all by itself:

Your first post said the app started at 12 MB and then rose to 38 MB. Well 38 - 12 = 16 which is about 17.78 MB, so I think this explains the issue.

from tesseract-ocr-ios.

One other thing worth mentioning is that the size of the DAWG is related to the specific language file you are using.

I'm assuming we both used "eng" in our tests, but I can confirm that using a custom language file (with less training data) can reduce the static memory usage. In my test just now, my custom language file (for a custom font I am recognizing) used 30 MB less memory than the "eng" language file.

from tesseract-ocr-ios.

how to make the custom language file? is there any tutorial?

from tesseract-ocr-ios.

The upstream project will be able to help you out with custom language files:

• Existing language files
• Training Tesseract

from tesseract-ocr-ios.

You may also find this tutorial useful if the font you want to train on is a font you have installed on your computer (so you can create the training document in Microsoft Word): http://michaeljaylissner.com/posts/2012/02/11/adding-new-fonts-to-tesseract-3-ocr-engine/

If the font is not one you can install on your computer, then you have to basically make the image of the training characters yourself, whether that means taking pictures of the characters or drawing the characters yourself in a drawing program.

from tesseract-ocr-ios.

I'm tagging this as "wontfix", although really it should be "can'tfix" since this is just how the upstream Tesseract library works.

from tesseract-ocr-ios.