Why Tesseract OCR library (iOS) cannot recognize text at all?

Question

I'm trying to use Tesseract OCR library in my iOS application. I downloaded tesseract-ios library from github and when I tried to recognize a simple text image I got garbage instead. Here is an image of what I tried to recognize:

enter image description here

I got unreadable text:

T0I1101T0W KIR1 H1I1101T0W KIR1 H1I1101T0W CIBEPS H1 ES PBHY P306 EHH11 133I R1 11335 11I1H1 19 13S SYIL 3B19 M H300H1911 H1113 AIR1 J1 OIII 3I9SH5H133IS 13V9 I1 Q1H211 E015 19 W331 H1 111SW

Why Tesseract can't recognise even simple image? Here is code which I used to instantiate Tesseract:

Tesseract* tesseractObject = [[Tesseract alloc] initWithDataPath:@"tessdata" language:@"eng"];
[tesseractObject setVariableValue:@"0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ" forKey:@"tessedit_char_whitelist"];
[tesseractObject setImage:image];
[tesseractObject recognize];
NSLog(@"RECOGNISED= %@" , [tesseractObject recognizedText]);

Here is my project structure:

enter image description here

I added English testdata folder by reference. So what am I doing wrong? How can I fix this?

do you have all the eng tessdata files included as it looks like your language files are not working as it is scanning and getting letter shapes — Adam Richardson, Jun 18 '13 at 13:22
Where do you get the image from, is it part of your app bundle, downloaded from a website by the app or is it from the camera? — Adam Richardson, Jun 18 '13 at 13:32
@AdamRichardson I use images from camera and images from the app bundle.Result still the same — MainstreamDeveloper00, Jun 18 '13 at 13:34
Have you tried it without [tesseractObject setVariableValue:@"0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ" forKey:@"tessedit_char_whitelist"]; as that is something that I dont have in my code — Adam Richardson, Jun 18 '13 at 13:40
@AdamRichardson Thanks man:)It really helped. I removed this and all text was recognized precisely. But it really strange...Restricting characters to a specific set must only improve recognizing accuracy.So now i don't understand things at all. — MainstreamDeveloper00, Jun 18 '13 at 13:45
glad it worked it is really odd as restricting characters would seem the logical thing to do — Adam Richardson, Jun 18 '13 at 13:47
you may find this thread useful as well as it is about improving the accuracy of OCR from images taken with the camera http://stackoverflow.com/questions/13511102/ios-tesseract-ocr-image-preperation/15528047#15528047 — Adam Richardson, Jun 18 '13 at 13:48
Perhaps you should try again, but this time include lower case letters in the whitelist... — Anders Johansen, Aug 12 '13 at 17:03
how to all language pack use in my application but more than one size for language package , how to manage this one issue , if any now know please let me know.. — Ramani Hitesh, Jan 02 '19 at 07:03

Médéric Petit · Answer 1 · 2013-12-16T09:52:26.437

You are using the option tessedit_char_whitelist with the value "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ" which limits the character recognition to this list only. However the image that you want to process contains lower case characters, if you want to use this option you will have to include lower cases char too.

[tesseractObject setVariableValue:@"0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ" forKey:@"tessedit_char_whitelist"];

score 19 · Accepted Answer · answered Jun 18 '13 at 13:25

19

Make sure you have the latest tessdata file from Google code

http://code.google.com/p/tesseract-ocr/downloads/list

This will provide you with a list of tessdata files that you need to download and include in your app if you haven't already. In your case you will need tesseract-ocr-3.02.eng.tar.gz as you are looking for the English language files

The following article will show you where you need to install it. I read through this tutorial when I built my first Tesseract project and found it really useful

http://lois.di-qual.net/blog/install-and-use-tesseract-on-ios-with-tesseract-ios/

answered Jun 18 '13 at 13:25

Adam Richardson

2,518
1
27
31

I used this tutorial when configured Tesseract. So don't you have some ideas how to solve this? I really need a help.Thanks – MainstreamDeveloper00 Jun 18 '13 at 13:37
You should use pod so pod will take care of any files that need to be updated. – GeneCode Oct 31 '16 at 08:20

score 12 · Answer 3 · answered Jun 22 '13 at 06:00

12

Like Adam said, if you want good results, you'll have to do some image processing and configure some settings (white-listing certain characters, etc).

For anyone else stumbling upon this question, I've put together a sample project here that does some white-listing and image processing:https://github.com/mstrchrstphr/OCR-iOS-Example

answered Jun 22 '13 at 06:00

Christopher

193
7

@user3823935 in build settings under architectures remove the default value and add armv7 and armv7s. this lib was'nt build against arm64 which is probably why you get the errors – Boaz Saragossi Dec 30 '14 at 10:28
if linker error then use "TesseractOCR.framework" use – Ramani Hitesh Dec 21 '18 at 11:44

score 0 · Answer 4 · answered Nov 08 '16 at 08:29

and my output is

Solution :

 tesseract.language = @"eng+fra";

tesseract.pageSegmentationMode = G8PageSegmentationModeAuto;
tesseract.engineMode  = G8OCREngineModeTesseractCubeCombined;
tesseract.image = [image.image g8_blackAndWhite];

tesseract.maximumRecognitionTime = 60.0;
[tesseract recognize];

NSLog(@"%@", tesseract.recognizedText);

reco_area.text = [tesseract recognizedText];

for tessdata click here

Sujatha Girijala · Answer 5 · 2016-12-21T07:26:25.253

whatever @ Adam Richardson explained is correct along with that add this 1) scaleimage method for increase size of the image(dimensions increase)

func scaleImage(image: UIImage, maxDimension: CGFloat) -> UIImage {

    var scaledSize = CGSize(width: maxDimension, height: maxDimension)
    var scaleFactor: CGFloat

    if image.size.width > image.size.height {
        scaleFactor = image.size.height / image.size.width
        scaledSize.width = maxDimension
        scaledSize.height = scaledSize.width * scaleFactor
    } else {
        scaleFactor = image.size.width / image.size.height
        scaledSize.height = maxDimension
        scaledSize.width = scaledSize.height * scaleFactor
    }

    UIGraphicsBeginImageContext(scaledSize)
    image.draw(in: CGRect(x: 0, y: 0, width: scaledSize.width, height: scaledSize.height))
    let scaledImage = UIGraphicsGetImageFromCurrentImageContext()
    UIGraphicsEndImageContext()

    return scaledImage!
}

2) store this eng.traineddata language file in filemanager

 func storeLanguageFile() throws{
    var fileManager: FileManager = FileManager.default
    let nsDocumentDirectory = FileManager.SearchPathDirectory.documentDirectory
    let nsUserDomainMask = FileManager.SearchPathDomainMask.userDomainMask
    let docDirectory = NSSearchPathForDirectoriesInDomains(nsDocumentDirectory, nsUserDomainMask, true)[0] as NSString
    let path: String = docDirectory.appendingPathComponent("/tessdata/eng.traineddata")
    if fileManager.fileExists(atPath: path){
        var data: NSData = NSData.dataWithContentsOfMappedFile((Bundle.main.resourcePath?.appending("/tessdata/eng.traineddata"))!)! as! NSData
        var error: NSError
        try FileManager.default.createDirectory(atPath: docDirectory.appendingPathComponent("/tessdata"), withIntermediateDirectories: true, attributes: nil)
        data.write(toFile: path, atomically: true)
    }
}

3) after that you can use https://github.com/BradLarson/GPUImage for increase clarity of the image

you can use this

func preprocessedImage(for tesseract: G8Tesseract!, sourceImage: UIImage!) -> UIImage! {
    var stillImageFilter: GPUImageAdaptiveThresholdFilter = GPUImageAdaptiveThresholdFilter()
    stillImageFilter.blurRadiusInPixels = 4.0
    var filterImage: UIImage = stillImageFilter.image(byFilteringImage: sourceImage)
    return filterImage
}

these 3 steps will help you to increase the accuracy of the tesseract upto 60 ~ 70 %

I got 70% accuracy text result from tesseract by using above 3 steps. But, I want to get text detection like address, firstName, lastName from this text — Sujatha Girijala, Dec 21 '16 at 07:01

Why Tesseract OCR library (iOS) cannot recognize text at all?

5 Answers5

Linked