Develop your own text recognition application with Tesseract

前言

这次和⼤Chat at home⽂Word recognition related topics.⼤At home, there must be various types of scanning APP 不陌⽣.拿着⼿camera facing any⽂字,directly into the camera⽂Word content is converted into ⼿Onboard editable strings.

完整的 OCR 算法,并不简单.It involves image recognition algorithms,as well as from Fig⽚identified in⽂Words can be converted into our program⽤的⼆进制形式.This also includes different languages⾔how to recognize characters,How to more accurately identify and so on.

设想⼀下,如果作为开发者,让你⾃⼰Develop this from scratch⼀套算法,I'm afraid it will cost a lot⼤⼒⽓.不过好消息是,我们⽣Live in this age of open source,There are many seniors who have already made the way for us.That's what we're going to talk about today, Tesseract就是这样⼀个开源库,给定任意⼀张图⽚,它可以识别出⾥⾯所有的⽂字内容,并且 API 接⼝使⽤⾮常简单.

Tesseract

Tesseract 是 Google 发布的⼀款 OCR 开源库,它⽀Holds multiple idioms⾔环境,以及运⾏环境,Including us⾥将要介绍的 iOS 环境.今天就⽤It leads⼤家开发⼀one belongs to you⾃⼰的 OCR 应⽤.

1.安装 Tesseract

最简单的⽅式是⽤过 Cocoapods,进⼊你的项⽬根⽬录,输⼊：

pod init

然后,编辑⽣成的 Podfile 配置⽂件：

target 'ocrSamples' do
	use_frameworks!
	pod 'TesseractOCRiOS'
	
end

将TesseractOCRiOS加⼊配置列表中,最后输⼊：

pod install

2,.配置⼯作

安装完成后,我们还需要进⾏⼀Simple configuration below,⾸先要在 Build Phases -> Link Binary With Libraries 中配置,项⽬需要的依赖库, CoreImage, libstdc++, 以及 TesseractOCRiOS ⾃⾝：

After configuring the dependency library,我们还需要将⽂Word recognition training data添加进来,What is the training data,TesseractOCRiOS识别图⽚的时候,It will be recognized according to the rules of this training data⽂字.⽐如中⽂,英⽂等,都有对应的训练数据,This can be understood as deep learning pre-trained models for us.⽤它来进⾏核⼼的识别算法.

Training data is required to follow不同语言来区分的,Tesseract There is a dedicated page listing all available training data：https://github.com/tesseract-ocr/tessdata/tree/3.04.00

For example, we need to recognize Simplified Chinese here,就可以下载 chi-sim this training data：

Then drag and drop the training data into ⼯程中：

上图中的 chi_sim.traineddata is the training data we downloaded,这⾥⾯有⼀The point to note is that,这个⽂item must be present testdata 这个⽂件夹中.并且这个⽂folder to start with "Create Folder Reference" 的⽅drag and drop in：

This citation⽤⽅formula and other⼀种 "Create Groups" 的⽅What's the difference？

主要区别在于:

• 使⽤引⽤⽅式Drag and drop in⽬录,在最后⽣成 APP 包的时候, Main Bunlde 中是以 testdata/chi_sim.traineddata This path form saves our training data resources
• 使⽤ "Create Groups" ⽅式,It will be when it is finally stored忽略⽂Folder name的,最后存储在 Main Bundle 中的⽂件是以/chi_sim.traineddata stored in this path.

注意: ⽽ TesseractOCRiOS,By default it will be there testdata/chi_sim.traineddata This path looks for training data,所以如果使⽤"Create Groups" ⽅drag⼊,will cause luck⾏No training data was found,⽽报错.This detail requires special attention.

最后, 为了让 TesseractOCRiOS can be shipped correctly⾏,We also need to turn off BitCode,否则会报编译错误,We need to turn it off in both places,⼀个是⾄⼯程,另外⼀个是 Pods 模块,如下图：

3. 开始编码

Complete the above preparations⼯作后,我们就可以开始编码了,这⾥只给⼤Home Show⽰最精简的代码.⾸First we need to be in the main world⾯上显⽰两个控件,⼀One is our pre-stored band⽂photo of words⽚,另外⼀个是⽤于展⽰识别结果的⽂本框：

override func viewDidLoad() {
    
	super.viewDidLoad()
	self.imageView = UIImageView(frame: CGRect(x: 0, y: 80, width: self.view.frame.size.width, height:
	self.view.frame.size.width * 0.7))
	self.imageView?.image = UIImage(named: "textimg")
	self.view.addSubview( self.imageView!)
	self.textView = UITextView(frame: CGRect(x: 0, y: labelResult.frame.origin.y + labelResult.frame.size.height + 20, width: self.view.frame.size.width, height: 200))
	self.view.addSubview( self.textView!)
}

这⾥⾯We only write out the key code for initialization of the two controls,Other unimportant codes are omitted.Then you can tune⽤ TesseractOCRiOS 来进⾏⽂word recognized：

func recognizeText(image: UIImage) {
    
	if let tesseract= G8Tesseract(language: "chi_sim") {
    
		tesseract. engineMode= .tesseractOnly
		tesseract. pageSegmentationMode= .auto
		tesseract. image= image
		tesseract.recognize()
		self.textView?. text= tesseract.recognizedText
	}
}

All identification related code is here⾥了.⾸先调⽤ G8Tesseract 进⾏初始化,我们传⼊The name of the training data,这⾥是 "chi_sim"Represents Simplified Chinese⽂.

engineModeThere are three modes to choose from, tesseractOnly,tesseractCubeCombined 和 cubeOnly.我们使⽤第⼀种模式,采⽤训练数据的⽅式. cubeOnly Consciousness is to make⽤更精准的 cube ⽅式, tesseractCubeCombined It is a combination of the two modes⽤.

cube Schema requires additional model data,⼤The example is this：

The picture above is in English⽂的 cube 识别模型.简体中⽂I haven't found the model yet,So we just in this instance⽤到了 tesseractOnly 模式.并且在没有 cube in the case of model data,我们是不⽤使⽤ tesseractCubeCombined 和 cubeOnly 的,Otherwise it will be because the model data does not exist⽽报错.⼤Home if you can find it⽂的 cube 模型,You are also welcome to stay⾔中反馈,This will make this⽂Word recognition is more accurate.

other tones⽤No need to say more,will be identified image 对象设置给 Tesseract.然后调⽤它的 recognize ⽅法进⾏识别,Finally the result will be identified tesseract.recognizedText 设置给 TextView.

Final luck⾏效果如下：

上⾯的图⽚I took it SwiftCafe ⽹站上⼀篇⽂Chapter photo⽚,from the identification results,还算⽐较准确.

总结

⼤Home from above⾯The code should also feel it, Tesseract Although provided essentially counts⽐较复杂的⽂Word recognition algorithm,But it provides access to developers⼝可以说得上是⾮常简单. OCR ⽂Word recognition as a whole,也可以算得上是 AI 的⼀个应⽤分⽀,在这个 AI ⼤⾏its time,即便掌握⼀些应⽤技术,It can also be very good for us developers⼤to broaden our horizons.发挥你的创意,也许类似Tesseract These components can help you create AI A newcomer of the times⽤.

Tesseract These components can help you create AI A newcomer of the times⽤.

当然, Tesseract ⽬前⾃⾝还有⼀些缺陷,⽐as it isOnly prints are recognized字体,⽽cannot be accurately identified⼿write font.But that⼜怎么样呢,It would have been quite complicated OCR 算法,可以让我们⽤⾮Often a small price should be paid⽤起来,还是⼀piece to the developer⾮A very happy thing.

照例,The sample project code in this article has been placed Github 中,You can download it directly if you need it：https://github.com/swiftcafex/Samples/tree/master/ocrSamples.