Spark OCR 2.3.x (Licensed)

Spark NLP comes with an OCR module that can read both PDF files and scanned images (requires Tesseract 4.x+).

Installation

Installing Tesseract

As mentioned above, if you are dealing with scanned images instead of test-selectable PDF files you need to install tesseract 4.x+ on all the nodes in your cluster. Here how you can install it on Ubuntu/Debian:

apt-get install tesseract-ocr

In Databricks this command may result in installing tesseract 3.x instead of version 4.x.

You can simply run this init script to install tesseract 4.x in your Databricks cluster:

#!/bin/bash
sudo apt-get install -y g++ # or clang++ (presumably)
sudo apt-get install -y autoconf automake libtool
sudo apt-get install -y pkg-config
sudo apt-get install -y libpng-dev
sudo apt-get install -y libjpeg8-dev
sudo apt-get install -y libtiff5-dev
sudo apt-get install -y zlib1g-dev
​
wget http://www.leptonica.org/source/leptonica-1.74.4.tar.gz
tar xvf leptonica-1.74.4.tar.gz
cd leptonica-1.74.4
./configure
make
sudo make install
​
git clone --single-branch --branch 4.1 https://github.com/tesseract-ocr/tesseract.git
cd tesseract
./autogen.sh
./configure
make
sudo make install
sudo ldconfig
​
tesseract -v

Quick start

Let’s read a PDF file:

import com.johnsnowlabs.nlp._
val ocrHelper = new OcrHelper()

//If you do this locally you can use file:/// or hdfs:/// if the files are hosted in Hadoop
val dataset = ocrHelper.createDataset(spark, "/tmp/sample_article.pdf")

If you are trying to extract text from scanned images in the format of PDF, please keep in mind to use these configs:

ocrHelper.setPreferredMethod("image")
ocrHelper.setFallbackMethod(false)
ocrHelper.setMinSizeBeforeFallback(0)

Configuration

setPreferredMethod(text/image = text) either text or image will work. Defaults to text. Text mode works better and faster for digital or text scanned PDFs
setFallbackMethod(boolean) on true, when text or image fail, it will fallback to the alternate method
setMinSizeBeforeFallback(int = 1) number of characters to have at a minimum, before falling back.
setPageSegMode(int = 3) image mode page segmentation mode
setEngineMode(int = 1) image mode engine mode
setPageIteratorLevel(int = 0) image mode page iteratior level
setScalingFactor(float) Specifies the scaling factor to apply to images, in both axes, before OCR. It can scale up the image(factor > 1.0) or scale it down(factor < 1.0)
setSplitPages(boolean = true) Whether to split pages into different rows and documents
setSplitRegions(boolean = true) Whether to split by document regions. Works only in image mode. Enables split pages as well.
setIncludeConfidence(boolean = false)
setAutomaticSkewCorrection(use: boolean, half_angle: double = 5.0, resolution: double = 1.0)
setAutomaticSizeCorrection(use: boolean, desired_size: int = 34)
setEstimateNoise(string) image mode estimator noise level
useErosion(use: boolean, kernel_size: int = 2, kernel_shape: Int = 0) image mode erosion

Utilizing Spark NLP OCR Module

Spark NLP OCR Module is not included within Spark NLP. It is not an annotator and not an extension to Spark ML.

You can use OcrHelper to directly create spark dataframes from PDF. This will hold entire documents in single rows, meant to be later processed by a SentenceDetector. This way, you won’t be breaking the content in rows as if you were reading a standard document. Metadata columns are added automatically and will include page numbers, file name and other useful information per row.

Python code

from pyspark.sql import SparkSession
from sparknlp.ocr import OcrHelper
from sparknlp import DocumentAssembler

data = OcrHelper().createDataset(spark = spark, input_path = "/your/example.pdf" )
documentAssembler = DocumentAssembler().setInputCol("text")
annotations = documentAssembler.transform(data)
annotations.columns

['text', 'pagenum', 'method', 'noiselevel', 'confidence', 'positions',
 'filename', 'document']

Scala code

import com.johnsnowlabs.nlp.util.io.OcrHelper
import com.johnsnowlabs.nlp.DocumentAssembler

val myOcrHelper = new OcrHelper
val data = myOcrHelper.createDataset(spark, "/your/example.pdf")
val documentAssembler = new DocumentAssembler().setInputCol("text")
val annotations = documentAssembler.transform(data)
annotations.columns

Array[String] = Array(text, pagenum, method, noiselevel, confidence, positions, filename, document)

… where the text column of the annotations spark dataframe includes the text content of the PDF, pagenum the page number, etc…

Creating an Array of Strings from PDF (For LightPipeline)

Another way, would be to simply create an array of strings. This is useful for example if you are parsing a small amount of pdf files and would like to use LightPipelines instead. See an example below.

Scala code

import com.johnsnowlabs.nlp.util.io.OcrHelper
import com.johnsnowlabs.nlp.{DocumentAssembler,LightPipeline}
import com.johnsnowlabs.nlp.annotator.SentenceDetector
import org.apache.spark.ml.Pipeline

val myOcrHelper = new OcrHelper
val raw = myOcrHelper.createMap("/pdfs/")
val documentAssembler = new DocumentAssembler().setInputCol("text").setOutputCol("document")
val sentenceDetector = new SentenceDetector().setInputCols("document").setOutputCol("sentence")
val lightPipeline = new LightPipeline(new Pipeline().setStages(Array(documentAssembler, sentenceDetector)).fit(Seq.empty[String].toDF("text")))
val annotations = ligthPipeline.annotate(raw.values.toArray)

Now to get the whole first PDF content in your /pdfs/ folder you can use:

annotations(0)("document")(0)

and to get the third sentence found in that first pdf:

annotations(0)("sentence")(2)

To get from the fifth pdf the second sentence:

annotations(4)("sentence")(1)

Similarly, the whole content of the fifth pdf can be retrieved by:

annotations(4)("document")(0)

PREVIOUSRelease notes