OCR models overview

This page gives you an overview of every OCR model in NLU which are provided by Spark OCR.

Additionally you can refer to the OCR tutorial Notebooks

• OCR Tutorial for extracting Text from Image/PDF/DOC(X) files
• OCR Tutorial for extracting Tables from Image/PDF/DOC(X) files

Overview of all OCR features

Overview of OCR Text Extractors
These models grab the text directly from your input file and returns it as a Pandas DataFrame

Overview of OCR Table Extractors
These models grab all Table data from the files detected and return a list of Pandas DataFrames, containing Pandas DataFrame for every table detected

File Path handling for OCR Models

When your nlu pipeline contains a ocr spell the predict method will accept the following inputs :

• a string pointing to a folder or to a file
• a list, numpy array or Pandas Series containing paths pointing to folders or files
• a Pandas Dataframe or Spark Dataframe containing a column named path which has one path entry per row pointing to folders or files

For every path in the input passed to the predict() method, nlu will distinguish between two cases:

1. If the path points to a file, nlu will apply OCR transformers to it, if the file type is processable with the currently loaded OCR pipeline.
2. If the path points to a folder, nlu will recursively search for files in the folder and sub-folders which have file types which are applicable with the loaded OCR pipeline.

NLU checks the file endings to determine whether the OCR models can be applied or not, i.e. .pdf, .img etc.. If your files lack these endings, NLU will not process them.

Image to Text

Sample image:

nlu.load('img2text').predict('path/to/haiku.png')

Output of IMG OCR:

PDF to Text

Sample PDF:

nlu.load('pdf2text').predict('path/to/haiku.pdf')

Output of PDF OCR:

DOCX to text

Sample DOCX:

nlu.load('doc2text').predict('path/to/haiku.docx')

Output of DOCX OCR:

PDF with Tables

Sample PDF:

nlu.load('pdf2table').predict('/path/to/sample.pdf')

Output of PDF Table OCR :

DOCX with Tables

Sample DOCX:

nlu.load('doc2table').predict('/path/to/sample.docx')

Output of DOCX Table OCR :

PPT with Tables

Sample PPT with two tables:

nlu.load('ppt2table').predict('/path/to/sample.docx')

Output of PPT Table OCR :

and

Combine OCR and NLP models

Sample image containing named entities from U.S. Presidents Wikipedia:

nlu.load('img2text ner').predict('path/to/presidents.png')

Output of image OCR and NER NLP :

Authorize NLU for OCR

You need a set of credentials to access the licensed OCR features. You can grab one here

Authorize anywhere via providing via JSON file

If you provide a JSON file with credentials, nlu will check whether there are only OCR or also Healthcare secrets. If both are contained in the JSON file, nlu will give you access to healthcare and OCR features, if only one of them is present you will be accordingly only authorized for one set of the features. You can specify the location of your secrets.json like this :

path = '/path/to/secrets.json'
nlu.auth(path).load('licensed_model').predict(data)

Authorize via providing String parameters

You can manually enter your secrets and authorize nlu for OCR and Healthcare features

import nlu
AWS_ACCESS_KEY_ID = 'YOUR_SECRETS'
AWS_SECRET_ACCESS_KEY = 'cgsHeZR+YOUR_SECRETS'
OCR_SECRET = 'YOUR_SECRETS'
JSL_SECRET = 'YOUR_SECRETS'
OCR_LICENSE = "YOUR_SECRETS"
SPARK_NLP_LICENSE = 'YOUR_SECRETS'
# this will automatically install the OCR library and NLP Healthcare library when credentials are provided
nlu.auth(SPARK_NLP_LICENSE,AWS_ACCESS_KEY_ID,AWS_SECRET_ACCESS_KEY,JSL_SECRET, OCR_LICENSE, OCR_SECRET)