This project implements Apache Tika running as a web service using Spring Boot. It exposes a REST API so that a client can send a document in binary format and receive back the extracted text. The supported document formats are the ones as in Tika. Some of the key motivation behind developing own wrapper over Tika instead of using the already available Tika server is a better control over used document parsers (such as PDFParser, Tesseract OCR and the legacy one taken from CogStack-Pipeline) and control over returned results with HTTP return codes.

Feel free to ask questions on the github issue tracker or on our discourse website which is frequently used by our development team!

To build the application, run in the main directory:

./gradlew build --console=plain

The build artifacts will be placed in ./build directory.

During the build, the tests will be run, where the failed tests can also signify missing third-party dependencies (see below). However, to skip running the tests and just build the application, one can run:

./gradlew build -x test

To run the available tests, run:

./gradlew test --console=plain.

Please note that failed tests may signify missing third-party dependencies.

In the minimal setup, for proper text extraction Apache Tika requires the following applications to be present on the system:

• Tesseract OCR,
• ImageMagick,
• Ghostscript (required by ImageMagick for documents conversion).

ImageMagick also requires its configuration file policy.xml to be overriden by the provided extras/ImageMagick/policy.xml (in order to increase the available resources for file processing and to override security policy related with Ghostscript).

Moreover, in order to enable additional image processing capabilities of Tesseract OCR, few other dependencies need to be present in the system, such as Python environment. Please see the provided Dockerfile for the full list.

The application can be either run as a standalone Java application or inside a Docker container. The application configuration can be changed in the application.yaml file. The default version of configuration file is embeded in the jar file, but can be specified manually (see below).

Please note that the recommended way is to use the provided Docker image since a number of dependencies need to be satisfied on a local machine.

Assuming that the build went correctly, to run the Tika service on a local machine:

java -jar build/jar/service-*.jar

The running service will be listening on port 8090 (by default) on the host machine.

The latest stable Docker image is available in the Docker Hub under cogstacksystems/tika-service:latest tag. Alternatively, the latest development version is available under cogstacksystems/tika-service:dev-latest tag. The image can be also build locally using the provided Dockerfile.

To run Tika service container:

docker run -p 8090:8090 cogstacksystems/tika-service:latest

The service will be listening on port 8090 on the host machine.

The application.yaml now has SSL mod enable by default.

The .p12 certificate is taken directly from the NiFi repo

When processing large documents (and in large amounts) it is important to benchmark the performance depending on the use case. An important setting which is enabled by default is the Tesseract thread limit option which is suitable for deployments where only small-sized documents are to be processed, this is because threading in such cases can do more harm than good since the program will have to start up and manage threads for small jobs most of the time.

By default, this setting is found in the '/docker/docker_compose.yml' file, enabled as a global variable:

  environment:
    OMP_THREAD_LIMIT=1

For more information: https://github.com/tesseract-ocr/tesseract/blob/7c3ac569f9c320bdc4bacea0ec66c69e2cf06a32/doc/tesseract.1.asc#environment-variables AND https://ocrmypdf.readthedocs.io/en/latest/advanced.html#overriding-default-tesseract

Tika Service, by default, will be listening on port 8090 and the returned content extraction result will be represented in JSON format.

The service exposes such endpoints:

• GET /api/info - returns information about the service with its configuration,
• POST /api/process - processes a binary data stream with the binary document content,
• POST /api/process_file - processes a document file (multi-part request).

The extraction results are represented in JSON format where the available main fields are:

• result - the content extraction result with metadata,
• timestamp - the content processing timestamp,
• success - specifies whether the extraction accomplished successfully,
• error - the message in case of processing error (assumes success : false).
• processingElapsedTime - how long did it take to process the document, useful for benchmarking

The content extraction result can contain such fields:

• text - the extracted text,
• metadata - the metadata associated with the document and the used parsers.

The provided metadata associated with the document and the used parsers can include such fields:

• X-Parsed-By - an array of names of the parsers used during the content extraction,
• X-OCR-Applied - a flag specifying whether OCR was applied,
• Content-Type - the content type of the document, as identified by Tika,
• Page-Count - the document page count (extracted from the document metadata by Tika),
• Creation-Date - the document creation date (extracted from the document metadata by Tika).

Additional metadata may be available depending on the file type:

 MS OFFICE metadata tags (for docx, doc, xls, etc. documents):

• meta:comments - doc comments
• meta:last-author - last user who edited the doc
• Category - doc category (if available)
• dc:creator - user that created the doc
• Keywords - doc keywords (if available)
• meta:word-count - word count (built in, it is not done for other doc types)
• meta:character-count - build in char count for MS documents
• Last-Save-Date - datetime at which the doc was last saved
• dcterms:modified - datetime at which the doc was last modified
• extended-properties:Application - application used to create the doc
• extender-properties:Company - company that owns the doc
• dcterms:created - doc creation date
• dc:description - doc description
• dc:identifier - doc identifier
• dc:subject - doc subject

  HTML metadata tags:

• Last-Modified - datetime at which doc was last modified
• Content-Encoding - encoding detected in the original doc

Using curl to send the document to Tika server instance running on localhost on 8090 port:

curl -k -F [email protected] https://localhost:8090/api/process_file | jq

Returned result:

{
  "result": {
    "text": "Sample Type / Medical Specialty: Lab Medicine - Pathology",
    "metadata": {
      "X-Parsed-By": [
        "org.apache.tika.parser.CompositeParser",
        "org.apache.tika.parser.DefaultParser",
        "org.apache.tika.parser.microsoft.ooxml.OOXMLParser"
      ],
      "X-OCR-Applied": "false",
      "Content-Type": "application/vnd.openxmlformats-officedocument.wordprocessingml.document"
    },
    "success": true,
    "timestamp": "2019-08-13T15:14:58.022+01:00"
  }
}

For this feature to work you must set the following use-legacy-tika-processor-as-default: false in application.yaml.

curl -k -F [email protected] -F [email protected] https://localhost:8090/api/process_bulk

Returned result:

{
   "results":[
      {
         "resourceId":"sample.pdf",
         "text":"\n A Simple PDF File \n This is a small demonstration .pdf file - \n\n just for use in the Virtual Mechanics tutorials. More text. And more \n text. And more text. And more text. And more text. \n\n And more text. And more text. And more text. And more text. And more \n text. And more text. Boring, zzzzz. And more text. And more text. And \n more text. And more text. And more text. And more text. And more text. \n And more text. And more text. \n\n And more text. And more text. And more text. And more text. And more \n text. And more text. And more text. Even more. Continued on page 2 ...\n\n\n\n Simple PDF File 2 \n ...continued from page 1. Yet more text. And more text. And more text. \n And more text. And more text. And more text. And more text. And more \n text. Oh, how boring typing this stuff. But not as boring as watching \n paint dry. And more text. And more text. And more text. And more text. \n Boring.  More, a little more text. The end, and just as well. \n\n\n",
         "metadata":{
            "X-Parsed-By":[
               "org.apache.tika.parser.pdf.PDFParser"
            ],
            "X-OCR-Applied":false,
            "Creation-Date":"2006-03-01T07:28:26Z",
            "Content-Type":"application/pdf",
            "Page-Count":2
         },
         "success":true,
         "timestamp":"2021-05-14T11:51:47.867+01:00"
      },
      {
         "resourceId":"sample_1.pdf",
         "text":"\n A Simple PDF File \n This is a small demonstration .pdf file - \n\n just for use in the Virtual Mechanics tutorials. More text. And more \n text. And more text. And more text. And more text. \n\n And more text. And more text. And more text. And more text. And more \n text. And more text. Boring, zzzzz. And more text. And more text. And \n more text. And more text. And more text. And more text. And more text. \n And more text. And more text. \n\n And more text. And more text. And more text. And more text. And more \n text. And more text. And more text. Even more. Continued on page 2 ...\n\n\n\n Simple PDF File 2 \n ...continued from page 1. Yet more text. And more text. And more text. \n And more text. And more text. And more text. And more text. And more \n text. Oh, how boring typing this stuff. But not as boring as watching \n paint dry. And more text. And more text. And more text. And more text. \n Boring.  More, a little more text. The end, and just as well. \n\n\n",
         "metadata":{
            "X-Parsed-By":[
               "org.apache.tika.parser.pdf.PDFParser"
            ],
            "X-OCR-Applied":false,
            "Creation-Date":"2006-03-01T07:28:26Z",
            "Content-Type":"application/pdf",
            "Page-Count":2
         },
         "success":true,
         "timestamp":"2021-05-14T11:51:47.871+01:00"
      }
   ]
}

All the available service and document processors parameters are stored in a single src/main/resources/application.yaml file.

Although the initial configuration file is bundled with the application jar file, a modified one can be provided as a parameter when running the Java application. For example, when running the Tika service in the Docker container, the script scripts/run.sh runs the Tika service with custom configuration file application.yaml located in /app/config/ directory: java -Dspring.config.location=/app/config/ -jar /app/service-*.jar

The configuration file is stored in yaml format with the following available properties.

• application.version - specifies the application version,
• server.port - the port number on which the service will be run (default: 8090),
• spring.servlet.multipart.max-file-size and spring.servlet.multipart.max-request-size - specifies the max file size when processing file requests (default: 100MB).

The following keys reside under tika.processing node:

• use-legacy-tika-processor-as-default - whether to use the legacy Tika PDF parser (as used in CogStack Pipeline) for backward compatibility (default: true),
• fail-on-empty-files - whether to fail the request and report an error when client provided an empty document (default: false),
• fail-on-non-document-types - whether to fail the request and report an error when client provided a not supported and/or non-document content (default: true).

The following keys reside under tika.parsers node.

The keys under tesseract-ocr define the default behavior of the Tika Tesseract OCR parser:

• language - the language dictionary used by Tesseract (default: eng),
• timeout - the max time (ms) to process documents before reporting error (default: 300),
• enable-image-processing - whether to use additional pre-processing of the images using ImageMagick (default: true), disable this option of if the files that will be processed do not contain any images.
• apply-rotation - whether to apply de-rotating of the images (default: false), Please note that enabling enable-image-processing and/or apply-rotation although might improve the quality of the extracted text can significantly slower the extraction process.
• depth - recursion depth before passing to the next parser, default is 16, has to be a power of 2
• filter - smoothing filter applied to the images, 'triangle' is fastest
• resize - image resizing, impacts performance a lot at the cost of extracted content quality (high values=better,slow, default: 100), must be set in increments of 100 otherwise it will error
• density - DPI at which the images detected are represented, lower values provide faster processing, but possible quality loss (default:150)

The keys under pdf-ocr-parser define the default behavior of the PDF parser that uses Tesseract OCR to extract the text:

• ocr-only-strategy - whether to use only OCR or to apply additional text extraction from the content (default: true),

• min-doc-text-length - if the available text in the document (before applying OCR) is higher than this value then skip OCR (default: 100),

• min-doc-byte-size - the minimum size of the image data (in bytes) that should have the content to be extracted, otherwise is skipped (default: 10000),

• use-legacy-ocr-parser-for-single-page-doc - in case of single-page PDF documents, whether to use the legacy parser (default: false).

• ocr-detect-angles - used in conjunction with apply rotation, so if apply-rotation is not used this should be false (default: false)

• ocr-dpi - a bit ambiguous as this is also provided in the tesseract-ocr-parser, DPI used to render the image (default: 150)

• ocr-rendering-strategy, values: "ALL" -render everything (images and text), "NO_TEXT", "TEXT_ONLY", "VECTOR_GRAPHICS_ONLY" - only vector images are rendered

values: "NONE", "RAW_IMAGES", "RENDER_PAGES_BEFORE_PARSE", "RENDER_PAGES_AT_PAGE_END"

• ocr-image-strategy - how do are the images rendered, raw_images renders them on the go, while the others are before parse and at the end

• ocr-strategy - this is the value that affects if OCR is performed at ALL, options:
  • "AUTO"
  • "NO_OCR" - exactly what it says, it WONT do any OCR
  • "OCR_AND_TEXT_EXTRACTION" - OCR and text extraction, WARNING, this will result in DUPLICATED TEXT
  • "OCR_ONLY" - only do OCR, without text extraction (this is done separately)

With the above setting you can speed up the workflows of certain processes, if you know you DO NOT need to OCR images, then it makes total sense to disable this as it will improve the service's speed by a substantial amount, resulting in near instant results. OCR should definitely be disabled for use with process_bulk.

Please check the application.yaml file for range limits on these settings.

The keys under legacy-pdf-parser define the behavior of the Tika PDF parser used in CogStack Pipeline (the 'legacy' parser), that is used for backward compatibility:

• image-magick.timeout - the max timeout value (in ms) when performing document conversion using ImageMagick (default: 1200),

• tesseract-ocr.timeout - the max timeout value (in ms) when performing text extraction using Tesseract OCR (default: 1200),

• min-doc-text-length - if the available text in the document (before applying OCR) is higher than this value then skip OCR (default: 10). if you have documents containing only embedded images this should probably be set to 1 (if nothing is processed, because if the document contains no text and just images it might be skipped)

Post ocr settings to force ENCODING, post-ocr subsection of tika:

• output-encoding - default "UTF-8", this is what we most likely will need in 99% of the cases that handle the English language.
• enforce-encoding-output - default "false", if this is set to true, the output text will be UTF-8 (or the value of output-encoding) compliant, all other chars are removed, there may be problems when enabling this option in some cases, please ensure the output is correct and the docs can be processed, if not then it is best to leave this option off.

The service is able to extract free-text from most known file types (PDF, docx, tiff, png, html etc.), please make sure to enable the enable-image-processing setting.

HTML files will have all text content extracted, images will only be OCRed if they are embedded within the doc (for example in base64 format).