Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/infosys/document-extraction-libraries
A suite of libraries to extract information from documents and build RAG-based solutions for semantic search and Q&A.
https://github.com/infosys/document-extraction-libraries
Last synced: about 23 hours ago
JSON representation
A suite of libraries to extract information from documents and build RAG-based solutions for semantic search and Q&A.
- Host: GitHub
- URL: https://github.com/infosys/document-extraction-libraries
- Owner: Infosys
- License: apache-2.0
- Created: 2022-02-02T12:05:26.000Z (almost 3 years ago)
- Default Branch: main
- Last Pushed: 2024-10-31T10:45:16.000Z (18 days ago)
- Last Synced: 2024-10-31T11:28:37.834Z (18 days ago)
- Language: Python
- Homepage:
- Size: 34.8 MB
- Stars: 9
- Watchers: 3
- Forks: 4
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- License: LICENSE
Awesome Lists containing this project
README
# Document Extraction Libraries 3.9.0
Document Extraction Libraries are a suite of python/java libraries that provide APIs to **extract information from documents** (e.g. scanned/native PDFs, images etc.). For semi-structured documents (e.g. form-like documents), this can be done in a simple and predictable manner. For unstructured documents, it can extract the raw content, retrieve relevant text from it using **semantic search** and use a **Large Language Model (LLM)** to extract information.
The suite consists of libraries that can be used to **generate** OCR files using free/commercial tools; **parse** OCR files to extract regions of interest; and **extract** texts and selection field values from the regions of interest. It can **extract segments** from pages, **generate chunks**, **generate embeddings**, and save them to a vector DB which can then be **retrieved as context** and given to LLM to extract information from documents.
Additionally, it includes a data pipeline framework - **Document Processor Platform (DPP)** - for creating **reusable components** and **configurable pipelines**. The pipeline brings all the libraries together to create a logical workflow.
These libraries can be used as SDKs to solve document digitization problems and help with semantic search and information extraction requirements.
## Prerequisites
- Python >=3.10 or <=3.11
- Java >=8
- OCR Tool (Tesseract / Azure Read OCR V 3.2) _(for non-digital documents)_
## Libraries
The details of each library and its core functionality is given below. For more details, please read the [docs](docs).
S# | Library | Description |
---|-------|---------------|
1 | infy_ocr_generator| Provides APIs to generate OCR files by specifying an OCR provider.
2 | infy_ocr_parser| Provides APIs to parse OCR files and detect regions of interest (bounding boxes) when given a search criteria.
3| infy_field_extractor|Provides APIs for extracting free text and selection fields (checkboxes and radio buttons) from image files using regions of interest (bounding boxes) as input.
4| infy_table_extractor|Provides APIs to extract rows and columns from an image of a table.
5| infy_common_utils | Provides APIs to invoke external tools like JAR files.
6| infy_fs_utils | Provides APIs to abstract underlying file system and object stores.
7| infy_gen_ai_sdk | Provides APIs for using embeddings, Large Language Models (LLM), vector DB etc.
8| InfyFormatConverterJAR | Provides APIs to convert documents from one format to another. E.g., PDF to image, JSON etc.
9| InfyOcrEngineJAR | Provides APIs to invoke OCR engines. Currently, it has tesseract.
10| infy_dpp_sdk|The sdk for document processor platform (DPP) containing the interfaces for processors, schema definition for document data and in-built orchestrators to execute a data pipeline made from processors.
11| infy_dpp_core|A collection of processors for core tasks like request creation, meta-data extraction etc.
12| infy_dpp_segmentation|A collection of processors for tasks like document segmentation, chunk creation etc.
13| infy_dpp_ai | A collection of processors for tasks like generating embeddings, calling LLMs with prompt templates etc.
14 | infy_dpp_storage | A collection of processors to help store data to graph DB etc.
15 | infy_dpp_content_extractor | A collection of processors for extracting raw contents from documents.
## Apps
The details of each app and its core functionality is given below.
S# | App | Description |
---|-------|---------------|
1 | infy_dpp_processor| It is implementation of indexing pipeline. This apps can be deployed as docker image on Kubernetes cluster and managed via an orchestrator like Airflow or Kubeflow to run the indexing pipeline.
2 | infy_db_service| There are two ways to store the created indexes, one is locally in the environment where indexing pipeline is running, and the other one is using infy_db_service which provides way to store indexes in a central environment where this service is hosted.
3 | infy_search_service| This is implementation of inferencing pipeline.If infy_db_service is used to store created indexes then use infy_search_service to query on those documents.
## How does it Work
### Semi-structured documents
The libraries use **computer vision** in the form of an **OCR engine** (e.g. Tesseract, Azure OCR Read etc.) for positional text detection. It then takes a "region definition" as input and applies techniques to detect regions of interest within the document.
This makes it possible to extract attributes -**free text, selection fields (checkboxes, radio buttons) and bordered tables** - specifically from the regions of interest and eliminates the risk of potential errors in future should the document layout change but not the regions of interest.
### Unstructured documents
The libraries along with the data pipeline framework help create a workflow where raw content is extracted from documents and stored in a vector DB as embeddings. From this, useful information is extracted using the Retrieval augmented generation (RAG) approach.
The API logical input/output is given below.
Step |Library | Input | Output
---|---|---|---|
1 | infy_ocr_generator | `image file` | `OCR file`
2 | infy_ocr_parser | `OCR file`, `region definition` | `region of interest [x,y,w,h]`
3 | infy_field_extractor | `OCR file`, `region of interest [x,y,w,h]` | `text`, `checkbox state(T/F)`, `radio button state(T/F)`
4 | infy_table_extractor | `image file` | `table data with rows and cols`
5 | infy_dpp_core
infy_dpp_segmentation
infy_dpp_ai
infy_dpp_storage
infy_dpp_content_extractor
| `config_data,document_data, context_data` | `document_data, context_data`
## Examples
For code examples, please read [docs/notebook](docs/notebook).
For infy_dpp_processor, please read [apps/infy_dpp_processor/README.md](apps/infy_dpp_processor/README.md)
For infy_db_service, please read [apps/infy_db_service/README.md](apps/infy_db_service/README.md)
For infy_search_service,please read [apps/infy_search_service/README.md](apps/infy_search_service/README.md)
## Reference
For API specifications, please read [docs/reference](docs/reference).