img2table

Features

• Complete overhaul of the line detection algorithm to improve detection of lines defined by background color changes
• Improvement in detection of semi-bordered cells
• Update detection of rows in borderless tables
• Add support for Surya OCR
• Add detection of implicit columns via the implicit_columns parameter
• Optimization of code performance via numba refactoring
• Update of examples notebooks

Bug fixes

• Fix bug with text position when extracting text from rotated PDFs

• Simpler and more consistent line detection
• Detection of discontinuous columns in borderless tables

• Fix miscellaneous code left from legacy processing
• Add margin to top/bottom of borderless tables

What's Changed

• Update metrics computation and borderless table detection
• Add compatibility with Python 3.12
• Add support for documents with black backgrounds

• Fix division by zero bug introduced in previous release

• Fix bugs
• Improve computation of image metrics on noisy documents
• Modify row detection for borderless tables in order to account for merged cells
• Implement Adaptive Run Length Smoothing Algorithm in order to isolate text areas in images

• Fix bugs related to OCR / table content extraction

• Fix bug in line detection
• Fix bug in cell creation
• Optimization of algorithm performances

• Improved processing of tables with dotted lines
• Add detection of semi-bordered cells in tables
• Update borderless table algorithm
• Speed improvements and code optimization (2 to 4x faster depending on inputs)

• Add HTML representation to extracted tables
• Call OCR only on pages/images containing tables
• Bump Pillow requirements for vulnerabilities

• Add option to pass keyword arguments for PaddleOCR/EasyOCR/docTR constructors
• Fix bug with PaddleOCR on blank images
• Line filtering for borderless table recognition
• Update deprecated polars code

• Fix issues related to latest polars release
• Improve detection of columns in document layout
• Fix rare bug leading to no detected lines
• Add coherency checks on borderless tables
• Wrap text when exporting to xlsx

• Improvement on document layout analysis in order to detect borderless table areas
• Modification of handling of optional dependencies

• Add support for docTR
• Fixes on line detection

• Drop Python 3.7 support
• Allow PaddleOCR on Python 3.11
• Improve detection of intersection between lines and words

• Deprecate Python 3.7
• Fix PanicExceptions in polars code
• Replace deprecated polars functions
• Fix requirements for EasyOCR

• Fix paddlepaddle version to avoid errors on certain platforms
• Better handling of exceptions in polars crossjoins

• Fix error with merged cells when creating xlsx files

• Improvement on line detection accuracy
• Adapt detection of implicit rows to image characteristics
• Handle column based documents for borderless table detection

• Fix bug in kernel size when using medianBlur
• Fix bug for GCP Vision missing vertex coordinates