Output text with markdown style codes based on font properties. Parameter is a list of span dictionaries. The spans may come from one or more

Identify white text.

Output text without any markdown or other styling. Parameter is a list of span dictionaries. The spans may come from one or more original "tex

Extract the text spans from a TextPage in natural reading sequence. All spans roughly on the same line are joined to generate an improved line.

Serialize ParsedDocument to markdown text.

Return 1-based number if a bbox contains bb, else return 0.

Check if character is in the Private Use Area (PUA) of Unicode.

Check if rect is outside all rects in the list.

Optionally render the rect part of a page. We will ignore images that are empty or that have an edge smaller than x% of the correspon

Return 1-based number if a bbox contains bb, else return 0.

Loads list of documents from PDF file and also accepts extra information in dict format. Args: file_path (Union[Path, str]): The

Output the text found inside the given clip. This is an alternative for plain text in that it outputs text enriched with markdown sty

Compute a matrix to adjust the width of the rendered text.

Check if 'inner' rectangle is contained within 'outer' rectangle.

Determines whether rectangle 'temp' can be extended by 'bb' without intersecting any of the rectangles contained in 'bboxlist' or 'ver

Determine bboxes which wrap a column on the page. Args: footer_margin: ignore text if distance from bottom is less header_margin:

2D convolution (cross-correlation) with reflect padding. Vectorized over kernel (no pixel loops).

Map the layout box number of each list-item to its hierarchy level. Args: layout_info (list): the bbox list "page.layout_information"

Extract text from a rect-like 'cell' as plain or MD styled text. This function should ultimately be used to extract text from a table cell. M

Convert "footnote" bboxes to markdown. The first line is prefixed with "> ". Subsequent lines are appended without line break if they do

Convert "footnote" bboxes to text.

Detect text bboxes with all mono-spaced lines. Returns True if all lines are mono-spaced. This may be used to output code blocks.

Create a page chunk dictionary for output. Args: doc: the ParsedDocument object page: the PageLayout object text: the pag

Output images and graphics above text rectangle.

Output tables above given text rectangle.

Sort and join the spans in a re-synthesized line. The PDF may contain "broken" text with words cut into pieces. This funtion joins sp

Process the document and return the text of the selected pages. Args: doc: pymupdf.Document or string. pages: list of page number

Processes metas of a PDF document.

Processes a single page of a PDF document.

Compute matrix to adjust the width of the rendered text.

Carry out all four checks and calculate the weighted score. img_gray: 2D uint8 array.

Analyze the page for the OCR decision. Args: blocks: output of page.get_text("dict") if already available Returns: A dict wit

Check if rect is contained in a rect of the list.

Complete Canny edge detector, NumPy-only.

Do some elementary cleaning.

Within a stripe, group boxes into columns based on horizontal proximity. We use a small horizontal gap threshold to decide when a new column

Divide page into horizontal stripes based on vertical gaps. Args: boxes (list): List of bounding boxes. joined_boxes: rectan

Output a code block in markdown format.

Output a code block in plain text format. Basic difference is that lines are separated by line breaks.

Connected Components (Union-Find) 8-connectivity Connected Components, Union-Find-based two-pass approach. binary_img: 0 background, !=0

Compute reading order of boxes delivered by PyMuPDF-Layout. Args: boxes (list): List of bounding boxes. vertical_gap (float)

Create a rectangle of full box width. Its height equals vertical_gap and its top equals y.

Edge density: mean(edges)/255.0, expects 0/255 image.

ENTROPY CHECK Shannon entropy via 256-bin histogram.

Read table content only of all pages in the document. Chatbots typically have limitations on the amount of data that can can be passed in (nu

Convert text extracted from images to markdown format.

Convert text extracted from unrecognized tables. We hope for some sort of table structure being present in the text spans: The maximum span c

FFT-Ratio Check: Low-Frequency-Ratio in FFT spectrum. Expects arbitrary size; internally rescaled to 128x128.

Remove boxes that are fully contained within another box.

Return virtual lines for a given table bbox. This utility looks for: * horizontal non-stroke vectors and uses their top and bottom edges

Separable Gaussian Blur: first horizontal, then vertical.

1D Gaussian kernel, normalized.

Determine the background color of the page. The function returns a PDF RGB color triple or None. We check the color of 10 x 10 pixel

Determine whether the page contains text worthwhile to OCR. Args: page: PyMuPDF Page object dpi: DPI used for rasterization *if*

Process one page. Args: doc: pymupdf.Document pno: 0-based page number textflags: text extraction flag bi

Return OCR'd span text using Tesseract. Args: page: pymupdf Page bbox: pymupdf Rect or its sequence dpi: resolution for O

Use Tesseract to extract text from a given bounding box of the pixmap. The irect is expected to contain one line only, so we use tessedit_pag

Extract text by line keeping natural reading sequence. Notes: Internally uses "dict" to select lines and their spans. Returns pla

Hysteresis with iterative neighborhood activation, vectorized.

Return True if a bbox touches bb, else return False.

Check if inner box is fully contained within outer box.

Check if the boxes have a clean multi-column layout. If False, the boxes represent only 1 column Used to early exit from stripe clust

Check whether the rectangle "box" contains 'signifiant' drawings. This means that some path is contained in the "interior" of box. To this en

Check for relevant text.

Postprocess identified text blocks, phase 2. Increase the width of each text block so that small left or right border differences are

Convert "list-item" bboxes to markdown. The first line is prefixed with "- ". Subsequent lines are appended without line break if their r

Convert "list-item" bboxes to text.

Non-Maximum Suppression, mostly vectorized.

Otsu Threshold Otsu thresholding, NumPy-only.

Check if page exclusivley contains OCR text. For this to be true, all text must be written as "ignore-text".

Convert text extracted from images to markdown format.

Convert text extracted from images to plain text format.

Join any rectangles with a pairwise non-empty overlap. Accepts and returns a list of Rect items. Note that rectangles that only "touch" each

Bilinear resize (similar to OpenCV INTER_LINEAR), vectorized. img: 2D uint8/float32, shape (H, W)

Accept a span and return a markdown link string. Args: links: a list as returned by page.get_links() span: a span dic

Convert "section-header" bboxes to markdown. This is treated as a level 2 header (##). The line text itself is handled like normal text.

Join vectors that are horizontally adjacent and vertically aligned.