Ocr

domain.fetch_pipeline.utils.ocr

Summary

OCR helpers for statement images and PDFs.

Covers number cleaning, tesseract text normalization, PDF rasterization, and per-page OCR with optional binarization.

Functions

clean_num

clean_num(s: str) -> float

Parse a numeric token into a float amount.

Strips non-digits and divides by the correct power of ten based on the recorded decimal part; a token without digits yields 0.0.

Parameters:

Name	Type	Description	Default
s	str	Raw numeric string.	required

Returns:

Name	Type	Description
float	float	Parsed value.

normalize

normalize(text: str) -> str

Normalize a raw OCR string for parsing.

Fixes a tesseract digit misread, expands pipes to spaces, trims stray punctuation from the leading edge (including the replacement character), and removes commas.

Parameters:

Name	Type	Description	Default
text	str	Raw OCR output.	required

Returns:

Name	Type	Description
str	str	Normalized text.

ocr_all_pages

ocr_all_pages(
    pdf_bytes: bytes, covert_to_image: bool = None
) -> list

OCR every page of a PDF in a temporary directory.

With covert_to_image pages are binarized and OCRed with the numeric-mode OEM 3 config; otherwise they are OCRed with the default PSM 6 config.

Parameters:

Name	Type	Description	Default
pdf_bytes	bytes	Raw PDF content.	required
covert_to_image	bool	When truthy, binarize pages to images before OCR.	None

Returns:

Name	Type	Description
list	list	OCR text, one string per page.

rasterize_pdf

rasterize_pdf(
    pdf_bytes: bytes, out_dir: str, dpi: int = 300
) -> list

Rasterize a PDF into per-page JPEGs.

Writes the bytes as input.pdf in out_dir and invokes pdftoppm at the requested DPI.

Parameters:

Name	Type	Description	Default
pdf_bytes	bytes	Raw PDF content.	required
out_dir	str	Directory used to write the PDF and JPEG outputs.	required
dpi	int	Rasterization resolution, defaults to 300.	300

Returns:

Name	Type	Description
list	list	Sorted list of JPEG path strings.

Raises:

Type	Description
CalledProcessError	If pdftoppm fails.