vibe.review.parsing.layout.analyzer

Layout analysis: Convert extracted words into layout structure.

This analyzer: 1. Groups words into lines based on vertical alignment 2. Groups lines into blocks based on spacing and alignment 3. Detects page regions (header, footer, body, columns) 4. Identifies repeated headers/footers across pages

LayoutConfig

Configuration for layout analysis.

LayoutAnalyzer

Analyze document layout from extracted words.

Converts ExtractedWord[] into LayoutPage[] with hierarchical structure: Page → Region → Block → Line → Word.

__init__ 

__init__(config: LayoutConfig | None = None, rules_dir: Path | None = None, rule_engine: RuleEngine | None = None, doclayout_detector: YoloLayoutDetector | None = None, table_structure_detector: TableStructureDetector | None = None) -> None

Initialize the layout analyzer.

Parameters:
  • config (LayoutConfig | None, default: None ) –

    Layout analysis configuration. Uses defaults if not provided.

  • rules_dir (Path | None, default: None ) –

    Directory containing rule YAML files. Defaults to built-in rules.

  • rule_engine (RuleEngine | None, default: None ) –

    Pre-configured rule engine. If provided, rules_dir is ignored.

  • doclayout_detector (YoloLayoutDetector | None, default: None ) –

    Optional YOLO layout detector override.

  • table_structure_detector (TableStructureDetector | None, default: None ) –

    Optional Table Transformer detector override.

analyze 

analyze(extraction: ExtractionResult, page_progress: Callable[[int, int], None] | None = None) -> list[LayoutPage]

Analyze layout of all pages.

Parameters:
  • extraction (ExtractionResult) –

    ExtractionResult from the extraction layer.

  • page_progress (Callable[[int, int], None] | None, default: None ) –

    Optional callback invoked after each page is processed. Receives (page_number, total_pages).
Returns:

extract_table_from_words 

extract_table_from_words(words: list[ExtractedWord], page_num: int, table_bbox: BBox | None = None, element_label: str = 'table') -> LayoutTable | None

Extract table structure from words using position-based analysis.

This method detects table columns by clustering text x-coordinates, groups text into rows by y-coordinate, and merges continuation rows where the first column (typically a row identifier) is empty.

Parameters:
  • words (list[ExtractedWord]) –

    Words within the table region.

  • page_num (int) –

    Page number (1-based).

  • table_bbox (BBox | None, default: None ) –

    Bounding box of the table region (optional).

  • element_label (str, default: 'table' ) –

    The doclayout label that triggered detection.
Returns:
  • LayoutTable | None –

    LayoutTable if table structure was detected, None otherwise.