Function make_table_from_bbox

src/table.py:164–220

Detect table structure within a given rectangle.

(textpage, word_rects, rect)

def make_table_from_bbox(textpage, word_rects, rect):
    """Detect table structure within a given rectangle."""
    cells = []  # table cells as (x0,y0,x1,y1) tuples

    # calls fz_find_table_within_bounds
    block = get_table_dict_from_rect(textpage, rect)
    # No table structure found if not a grid block
    if block.get("type") != mupdf.FZ_STEXT_BLOCK_GRID:
        return cells
    bbox = pymupdf.Rect(block["bbox"])  # resulting table bbox

    # lists of (pos,uncertainty) tuples
    xpos = sorted(block["xpos"], key=lambda x: x[0])
    ypos = sorted(block["ypos"], key=lambda y: y[0])

    # maximum uncertainties in x and y directions
    xmaxu, ymaxu = block["max_uncertain"]

    # Modify ypos to remove uncertain positions, and y positions
    # that cut through words.
    nypos = []
    for y, yunc in ypos:
        if yunc > 0:  # allow no uncertain y values
            continue
        if intersects_words_h(bbox, y, word_rects):
            continue  # allow no y that cuts through words
        if nypos and (y - nypos[-1] < 3):
            nypos[-1] = y  # snap close positions
        else:
            nypos.append(y)

    # New max y uncertainty: 35% of remaining y positions.
    # Omit x positions that intersect too many words, otherwise
    # only remove x for the affected cells.
    ymaxu = max(0, round((len(nypos) - 2) * 0.35))

    # Exclude x positions with too high uncertainty
    # (we allow more uncertainty in x direction)
    nxpos = [x[0] for x in xpos if x[1] <= ymaxu]
    if bbox.x1 > nxpos[-1] + 3:
        nxpos.append(bbox.x1)  # ensure right table border

    # Compose cells from the remaining x and y positions.
    for i in range(len(nypos) - 1):
        row_box = pymupdf.Rect(bbox.x0, nypos[i], bbox.x1, nypos[i + 1])
        # Sub-select words in this row and sort them by left coordinate
        row_words = sorted(
            [r for r in word_rects if rect_in_rect(r, row_box)], key=lambda r: r.x0
        )
        # Sub-select x values that do not cut through words
        this_xpos = [x for x in nxpos if not any(r.x0 < x < r.x1 for r in row_words)]
        for j in range(len(this_xpos) - 1):
            cell = pymupdf.Rect(this_xpos[j], nypos[i], this_xpos[j + 1], nypos[i + 1])
            if not cell.is_empty:  # valid cell
                cells.append(tuple(cell))
    # Add new table to TableFinder tables
    return cells