Run1 pdf calendar parsing

PDF Calendar Parsing

Approach Overview

To extract schedule data from a visual PDF calendar, you need to:

1. Convert the PDF to an image or parse its vector/text content
2. Identify the time axis and grid lines
3. Detect colored blocks representing appointments
4. Map pixel positions to actual times

Tools and Libraries

Python libraries:

• pdfplumber — extracts text, lines, and rectangles from PDFs with precise coordinates
• PyMuPDF (fitz) — renders PDF pages to images, extracts drawings and text with positions
• pdf2image + Pillow — converts PDF to images for pixel-level analysis
• tabula-py or camelot — for table extraction (less useful for visual calendars)

Extracting Horizontal Lines with pdfplumber

import pdfplumber

with pdfplumber.open("/root/calendar.pdf") as pdf:
    page = pdf.pages[0]
    # Get all lines
    lines = page.lines
    # Filter horizontal lines (y0 == y1 or nearly equal)
    horizontal_lines = [l for l in lines if abs(l['top'] - l['bottom']) < 2]
    # Sort by vertical position
    horizontal_lines.sort(key=lambda l: l['top'])

Extracting Rectangles/Blocks with pdfplumber

    rects = page.rects
    for rect in rects:
        print(rect)  # Keys: x0, y0, x1, y1, top, bottom, fill color, etc.

Extracting Colored Blocks with PyMuPDF

import fitz

doc = fitz.open("/root/calendar.pdf")
page = doc[0]

# Get drawings (vector graphics)
drawings = page.get_drawings()
for d in drawings:
    for item in d["items"]:
        # item is a tuple like ("re", rect) for rectangles
        pass
    fill_color = d.get("fill")  # RGB tuple, e.g., (0, 0, 1) for blue
    rect = d.get("rect")        # fitz.Rect object

Identifying Blue Blocks

Blue blocks typically have RGB fill values where:

• Blue channel is dominant (close to 1.0)
• Red and Green channels are low

def is_blue(color):
    if color is None:
        return False
    r, g, b = color[:3]
    return b > 0.5 and r < 0.5 and g < 0.5

Mapping Positions to Times

Given that the space between two adjacent horizontal lines = 15 minutes:

# After sorting horizontal lines by y-position
line_positions = sorted(set(l['top'] for l in horizontal_lines))

# Determine the starting time from text labels on the calendar
# e.g., if the first line corresponds to 8:00 AM
start_time = datetime.strptime("08:00 AM", "%I:%M %p")

def y_to_time(y, line_positions, start_time):
    """Convert a y-coordinate to a time based on grid lines."""
    from datetime import timedelta
    # Find which interval the y falls in
    for i in range(len(line_positions) - 1):
        if line_positions[i] <= y <= line_positions[i + 1]:
            # Interpolate within the interval
            frac = (y - line_positions[i]) / (line_positions[i + 1] - line_positions[i])
            minutes = (i + frac) * 15
            return start_time + timedelta(minutes=minutes)
    # If beyond last line
    idx = len(line_positions) - 1
    extra = (y - line_positions[-1]) / (line_positions[1] - line_positions[0])
    minutes = (idx + extra) * 15
    return start_time + timedelta(minutes=minutes)

Extracting Text Labels

# With pdfplumber
words = page.extract_words()
for w in words:
    print(w['text'], w['top'], w['x0'])  # text content and position

Determining Day Columns

If the calendar has multiple days (columns):

1. Extract day/date headers from text at the top
2. Identify column boundaries from vertical lines or header positions
3. Map each block's x-position to the appropriate day column

vertical_lines = [l for l in lines if abs(l['x0'] - l['x1']) < 2]
vertical_lines.sort(key=lambda l: l['x0'])
column_boundaries = [l['x0'] for l in vertical_lines]

def x_to_day(x, column_boundaries, day_labels):
    for i in range(len(column_boundaries) - 1):
        if column_boundaries[i] <= x < column_boundaries[i + 1]:
            return day_labels[i]
    return day_labels[-1]

Image-Based Color Detection (Fallback)

If vector extraction doesn't yield colors reliably:

from pdf2image import convert_from_path
from PIL import Image
import numpy as np

images = convert_from_path("/root/calendar.pdf", dpi=200)
img = np.array(images[0])

# Check pixel color at specific coordinates
# Blue pixels: high B channel, low R and G
blue_mask = (img[:,:,2] > 150) & (img[:,:,0] < 100) & (img[:,:,1] < 100)

Tips

• PDF coordinate systems typically have origin at bottom-left, but pdfplumber uses top-left
• Always verify coordinate system by cross-referencing text positions with known labels
• Account for small floating-point differences when comparing positions
• Group nearby horizontal lines that might be duplicates (borders vs grid lines)