Working with Regions¶

Regions are rectangular areas on a page that define boundaries for operations like text extraction, element finding, or visualization. They're one of Natural PDF's most powerful features for working with specific parts of a document.

Setup¶

Let's set up a PDF to experiment with regions.

In [1]:

from natural_pdf import PDF

# Load the PDF
pdf = PDF("https://github.com/jsoma/natural-pdf/raw/refs/heads/main/pdfs/01-practice.pdf")

# Get the first page
page = pdf.pages[0]

# Display the page
page.show(width=700)

from natural_pdf import PDF # Load the PDF pdf = PDF("https://github.com/jsoma/natural-pdf/raw/refs/heads/main/pdfs/01-practice.pdf") # Get the first page page = pdf.pages[0] # Display the page page.show(width=700)

CropBox missing from /Page, defaulting to MediaBox

Out[1]:

Creating Regions¶

There are several ways to create regions in Natural PDF.

Using create_region() with Coordinates¶

This is the most direct method - provide the coordinates directly.

In [2]:

# Create a region by specifying (x0, top, x1, bottom) coordinates
# Let's create a region in the middle of the page
mid_region = page.create_region(
    x0=100,         # Left edge
    top=200,        # Top edge
    x1=500,         # Right edge
    bottom=400      # Bottom edge
)

# Highlight the region to see it
mid_region.highlight(color="blue").show()

# Create a region by specifying (x0, top, x1, bottom) coordinates # Let's create a region in the middle of the page mid_region = page.create_region( x0=100, # Left edge top=200, # Top edge x1=500, # Right edge bottom=400 # Bottom edge ) # Highlight the region to see it mid_region.highlight(color="blue").show()

Out[2]:

Using Element Methods: above(), below(), left(), right()¶

You can create regions relative to existing elements.

In [3]:

# Find a heading-like element
heading = page.find('text[size>=12]:bold')

# Create a region below this heading element
if heading:
    region_below = heading.below()
    
    # Highlight the heading and the region below it
    heading.highlight(color="red")
    region_below.highlight(color="blue")
    page.show()

# Find a heading-like element heading = page.find('text[size>=12]:bold') # Create a region below this heading element if heading: region_below = heading.below() # Highlight the heading and the region below it heading.highlight(color="red") region_below.highlight(color="blue") page.show()

In [4]:

# Create a region with height limit
if heading:
    # Only include 100px below the heading
    small_region_below = heading.below(height=100)
    
    page.clear_highlights()
    heading.highlight(color="red")
    small_region_below.highlight(color="green")
    page.show()

# Create a region with height limit if heading: # Only include 100px below the heading small_region_below = heading.below(height=100) page.clear_highlights() heading.highlight(color="red") small_region_below.highlight(color="green") page.show()

In [5]:

# Find a line or other element to create a region above
line = page.find('line')
if line:
    # Create a region above the line
    region_above = line.above()
    
    page.clear_highlights()
    line.highlight(color="black")
    region_above.highlight(color="purple")
    page.show()

# Find a line or other element to create a region above line = page.find('line') if line: # Create a region above the line region_above = line.above() page.clear_highlights() line.highlight(color="black") region_above.highlight(color="purple") page.show()

Creating a Region Between Elements with until()¶

In [6]:

# Find two elements to use as boundaries
first_heading = page.find('text[size>=11]:bold')
next_heading = first_heading.next('text[size>=11]:bold') if first_heading else None

if first_heading and next_heading:
    # Create a region from the first heading until the next heading
    section = first_heading.below(until=next_heading, include_endpoint=False)
    
    # Highlight both elements and the region between them
    page.clear_highlights()
    first_heading.highlight(color="red")
    next_heading.highlight(color="red")
    section.highlight(color="yellow")
    page.show()

# Find two elements to use as boundaries first_heading = page.find('text[size>=11]:bold') next_heading = first_heading.next('text[size>=11]:bold') if first_heading else None if first_heading and next_heading: # Create a region from the first heading until the next heading section = first_heading.below(until=next_heading, include_endpoint=False) # Highlight both elements and the region between them page.clear_highlights() first_heading.highlight(color="red") next_heading.highlight(color="red") section.highlight(color="yellow") page.show()

Jump to the enclosing area with parent()¶

Need to know which table, figure or coloured panel an element sits inside?

The new parent() helper finds the smallest element or detected region that spatially encloses the current one.

In [7]:

# After running a layout model
page.analyze_layout('tatr')

text  = page.find(text="Hazardous Materials")
row       = text.parent('region[type=table-row]')
row.show(crop=True)

# After running a layout model page.analyze_layout('tatr') text = page.find(text="Hazardous Materials") row = text.parent('region[type=table-row]') row.show(crop=True)

Out[7]:

Parameters

parent(selector=None, *, mode="contains")

mode = "contains"  # candidate fully covers the element (default)
     | "center"    # candidate contains element's centroid
     | "overlap"   # any intersection > 0pt²

If no enclosing object matches the selector (or exists at all) it returns None, so chaining is safe.

Using Regions¶

Once you have a region, here's what you can do with it.

Extract Text from a Region¶

In [8]:

# Find a region to work with (e.g., from a title to the next bold text)
title = page.find('text:contains("Site")')  # Adjust if needed
if title:
    # Create a region from title down to the next bold text
    content_region = title.below(until='line:horizontal', include_endpoint=False)
    
    # Extract text from just this region
    region_text = content_region.extract_text()
    
    # Show the region and the extracted text
    page.clear_highlights()
    content_region.highlight(color="green")
    page.show()
    
    # Displaying the text (first 300 chars if long)
    print(region_text[:300] + "..." if len(region_text) > 300 else region_text)

# Find a region to work with (e.g., from a title to the next bold text) title = page.find('text:contains("Site")') # Adjust if needed if title: # Create a region from title down to the next bold text content_region = title.below(until='line:horizontal', include_endpoint=False) # Extract text from just this region region_text = content_region.extract_text() # Show the region and the extracted text page.clear_highlights() content_region.highlight(color="green") page.show() # Displaying the text (first 300 chars if long) print(region_text[:300] + "..." if len(region_text) > 300 else region_text)

Date: February 3, 1905
Violation Count: 7
Summary: Worst of any, however, were the fertilizer men, and those who served in the cooking rooms.
These people could not be shown to the visitor - for the odor of a fertilizer man would scare any ordinary
visitor at a hundred yards, and as for the other me...

Find Elements Within a Region¶

You can use a region as a "filter" to only find elements within its boundaries.

In [9]:

# Create a region in an interesting part of the page
test_region = page.create_region(
    x0=page.width * 0.1, 
    top=page.height * 0.25, 
    x1=page.width * 0.9, 
    bottom=page.height * 0.75
)

# Find all text elements ONLY within this region
text_in_region = test_region.find_all('text')

# Display result
page.clear_highlights()
test_region.highlight(color="blue")
text_in_region.highlight(color="red")
page.show()

len(text_in_region)  # Number of text elements found in region

# Create a region in an interesting part of the page test_region = page.create_region( x0=page.width * 0.1, top=page.height * 0.25, x1=page.width * 0.9, bottom=page.height * 0.75 ) # Find all text elements ONLY within this region text_in_region = test_region.find_all('text') # Display result page.clear_highlights() test_region.highlight(color="blue") text_in_region.highlight(color="red") page.show() len(text_in_region) # Number of text elements found in region

Out[9]:

17

Generate an Image of a Region¶

In [10]:

# Find a specific region to capture
# (Could be a table, figure, or any significant area)
region_for_image = page.create_region(
    x0=100, 
    top=150,
    x1=page.width - 100,
    bottom=300
)

# Generate an image of just this region
region_for_image.to_image(crop=True)  # Shows just the region

# Find a specific region to capture # (Could be a table, figure, or any significant area) region_for_image = page.create_region( x0=100, top=150, x1=page.width - 100, bottom=300 ) # Generate an image of just this region region_for_image.to_image(crop=True) # Shows just the region

Out[10]:

Adjust and Expand Regions¶

In [11]:

# Take an existing region and expand it
region_a = page.create_region(200, 200, 400, 400)

# Expand by a certain number of points in each direction
expanded = region_a.expand(left=20, right=20, top=20, bottom=20)

# Visualize original and expanded regions
page.clear_highlights()
region_a.highlight(color="blue", label="Original")
expanded.highlight(color="red", label="Expanded")
page.to_image()

# Take an existing region and expand it region_a = page.create_region(200, 200, 400, 400) # Expand by a certain number of points in each direction expanded = region_a.expand(left=20, right=20, top=20, bottom=20) # Visualize original and expanded regions page.clear_highlights() region_a.highlight(color="blue", label="Original") expanded.highlight(color="red", label="Expanded") page.to_image()

Out[11]:

Using Exclusion Zones with Regions¶

Exclusion zones are regions that you want to ignore during operations like text extraction.

In [12]:

# Create a region for the whole page
full_page_region = page.create_region(0, 0, page.width, page.height)

# Extract text without exclusions as baseline
full_text = full_page_region.extract_text()
print(f"Full page text length: {len(full_text)} characters")

# Create a region for the whole page full_page_region = page.create_region(0, 0, page.width, page.height) # Extract text without exclusions as baseline full_text = full_page_region.extract_text() print(f"Full page text length: {len(full_text)} characters")

Full page text length: 1255 characters

In [13]:

# Define an area we want to exclude (like a header)
# Let's exclude the top 10% of the page
header_zone = page.create_region(0, 0, page.width, page.height * 0.1)

# Add this as an exclusion for the page
page.add_exclusion(header_zone)

# Visualize the exclusion
page.clear_highlights()
header_zone.highlight(color="red", label="Excluded")
page.show()

# Define an area we want to exclude (like a header) # Let's exclude the top 10% of the page header_zone = page.create_region(0, 0, page.width, page.height * 0.1) # Add this as an exclusion for the page page.add_exclusion(header_zone) # Visualize the exclusion page.clear_highlights() header_zone.highlight(color="red", label="Excluded") page.show()

Out[13]:

In [14]:

# Now extract text again - the header should be excluded
text_with_exclusion = full_page_region.extract_text() # Uses apply_exclusions=True by default

# Compare text lengths
print(f"Original text: {len(full_text)} chars\nText with exclusion: {len(text_with_exclusion)} chars")
print(f"Difference: {len(full_text) - len(text_with_exclusion)} chars excluded")

# Now extract text again - the header should be excluded text_with_exclusion = full_page_region.extract_text() # Uses apply_exclusions=True by default # Compare text lengths print(f"Original text: {len(full_text)} chars\nText with exclusion: {len(text_with_exclusion)} chars") print(f"Difference: {len(full_text) - len(text_with_exclusion)} chars excluded")

Original text: 1255 chars
Text with exclusion: 1193 chars
Difference: 62 chars excluded

In [15]:

# When done with this page, clear exclusions
page.clear_exclusions()

# When done with this page, clear exclusions page.clear_exclusions()

Out[15]:

<Page number=1 index=0>

Document-Level Exclusions¶

PDF-level exclusions apply to all pages and use functions to adapt to each page.

In [16]:

# Define a PDF-level exclusion for headers
# This will exclude the top 30% of every page
pdf.add_exclusion(
    lambda p: p.create_region(0, 0, p.width, p.height * 0.3),
    label="Header zone"
)

# Define a PDF-level exclusion for footers
# This will exclude the bottom 20% of every page
pdf.add_exclusion(
    lambda p: p.create_region(0, p.height * 0.8, p.width, p.height),
    label="Footer zone"
)

# PDF-level exclusions are used whenever you extract text
# Let's try on the first three pages
for page in pdf.pages[:3]:
    text = page.extract_text()
    text_original = page.extract_text(use_exclusions=False)
    print(f"Page {page.number} – Before: {len(text_original)} After: {len(text)}")

# Define a PDF-level exclusion for headers # This will exclude the top 30% of every page pdf.add_exclusion( lambda p: p.create_region(0, 0, p.width, p.height * 0.3), label="Header zone" ) # Define a PDF-level exclusion for footers # This will exclude the bottom 20% of every page pdf.add_exclusion( lambda p: p.create_region(0, p.height * 0.8, p.width, p.height), label="Footer zone" ) # PDF-level exclusions are used whenever you extract text # Let's try on the first three pages for page in pdf.pages[:3]: text = page.extract_text() text_original = page.extract_text(use_exclusions=False) print(f"Page {page.number} – Before: {len(text_original)} After: {len(text)}")

Page 1 – Before: 456 After: 456

In [17]:

# Clear PDF-level exclusions when done
pdf.clear_exclusions()
print("Cleared all PDF-level exclusions")

# Clear PDF-level exclusions when done pdf.clear_exclusions() print("Cleared all PDF-level exclusions")

Cleared all PDF-level exclusions

Working with Layout Analysis Regions¶

When you run layout analysis, the detected regions (tables, titles, etc.) are also Region objects.

In [18]:

# First, run layout analysis to detect regions
page.analyze_layout()  # Uses 'yolo' engine by default

# Find all detected regions
detected_regions = page.find_all('region')
print(f"Found {len(detected_regions)} layout regions")

# First, run layout analysis to detect regions page.analyze_layout() # Uses 'yolo' engine by default # Find all detected regions detected_regions = page.find_all('region') print(f"Found {len(detected_regions)} layout regions")

image 1/1 /var/folders/25/h3prywj14qb0mlkl2s8bxq5m0000gn/T/tmpwn8a4mv6/temp_layout_image.png: 1024x800 1 title, 3 plain texts, 2 abandons, 1 table, 1248.8ms

Speed: 6.3ms preprocess, 1248.8ms inference, 1.2ms postprocess per image at shape (1, 3, 1024, 800)

Found 71 layout regions

In [19]:

# Highlight all detected regions by type
detected_regions.highlight(group_by='region_type').show()

# Highlight all detected regions by type detected_regions.highlight(group_by='region_type').show()

Out[19]:

In [20]:

# Extract text from a specific region type (e.g., title)
title_regions = page.find_all('region[type=title]')
if title_regions:
    titles_text = title_regions.extract_text()
    print(f"Title text: {titles_text}")

# Extract text from a specific region type (e.g., title) title_regions = page.find_all('region[type=title]') if title_regions: titles_text = title_regions.extract_text() print(f"Title text: {titles_text}")

Title text: 

Next Steps¶

Now that you understand regions, you can:

• Extract tables from table regions
• Ask questions about specific regions
• Exclude content from extraction