Create findInitialMarkups.py

findInitialMarkups.py

@@ -0,0 +1,572 @@
+import re
+from collections import defaultdict, Counter
+import fitz  # PyMuPDF
+import requests
+from io import BytesIO
+
+def normalize_text(text):
+    if text is None:
+        return ""
+    return re.sub(r'\s+', ' ', text.strip().lower())
+
+def get_spaced_text_from_spans(spans):
+    return normalize_text(" ".join(span["text"].strip() for span in spans))
+
+def is_header(span, most_common_font_size, most_common_color, most_common_font):
+    fontname = span.get("font", "").lower()
+    # is_italic = "italic" in fontname or "oblique" in fontname
+    is_bold = "bold" in fontname or span.get("bold", False)
+    return (
+        (
+            span["size"] > most_common_font_size or
+            span["font"].lower() != most_common_font.lower() or
+            is_bold
+        )
+    )
+
+def add_span_to_nearest_group(span_y, grouped_dict, pageNum=None, threshold=0.5):
+    for (p, y) in grouped_dict:
+        if pageNum is not None and p != pageNum:
+            continue
+        if abs(y - span_y) <= threshold:
+            return (p, y)
+    return (pageNum, span_y)
+
+
+def get_regular_font_size_and_color(doc):
+    font_sizes = []
+    colors = []
+    fonts = []
+
+    # Loop through all pages
+    for page_num in range(len(doc)):
+        page = doc.load_page(page_num)
+        for span in page.get_text("dict")["blocks"]:
+            if "lines" in span:
+                for line in span["lines"]:
+                    for span in line["spans"]:
+                        font_sizes.append(span['size'])
+                        colors.append(span['color'])
+                        fonts.append(span['font'])
+
+    # Get the most common font size, color, and font
+    most_common_font_size = Counter(font_sizes).most_common(1)[0][0] if font_sizes else None
+    most_common_color = Counter(colors).most_common(1)[0][0] if colors else None
+    most_common_font = Counter(fonts).most_common(1)[0][0] if fonts else None
+
+    return most_common_font_size, most_common_color, most_common_font
+    
+def extract_headers(doc, toc_pages, most_common_font_size, most_common_color, most_common_font, top_margin, bottom_margin):
+    print("Font baseline:", most_common_font_size, most_common_color, most_common_font)
+
+    grouped_headers = defaultdict(list)
+    spans = []
+    line_merge_threshold = 1.5  # Maximum vertical distance between lines to consider as part of same header
+
+    for pageNum in range(len(doc)):
+        if pageNum in toc_pages:
+            continue
+        page = doc.load_page(pageNum)
+        page_height = page.rect.height
+        text_instances = page.get_text("dict")
+
+        # First pass: collect all potential header spans
+        potential_header_spans = []
+        for block in text_instances['blocks']:
+            if block['type'] != 0:
+                continue
+
+            for line in block['lines']:
+                for span in line['spans']:
+                    span_y0 = span['bbox'][1]
+                    span_y1 = span['bbox'][3]
+
+                    if span_y0 < top_margin or span_y1 > (page_height - bottom_margin):
+                        continue
+
+                    span_text = normalize_text(span.get('text', ''))
+                    if not span_text:
+                        continue
+                    if span_text.startswith('http://www') or span_text.startswith('www'):
+                        continue
+                    if any((
+                        'page' in span_text,
+                        not re.search(r'[a-z0-9]', span_text),
+                        'end of section' in span_text,
+                        re.search(r'page\s+\d+\s+of\s+\d+', span_text),
+                        re.search(r'\b(?:\d{1,2}[/-])?\d{1,2}[/-]\d{2,4}\b', span_text),
+                        # re.search(r'\b(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)', span_text),
+                        'specification:' in span_text
+                    )):
+                        continue
+
+                    cleaned_text = re.sub(r'[.\-]{4,}.*$', '', span_text).strip()
+                    cleaned_text = normalize_text(cleaned_text)
+
+                    if is_header(span, most_common_font_size, most_common_color, most_common_font):
+                        potential_header_spans.append({
+                            'text': cleaned_text,
+                            'size': span['size'],
+                            'pageNum': pageNum,
+                            'y0': span_y0,
+                            'y1': span_y1,
+                            'x0': span['bbox'][0],
+                            'x1': span['bbox'][2],
+                            'span': span
+                        })
+
+        # Sort spans by vertical position (top to bottom)
+        potential_header_spans.sort(key=lambda s: (s['pageNum'], s['y0']))
+
+        # Second pass: group spans that are vertically close and likely part of same header
+        i = 0
+        while i < len(potential_header_spans):
+            current = potential_header_spans[i]
+            header_text = current['text']
+            header_size = current['size']
+            header_page = current['pageNum']
+            min_y = current['y0']
+            max_y = current['y1']
+            spans_group = [current['span']]
+
+            # Look ahead to find adjacent lines that might be part of same header
+            j = i + 1
+            while j < len(potential_header_spans):
+                next_span = potential_header_spans[j]
+                # Check if on same page and vertically close with similar styling
+                if (next_span['pageNum'] == header_page and
+                    next_span['y0'] - max_y < line_merge_threshold and
+                    abs(next_span['size'] - header_size) < 0.5):
+                    header_text += " " + next_span['text']
+                    max_y = next_span['y1']
+                    spans_group.append(next_span['span'])
+                    j += 1
+                else:
+                    break
+
+            # Add the merged header
+            grouped_headers[(header_page, min_y)].append({
+                "text": header_text.strip(),
+                "size": header_size,
+                "pageNum": header_page,
+                "spans": spans_group
+            })
+            spans.extend(spans_group)
+            i = j  # Skip the spans we've already processed
+
+    # Prepare final headers list
+    headers = []
+    for (pageNum, y), header_groups in sorted(grouped_headers.items()):
+        for group in header_groups:
+            headers.append([
+                group['text'],
+                group['size'],
+                group['pageNum'],
+                y
+            ])
+
+    font_sizes = [size for _, size, _, _ in headers]
+    font_size_counts = Counter(font_sizes)
+
+      # Filter font sizes that appear at least 3 times
+    valid_font_sizes = [size for size, count in font_size_counts.items() if count >= 3]
+
+    # Sort in descending order
+    valid_font_sizes_sorted = sorted(valid_font_sizes, reverse=True)
+
+    # If only 2 sizes, repeat the second one
+    if len(valid_font_sizes_sorted) == 2:
+        top_3_font_sizes = [valid_font_sizes_sorted[0], valid_font_sizes_sorted[1], valid_font_sizes_sorted[1]]
+    else:
+        top_3_font_sizes = valid_font_sizes_sorted[:3]
+
+    # Get the smallest font size among valid ones
+    smallest_font_size = min(valid_font_sizes) if valid_font_sizes else None
+
+    print("Smallest font size in headers:", smallest_font_size)
+
+    return headers, top_3_font_sizes, smallest_font_size, spans
+
+import re
+import difflib
+
+def is_numbered(text):
+    return bool(re.match(r'^\d', text.strip()))
+
+def is_similar(a, b, threshold=0.85):
+    return difflib.SequenceMatcher(None, a, b).ratio() > threshold
+
+def normalize(text):
+    text = text.lower()
+    text = re.sub(r'\.{2,}', '', text)  # remove long dots
+    text = re.sub(r'\s+', ' ', text)    # replace multiple spaces with one
+    return text.strip()
+
+def clean_toc_entry(toc_text):
+    """Remove page numbers and formatting from TOC entries"""
+    # Remove everything after last sequence of dots/whitespace followed by digits
+    return re.sub(r'[\.\s]+\d+.*$', '', toc_text).strip('. ')
+
+def build_header_hierarchy(doc, toc_pages, most_common_font_size, most_common_color, most_common_font, top_margin=70, bottom_margin=70):
+    # Extract headers with margin handling
+    headers_list, top_3_font_sizes, smallest_font_size, spans = extract_headers(
+        doc,
+        toc_pages=toc_pages,
+        most_common_font_size=most_common_font_size,
+        most_common_color=most_common_color,
+        most_common_font=most_common_font,
+        top_margin=top_margin,
+        bottom_margin=bottom_margin
+    )
+
+    # Step 1: Collect and filter potential headers
+    headers = []
+    seen_headers = set()
+
+    # First extract TOC entries to get exact level 0 header texts
+    toc_entries = {}
+    for pno in toc_pages:
+        page = doc.load_page(pno)
+        toc_text = page.get_text()
+        for line in toc_text.split('\n'):
+            clean_line = line.strip()
+            if clean_line:
+                norm_line = normalize(clean_line)
+                toc_entries[norm_line] = clean_line  # Store original text
+  
+    for h in headers_list:
+        text, size, pageNum, y = h[:4]
+        page = doc.load_page(pageNum)
+        page_height = page.rect.height
+
+        # Skip margin areas
+        if y < top_margin or y > (page_height - bottom_margin):
+            continue
+
+        norm_text = normalize(text)
+        if len(norm_text) > 2 and size >= most_common_font_size:
+            headers.append({
+                "text": text,
+                "page": pageNum,
+                "y": y,
+                "size": size,
+                "bold": h[4] if len(h) > 4 else False,
+                # "italic": h[5] if len(h) > 5 else False,
+                "color": h[6] if len(h) > 6 else None,
+                "font": h[7] if len(h) > 7 else None,
+                "children": [],
+                "is_numbered": is_numbered(text),
+                "original_size": size,
+                "norm_text": norm_text,
+                "level": -1  # Initialize as unassigned
+            })
+
+    # Sort by page and vertical position
+    headers.sort(key=lambda h: (h['page'], h['y']))
+    # Step 2: Detect consecutive headers and assign levels
+    i = 0
+    while i < len(headers) - 1:
+        current = headers[i]
+        next_header = headers[i+1]
+
+        # Check if they are on the same page and very close vertically (likely consecutive lines)
+        if (current['page'] == next_header['page'] and
+            abs(current['y'] - next_header['y']) < 20):  # 20pt threshold for "same line"
+
+            # Case 1: Both unassigned - make current level 1 and next level 2
+            if current['level'] == -1 and next_header['level'] == -1:
+                current['level'] = 1
+                next_header['level'] = 2
+                i += 1  # Skip next header since we processed it
+
+            # Case 2: Current unassigned, next assigned - make current one level above
+            elif current['level'] == -1 and next_header['level'] != -1:
+                current['level'] = max(1, next_header['level'] - 1)
+
+            # Case 3: Current assigned, next unassigned - make next one level below
+            elif current['level'] != -1 and next_header['level'] == -1:
+                next_header['level'] = current['level'] + 1
+                i += 1  # Skip next header since we processed it
+        i += 1
+    # Step 2: Identify level 0 headers (largest and in TOC)
+    # max_size = max(h['size'] for h in headers) if headers else 0
+    max_size,subheaderSize,nbsheadersize=top_3_font_sizes
+    print(max_size)
+    toc_text_match=[]
+    # Improved TOC matching with exact and substring matching
+    toc_matches = []
+    for h in headers:
+        norm_text = h['norm_text']
+        matching_toc_texts = []
+
+        # Check both exact matches and substring matches
+        for toc_norm, toc_text in toc_entries.items():
+            # Exact match case
+            if norm_text == toc_norm and len(toc_text)>4 and h['size']==max_size:
+                matching_toc_texts.append(toc_text)
+            # Substring match case (header is substring of TOC entry)
+            elif norm_text in toc_norm and len(toc_text)>4 and h['size']==max_size:
+                matching_toc_texts.append(toc_text)
+            # Substring match case (TOC entry is substring of header)
+            elif toc_norm in norm_text and len(toc_text)>4 and h['size']==max_size:
+                matching_toc_texts.append(toc_text)
+        
+        if matching_toc_texts and h['size'] >= max_size * 0.9:
+            best_match = max(matching_toc_texts,
+                          key=lambda x: (len(x), -len(x.replace(norm_text, ''))))
+            h['text'] = normalize_text(clean_toc_entry(best_match))
+            h['level'] = 0
+            if h['text'] not in  toc_text_match:
+              toc_matches.append(h)
+              toc_text_match.append(h['text'])
+        elif matching_toc_texts and h['size'] < max_size * 0.9 and h['size'] > nbsheadersize : # h['size'] < max_size * 0.9 and h['size'] > max_size*0.75:
+             print(h['text'],matching_toc_texts)
+             headers.remove(h)
+             continue
+       
+    
+    # Remove duplicates - keep only first occurrence of each level 0 header
+    unique_level0 = []
+    seen_level0 = set()
+    for h in toc_matches:
+        # Use the cleaned text for duplicate checking
+        cleaned_text = clean_toc_entry(h['text'])
+        norm_cleaned_text = normalize(cleaned_text)
+
+        if norm_cleaned_text not in seen_level0:
+            seen_level0.add(norm_cleaned_text)
+            # Update the header text with cleaned version
+            h['text'] = cleaned_text
+            unique_level0.append(h)
+            print(f"Added unique header: {cleaned_text} (normalized: {norm_cleaned_text})")
+
+    # Step 3: Process headers under each level 0 to identify level 1 format
+    
+    # First, group headers by their level 0 parent
+    level0_headers = [h for h in headers if h['level'] == 0]
+    header_groups = []
+    
+    for i, level0 in enumerate(level0_headers):
+        start_idx = headers.index(level0)
+        end_idx = headers.index(level0_headers[i+1]) if i+1 < len(level0_headers) else len(headers)
+        group = headers[start_idx:end_idx]
+        header_groups.append(group)
+
+    # Now process each group to identify level 1 format
+    for group in header_groups:
+        level0 = group[0]
+        level1_candidates = [h for h in group[1:] if h['level'] == -1]
+        
+        if not level1_candidates:
+            continue
+            
+        # The first candidate is our reference level 1
+        first_level1 = level1_candidates[0]
+        level1_format = {
+            'font': first_level1['font'],
+            'color': first_level1['color'],
+            'starts_with_number': is_numbered(first_level1['text']),
+            'size': first_level1['size'],
+            'bold': first_level1['bold']
+            # 'italic': first_level1['italic']
+        }
+        
+        # Assign levels based on the reference format
+        for h in level1_candidates:
+            current_format = {
+                'font': h['font'],
+                'color': h['color'],
+                'starts_with_number': is_numbered(h['text']),
+                'size': h['size'],
+                'bold': h['bold']
+                # 'italic': h['italic']
+            }
+            
+            # Compare with level1 format
+            if (current_format['font'] == level1_format['font'] and
+                current_format['color'] == level1_format['color'] and
+                current_format['starts_with_number'] == level1_format['starts_with_number'] and
+                abs(current_format['size'] - level1_format['size']) <= 0.1 and
+                current_format['bold'] == level1_format['bold'] ): #and
+                # current_format['italic'] == level1_format['italic']):
+                h['level'] = 1
+            else:
+                h['level'] = 2
+
+    # Step 4: Assign levels to remaining unassigned headers
+    unassigned = [h for h in headers if h['level'] == -1]
+    if unassigned:
+        # Cluster by size with tolerance
+        sizes = sorted({h['size'] for h in unassigned}, reverse=True)
+        clusters = []
+
+        for size in sizes:
+            found_cluster = False
+            for cluster in clusters:
+                if abs(size - cluster['size']) <= max(size, cluster['size']) * 0.1:
+                    cluster['headers'].extend([h for h in unassigned if abs(h['size'] - size) <= size * 0.1])
+                    found_cluster = True
+                    break
+            if not found_cluster:
+                clusters.append({
+                    'size': size,
+                    'headers': [h for h in unassigned if abs(h['size'] - size) <= size * 0.1]
+                })
+
+        # Assign levels starting from 1
+        clusters.sort(key=lambda x: -x['size'])
+        for i, cluster in enumerate(clusters):
+            for h in cluster['headers']:
+                base_level = i + 1
+                if h['bold']:
+                    base_level = max(1, base_level - 1)
+                h['level'] = base_level
+
+    # Step 5: Build hierarchy
+    root = []
+    stack = []
+
+    # Create a set of normalized texts from unique_level0 to avoid duplicates
+    unique_level0_texts = {h['norm_text'] for h in unique_level0}
+
+    # Filter out any headers from the original list that match unique_level0 headers
+    filtered_headers = []
+    for h in headers:
+        if h['norm_text'] in unique_level0_texts and h not in unique_level0:
+              h['level'] = 0
+        filtered_headers.append(h)
+
+    # Combine all headers - unique_level0 first, then the filtered headers
+    all_headers = unique_level0 + filtered_headers
+    all_headers.sort(key=lambda h: (h['page'], h['y']))
+
+    # Track which level 0 headers we've already added
+    added_level0 = set()
+
+    for header in all_headers:
+        if header['level'] < 0:
+            continue
+
+        if header['level'] == 0:
+            norm_text = header['norm_text']
+            if norm_text in added_level0:
+                continue
+            added_level0.add(norm_text)
+
+        # Pop stack until we find a parent
+        while stack and stack[-1]['level'] >= header['level']:
+            stack.pop()
+
+        current_parent = stack[-1] if stack else None
+
+        if current_parent:
+            current_parent['children'].append(header)
+        else:
+            root.append(header)
+
+        stack.append(header)
+
+    # Step 6: Enforce proper nesting
+    def enforce_nesting(node_list, parent_level=-1):
+        for node in node_list:
+            if node['level'] <= parent_level:
+                node['level'] = parent_level + 1
+            enforce_nesting(node['children'], node['level'])
+
+    enforce_nesting(root)
+    root = [h for h in root if not (h['level'] == 0 and not h['children'])]
+        # NEW: Filter out level 1 headers containing 'installation' and their children
+    def filter_installation_headers(node_list):
+        filtered = []
+        for node in node_list:
+            # Skip if it's a level 1 header containing 'installation' (case insensitive)
+            if node['level'] == 1 and ('installation' in node['text'].lower() or 'execution' in node['text'].lower() or 'miscellaneous items' in node['text'].lower() ) :
+                continue
+            # Recursively filter children
+            node['children'] = filter_installation_headers(node['children'])
+            filtered.append(node)
+        return filtered
+    
+    root = filter_installation_headers(root)
+    return root
+
+def adjust_levels_if_level0_not_in_toc(doc, toc_pages, root):
+    def normalize(text):
+        return re.sub(r'\s+', ' ', text.strip().lower())
+
+    toc_text = ""
+    for pno in toc_pages:
+        page = doc.load_page(pno)
+        toc_text += page.get_text()
+    toc_text_normalized = normalize(toc_text)
+
+    def is_level0_in_toc_text(header):
+        return header['level'] == 0 and normalize(header['text']) in toc_text_normalized
+
+    if any(is_level0_in_toc_text(h) for h in root):
+        return  # No change needed
+
+    def increase_levels(node_list):
+        for node in node_list:
+            node['level'] += 1
+            increase_levels(node['children'])
+
+def assign_numbers_to_headers(headers, prefix=None):
+    for idx, header in enumerate(headers, 1):
+        current_number = f"{prefix}.{idx}" if prefix else str(idx)
+        header["number"] = current_number
+        assign_numbers_to_headers(header["children"], current_number)
+
+def print_tree_with_numbers(headers, listofheaders, indent=0):
+    for header in headers:
+        size_info = f"size:{header['original_size']:.1f}" if 'original_size' in header else ""
+        line = (
+            "  " * indent +
+            f"{header.get('number', '?')} {header['text']} " +
+            f"(Level {header['level']}, p:{header['page']+1}, {size_info})"
+        )
+        print(line)
+        listofheaders.append(line)
+        print_tree_with_numbers(header["children"], listofheaders, indent + 1)
+    return listofheaders
+    
+def get_toc_page_numbers(doc, max_pages_to_check=15):
+    toc_pages = []
+    for page_num in range(min(len(doc), max_pages_to_check)):
+        page = doc.load_page(page_num)
+        blocks = page.get_text("dict")["blocks"]
+
+        dot_line_count = 0
+        for block in blocks:
+            for line in block.get("lines", []):
+                line_text = get_spaced_text_from_spans(line["spans"]).strip()
+                if dot_pattern.search(line_text):
+                    dot_line_count += 1
+
+        if dot_line_count >= 3:
+            toc_pages.append(page_num)
+
+    return list(range(0, toc_pages[-1] +1)) if toc_pages else toc_pages
+
+    
+def headersfrompdf(filePath):
+    pdf_path=filePath
+    if pdf_path and ('http' in pdf_path or 'dropbox' in pdf_path):
+        pdf_path = pdf_path.replace('dl=0', 'dl=1')
+    
+    response = requests.get(pdf_path)
+    pdf_content = BytesIO(response.content)
+    if not pdf_content:
+        raise ValueError("No valid PDF content found.")
+    
+    doc = fitz.open(stream=pdf_content, filetype="pdf")
+    most_common_font_size, most_common_color, most_common_font = get_regular_font_size_and_color(doc)
+    toc_pages = get_toc_page_numbers(doc)
+    hierarchy = build_header_hierarchy(doc,toc_pages, most_common_font_size, most_common_color, most_common_font)
+    assign_numbers_to_headers(hierarchy)
+    listofheaders=print_tree_with_numbers(hierarchy,listofheaders=[])
+    print(listofheaders)
+    return listofheaders
+
+
+