import gradio as gr from PIL import Image, ImageDraw, ImageFont import xml.etree.ElementTree as ET import os # --- Helper Functions --- def get_alto_namespace(xml_file_path): try: tree = ET.parse(xml_file_path) root = tree.getroot() if '}' in root.tag: return root.tag.split('}')[0] + '}' except ET.ParseError: print(f"Error parsing XML to find namespace: {xml_file_path}") return '' def parse_alto_xml(xml_file_path): full_text_lines = [] ocr_data = [] if not xml_file_path or not os.path.exists(xml_file_path): return "Error: XML file not provided or does not exist.", [] try: ns_prefix = get_alto_namespace(xml_file_path) tree = ET.parse(xml_file_path) root = tree.getroot() for text_line in root.findall(f'.//{ns_prefix}TextLine'): line_text_parts = [] for string_element in text_line.findall(f'{ns_prefix}String'): text = string_element.get('CONTENT') if text: line_text_parts.append(text) try: hpos = int(float(string_element.get('HPOS'))) vpos = int(float(string_element.get('VPOS'))) width = int(float(string_element.get('WIDTH'))) height = int(float(string_element.get('HEIGHT'))) ocr_data.append({ 'text': text, 'x': hpos, 'y': vpos, 'w': width, 'h': height }) except (ValueError, TypeError) as e: print(f"Warning: Could not parse coordinates for '{text}': {e}") ocr_data.append({ 'text': text, 'x': 0, 'y': 0, 'w': 10, 'h': 10 }) if line_text_parts: full_text_lines.append(" ".join(line_text_parts)) return "\n".join(full_text_lines), ocr_data except ET.ParseError as e: return f"Error parsing XML: {e}", [] except Exception as e: return f"An unexpected error occurred during XML parsing: {e}", [] def draw_ocr_on_image(image_pil, ocr_data): """ Draws styled bounding boxes and text from ocr_data onto the image. """ if not image_pil or not ocr_data: return image_pil draw = ImageDraw.Draw(image_pil) for item in ocr_data: x, y, w, h = item['x'], item['y'], item['w'], item['h'] text = item['text'] # 1. Draw bounding box for the original OCR segment draw.rectangle([(x, y), (x + w, y + h)], outline="rgba(255,0,0,190)", width=1) # Red, bit transparent # 2. Determine font for this specific item # Adaptive font size based on item height, with min/max caps # Making it slightly smaller relative to box height for overlay text current_font_size = max(8, min(16, int(item['h'] * 0.60))) is_truetype = False try: font = ImageFont.truetype("arial.ttf", current_font_size) is_truetype = True except IOError: try: font = ImageFont.truetype("DejaVuSans.ttf", current_font_size) is_truetype = True except IOError: font = ImageFont.load_default() # Small bitmap font, size not controllable # print("Arial and DejaVuSans fonts not found, using default PIL font for some items.") # 3. Get actual text dimensions for precise placement text_actual_width = 0 text_actual_height = 0 text_draw_origin_is_top_left = False # Flag to know how to use coordinates with draw.text try: # Modern Pillow: use textbbox with anchor for precise bounds # Using 'lt' (left-top) anchor means (0,0) in textbbox is relative to top-left of text bbox = font.getbbox(text, anchor='lt') text_actual_width = bbox[2] - bbox[0] text_actual_height = bbox[3] - bbox[1] text_draw_origin_is_top_left = True # draw.text with anchor='lt' will use xy as top-left except (AttributeError, TypeError): # Fallback for older Pillow or if anchor not supported by getbbox try: # Try font.getsize (deprecated but widely available) size = font.getsize(text) text_actual_width = size[0] if is_truetype: # For TrueType, getsize height is usually baseline to top ascent, descent = font.getmetrics() text_actual_height = ascent + descent # Full line height else: # For default font, getsize height is total height text_actual_height = size[1] except AttributeError: # Ultimate fallback if getsize also fails (unlikely for loaded font) text_actual_width = len(text) * current_font_size // 2 # Very rough estimate text_actual_height = current_font_size # Very rough estimate # 4. Calculate position for the TOP-LEFT of the overlay text buffer = 3 # Buffer in pixels between OCR box and text overlay text_draw_x = x # Align left edge of text with left edge of box # Try to place text above the box tentative_text_top_y = y - text_actual_height - buffer if tentative_text_top_y < buffer: # If it goes off (or too close to) the top of the image tentative_text_top_y = y + h + buffer # Place it below the box # Ensure it doesn't go off the bottom either if tentative_text_top_y + text_actual_height > image_pil.height - buffer: tentative_text_top_y = image_pil.height - text_actual_height - buffer # Pin to bottom if tentative_text_top_y < buffer: # If image is too short for text, pin to top tentative_text_top_y = buffer final_text_top_y = tentative_text_top_y # 5. Draw background for the overlay text bg_padding = 2 bg_x0 = text_draw_x - bg_padding bg_y0 = final_text_top_y - bg_padding bg_x1 = text_draw_x + text_actual_width + bg_padding bg_y1 = final_text_top_y + text_actual_height + bg_padding draw.rectangle([(bg_x0, bg_y0), (bg_x1, bg_y1)], fill="rgba(255, 255, 220, 235)") # Light yellow, fairly opaque # 6. Draw the overlay text draw_coords = (text_draw_x, final_text_top_y) if text_draw_origin_is_top_left: try: draw.text(draw_coords, text, fill="black", font=font, anchor='lt') except (TypeError, AttributeError): # Fallback if anchor='lt' fails at draw time # This fallback means background might be slightly off if is_truetype and text_draw_origin_is_top_left was True due to getbbox working # but draw.text doesn't support anchor. For simplicity, draw at top_y assuming it's baseline. ascent = font.getmetrics()[0] if is_truetype else text_actual_height draw.text((text_draw_x, final_text_top_y + ascent if is_truetype else final_text_top_y), text, fill="black", font=font) else: # Older Pillow, (x,y) is baseline for TrueType, top-left for default if is_truetype: ascent, _ = font.getmetrics() draw.text((text_draw_x, final_text_top_y + ascent), text, fill="black", font=font) else: # Default font, (x,y) is top-left draw.text((text_draw_x, final_text_top_y), text, fill="black", font=font) return image_pil # --- Gradio Interface Function --- def process_image_and_xml(image_path, xml_path, show_overlay): if image_path is None: return None, "Please upload an image.", None try: img_pil = Image.open(image_path).convert("RGB") except Exception as e: return None, f"Error loading image: {e}", None if xml_path is None: return img_pil, "Please upload an OCR XML file.", None extracted_text, ocr_box_data = parse_alto_xml(xml_path) overlay_image_pil = None if show_overlay: if ocr_box_data: img_for_overlay = img_pil.copy() overlay_image_pil = draw_ocr_on_image(img_for_overlay, ocr_box_data) elif not (isinstance(extracted_text, str) and extracted_text.startswith("Error")): if isinstance(extracted_text, str): extracted_text += "\n(No bounding box data for overlay)" else: extracted_text = "(No bounding box data for overlay)" return img_pil, extracted_text, overlay_image_pil # --- Create Gradio App --- with gr.Blocks(theme=gr.themes.Soft()) as demo: gr.Markdown("# OCR Viewer (ALTO XML)") gr.Markdown( "Upload an image and its corresponding ALTO OCR XML file. " "The app will display the image, extract and show the plain text, " "and optionally overlay the OCR predictions on the image." ) with gr.Row(): with gr.Column(scale=1): image_input = gr.File(label="Upload Image (PNG, JPG, etc.)", type="filepath") xml_input = gr.File(label="Upload ALTO XML File (.xml)", type="filepath") show_overlay_checkbox = gr.Checkbox(label="Show OCR Overlay on Image", value=False) submit_button = gr.Button("Process Files", variant="primary") with gr.Row(): with gr.Column(scale=1): output_image_orig = gr.Image(label="Uploaded Image", type="pil", interactive=False) with gr.Column(scale=1): output_text = gr.Textbox(label="Extracted Plain Text", lines=15, interactive=False) output_image_overlay = gr.Image(label="Image with OCR Overlay", type="pil", interactive=False, visible=True) def update_interface(image_filepath, xml_filepath, show_overlay_val): if image_filepath is None and xml_filepath is None: return None, "Please upload an image and an XML file.", None img, text, overlay_img = process_image_and_xml(image_filepath, xml_filepath, show_overlay_val) return img, text, overlay_img submit_button.click( fn=update_interface, inputs=[image_input, xml_input, show_overlay_checkbox], outputs=[output_image_orig, output_text, output_image_overlay] ) show_overlay_checkbox.change( fn=update_interface, inputs=[image_input, xml_input, show_overlay_checkbox], outputs=[output_image_orig, output_text, output_image_overlay] ) gr.Markdown("---") gr.Markdown("### Example ALTO XML Snippet (for `String` element extraction):") gr.Code( value=""" ... ... """, interactive=False ) if __name__ == "__main__": try: img_test = Image.new('RGB', (2394, 3612), color = 'lightgray') img_test.save("dummy_image.png") print("Created dummy_image.png for testing.") example_xml_filename = "189819724.34.xml" if not os.path.exists(example_xml_filename): print(f"WARNING: Example XML '{example_xml_filename}' not found. Please create it (using the content from the prompt) or upload your own.") except ImportError: print("Pillow not installed, can't create dummy image.") except Exception as e: print(f"Error during setup: {e}") demo.launch()