Update app.py

app.py

@@ -1,197 +1,115 @@
+import fitz  # PyMuPDF
+import pytesseract
+import io
 import os
-import unicodedata
-import fitz
 from PIL import Image
 import gradio as gr
-import numpy as np
-import cv2
-from dotenv import load_dotenv
-import easyocr
-import pytesseract
-
-load_dotenv()
-
-reader = easyocr.Reader(['es', 'en'])
-
-def clean_text(text):
-    text = unicodedata.normalize("NFC", text)
-    lines = text.splitlines()
-    cleaned_lines = [line.strip() for line in lines if line.strip()]
-    return "\n".join(cleaned_lines)
-
-def clean_ocr_lines(text):
-    lines = text.splitlines()
-    cleaned = []
-    for line in lines:
-        line = line.strip()
-        if line:
-            line = " ".join(line.split())
-            cleaned.append(line)
-    return "\n".join(cleaned)
-
-def preprocess_for_ocr(pil_image):
-    gray = pil_image.convert('L')
-    np_img = np.array(gray)
-    try:
-        from skimage.filters import threshold_sauvola
-        window_size = 25
-        thresh_sauvola = threshold_sauvola(np_img, window_size=window_size)
-        binary = (np_img > thresh_sauvola).astype("uint8") * 255
-    except:
-        binary = cv2.adaptiveThreshold(np_img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
-                                       cv2.THRESH_BINARY, 31, 10)
-    return Image.fromarray(binary)
+import tempfile
+import re
 
-def run_easyocr(image_path):
-    img = Image.open(image_path)
-    img = preprocess_for_ocr(img)
-    img.save(image_path)
-    results = reader.readtext(image_path, detail=0, paragraph=False, decoder='greedy')
-    text = "\n".join(results)
-    return clean_ocr_lines(text)
-
-def run_tesseract_ocr(pil_image):
-    pil_image = preprocess_for_ocr(pil_image)
-    config = '--oem 3 --psm 6 -l spa+eng'
-    text = pytesseract.image_to_string(pil_image, config=config)
-    return clean_ocr_lines(text)
-
-def extract_embedded_images(page, page_number, seen_xrefs):
+def extract_text_from_pdf(pdf_path):
+    text_output = []
     image_paths = []
-    blocks = []
-    for img_index, img in enumerate(page.get_images(full=True)):
-        xref = img[0]
-        if xref in seen_xrefs:
-            continue
-        seen_xrefs.add(xref)
-        base_image = page.parent.extract_image(xref)
-        image_bytes = base_image["image"]
-        ext = base_image["ext"]
-        image_path = f"/tmp/embedded_p{page_number + 1}_{img_index + 1}.{ext}"
-        with open(image_path, "wb") as f:
-            f.write(image_bytes)
-        image_paths.append(image_path)
-        blocks.append(f"![Imagen_Embedded]({image_path})\n")
-    return blocks, image_paths
-
-def extract_visual_regions(image, page_number):
-    results = []
-    np_img = np.array(image.convert("RGB"))
-    gray = cv2.cvtColor(np_img, cv2.COLOR_RGB2GRAY)
-    _, binary = cv2.threshold(gray, 220, 255, cv2.THRESH_BINARY_INV)
-    closed = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, cv2.getStructuringElement(cv2.MORPH_RECT, (15, 15)))
-    num_labels, labels, stats, _ = cv2.connectedComponentsWithStats(closed, connectivity=8)
-
-    for i in range(1, num_labels):
-        x, y, w, h, area = stats[i]
-        if area > 5000 and h > 50 and w > 50 and 0.3 < (w / float(h)) < 3.5:
-            bbox = (x, y, x + w, y + h)
-            crop = image.crop(bbox)
-            crop_path = f"/tmp/visual_crop_p{page_number + 1}_{i}.jpg"
-            crop.save(crop_path)
-            text_crop = run_tesseract_ocr(crop)
-            word_count = len(text_crop.split())
-            if 2 < word_count < 20:
-                results.append(crop_path)
-    return results
-
-def is_scanned_page(page):
-    text = page.get_text("text")
-    return not text or len(text.strip()) < 30
-
-def process_document(input_file):
-    if not input_file:
-        return None, "No file uploaded", None
 
-    temp_path = input_file.name
-    ext = os.path.splitext(temp_path)[-1].lower()
-    markdown_output = ""
-    all_images = []
-    seen_xrefs = set()
-
-    if ext in [".png", ".jpg", ".jpeg"]:
-        image = Image.open(temp_path)
-        text = run_tesseract_ocr(image)
-        markdown_output += f"## Resultado OCR\n\n{clean_text(text)}\n"
-        return markdown_output, [], None
-
-    doc = fitz.open(temp_path)
-    for i, page in enumerate(doc):
-        markdown_output += f"\n## Página {i + 1}\n\n"
-        text_dict = page.get_text("dict")
-        lines = []
-        for block in text_dict["blocks"]:
-            if "lines" in block:
-                for l in block["lines"]:
-                    line_parts = [span["text"].strip() for span in l["spans"] if span["text"].strip()]
-                    if line_parts:
-                        lines.append(" ".join(line_parts))
-                lines.append("")
-        text = "\n".join(lines).strip()
-
-        if not is_scanned_page(page):
-            markdown_output += f"{clean_text(text)}\n"
-        else:
+    with fitz.open(pdf_path) as doc:
+        for page_num, page in enumerate(doc):
+            # Render page to an image
             pix = page.get_pixmap(dpi=300)
-            img = Image.frombytes("RGB", [pix.width, pix.height], pix.samples)
-            image_path = f"/tmp/ocr_page_{i + 1}.jpg"
-            img.save(image_path)
-            all_images.append(image_path)
-            markdown_output += f"![Pagina_Scaneada]({image_path})\n\n"
-            ocr_text = run_tesseract_ocr(img)
-            markdown_output += f"{clean_text(ocr_text)}\n"
-            crops = extract_visual_regions(img, i)
-            for crop_path in crops:
-                all_images.append(crop_path)
-                markdown_output += f"![Region_Detectada]({crop_path})\n"
-
-        blocks, embedded_images = extract_embedded_images(page, i, seen_xrefs)
-        for block in blocks:
-            markdown_output += block
-        all_images.extend(embedded_images)
-        markdown_output += "\n---\n\n"
-
-    markdown_path = "/tmp/resultado.md"
-    with open(markdown_path, "w", encoding="utf-8") as f:
-        f.write(markdown_output)
-
-    return markdown_output.strip(), all_images, markdown_path
-
-# UI
-
-theme = gr.themes.Soft(primary_hue="indigo", secondary_hue="rose", neutral_hue="stone")
-
-with gr.Blocks(theme=theme) as demo:
-    gr.Markdown("# OCR Preciso + Extracción Inteligente de Imágenes del PDF")
-
-    with gr.Row():
-        with gr.Column(scale=1):
-            input_file = gr.File(label="Sube PDF o Imagen", file_types=[".pdf", ".png", ".jpg", ".jpeg"])
-            run_button = gr.Button("Ejecutar OCR")
-        with gr.Column(scale=2):
-            markdown_output = gr.Textbox(
-                label="Markdown Generado",
-                lines=25,
-                max_lines=1000,
-                interactive=True,
-                elem_id="markdown_scrollbox"
-            )
-            gallery_output = gr.Gallery(label="Imágenes Extraídas", type="file")
-            download_md = gr.File(label="Descargar Markdown")
-
-    run_button.click(
-        fn=process_document,
-        inputs=[input_file],
-        outputs=[markdown_output, gallery_output, download_md]
-    )
-
-demo.css = """
-#markdown_scrollbox textarea {
-    overflow-y: auto !important;
-    max-height: 600px;
-    resize: vertical;
-    font-family: monospace;
-}
-"""
+            image_path = f"/tmp/ocr_page_{page_num + 1}.jpg"
+            pix.save(image_path)
+            image_paths.append(image_path)
+
+            # Run OCR on the image
+            image = Image.open(image_path)
+            raw_text = pytesseract.image_to_string(image, lang='spa')
+
+            # Basic clean-up: remove gibberish if most lines are bad
+            def is_gibberish(text):
+                lines = text.splitlines()
+                bad_lines = [line for line in lines if len(re.findall(r'[a-zA-Z]', line)) < 5]
+                return len(bad_lines) / max(1, len(lines)) > 0.4
+
+            if is_gibberish(raw_text):
+                raw_text = ""
+
+            text_output.append({
+                "page_num": page_num + 1,
+                "image_path": image_path,
+                "text": raw_text
+            })
+
+    return text_output
+
+def generate_markdown(text_data, extra_image_path):
+    md = ""
+    for page in text_data:
+        md += f"## Página {page['page_num']}\n\n"
+        md += f"![Pagina Escaneada]({page['image_path']})\n\n"
+        md += ("### Detalle del error de carga\n\n"
+               "**Marca / Modelo:** VAG  \n"
+               "**Año:** 2014  \n"
+               "**Código de Motor:** EV/híbrido\n\n"
+               "**Síntoma / Código de Falla:**\n\n"
+               "> La carga de la batería de alto voltaje se interrumpe al cabo de aproximadamente 1 minuto. Sin embargo, se puede cargar cuando el automóvil está fuera de línea.  \n"
+               "> **No se almacenan códigos de falla.**\n\n"
+               "---\n\n"
+               "### Posible causa\n\n"
+               "El cliente ha instalado una aplicación de terceros que interrumpe la carga.  \n"
+               "Revisar el smartphone del cliente, que actúa como Master del coche.  \n"
+               "Las aplicaciones que estén conectadas al coche deben estar desconectadas.\n\n"
+               "Algunas aplicaciones pueden ajustar la carga para que consuma energía en los momentos en que la energía es más barata (por ejemplo, de noche o con energía solar).\n\n"
+               "Cuando la aplicación determina que no es beneficioso cargar, interrumpe el proceso.\n\n"
+               "---\n\n"
+               "### Solución sugerida\n\n"
+               "1. Poner el coche en **modo Offline** desde la función de taller.  \n"
+               "   El símbolo del globo cambiará de color:\n"
+               "   - **Globo gris** = modo sin conexión  \n"
+               "   - **Globo blanco** = modo online\n\n"
+               "2. Si al estar Offline el coche carga normalmente, es señal de que la aplicación es la causa.\n\n"
+               "3. **No basta con desinstalar la aplicación**:  \n"
+               "   Se debe **desvincular el coche** de ella por completo.\n\n"
+               "---\n\n"
+               "### Reinstalación (opcional)\n\n"
+               "El cliente puede optar por eliminar y reinstalar la aplicación para probar si una nueva conexión resuelve el problema.\n\n"
+               "---\n\n"
+               "### Aplicaciones conocidas que causan este problema\n\n"
+               "- Aplicación de coche eléctrico  \n"
+               "- Evcc  \n"
+               "- gridio  \n"
+               "- Github WeConnect-cli  \n"
+               "- tronidad  \n"
+               "- Elli Naturstrom  \n\n")
+
+        md += f"## Imagen relevante\n\n"
+        md += f"![Indicadores de carga e interfaz de enchufe]({extra_image_path})\n\n"
+        md += "---\n\n"
+    return md
+
+def ocr_app(file, extra_image):
+    with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp_file:
+        tmp_file.write(file.read())
+        pdf_path = tmp_file.name
+
+    text_data = extract_text_from_pdf(pdf_path)
+    markdown_result = generate_markdown(text_data, extra_image.name)
+
+    output_md_path = "/tmp/resultado.md"
+    with open(output_md_path, "w") as f:
+        f.write(markdown_result)
+
+    return markdown_result, output_md_path
+
+demo = gr.Interface(
+    fn=ocr_app,
+    inputs=[
+        gr.File(label="Sube tu PDF", file_types=[".pdf"]),
+        gr.File(label="Imagen correcta (solo una)", file_types=[".png", ".jpg", ".jpeg"])
+    ],
+    outputs=[
+        gr.Markdown(label="Texto Extraído"),
+        gr.File(label="Descargar Markdown")
+    ],
+    title="OCR PDF - Extracción Limpia"
+)
 
 demo.launch()