#!/usr/bin/env python3 """ Google Cloud Vision API + Gemini Çeviri Vision: Kesin koordinatlar Gemini: Türkçe çeviri """ import fitz import json import base64 import requests import google.generativeai as genai # API Keys VISION_API_KEY = "AIzaSyCkYVKLxWDPNbQZr0-HM0nlSneEeM9KMjs" GEMINI_API_KEY = "AIzaSyCkYVKLxWDPNbQZr0-HM0nlSneEeM9KMjs" VISION_API_URL = f"https://vision.googleapis.com/v1/images:annotate?key={VISION_API_KEY}" def pdf_page_to_image(pdf_path: str, page_num: int = 0, dpi: int = 200): """PDF sayfasını PNG'ye çevir""" doc = fitz.open(pdf_path) page = doc.load_page(page_num) rect = page.rect page_width = rect.width page_height = rect.height mat = fitz.Matrix(dpi/72, dpi/72) pix = page.get_pixmap(matrix=mat) img_bytes = pix.tobytes("png") doc.close() return img_bytes, page_width, page_height def vision_ocr(image_bytes: bytes) -> dict: """Google Cloud Vision API ile OCR""" img_base64 = base64.b64encode(image_bytes).decode() request_body = { "requests": [{ "image": {"content": img_base64}, "features": [{"type": "DOCUMENT_TEXT_DETECTION"}] }] } response = requests.post( VISION_API_URL, headers={"Content-Type": "application/json"}, json=request_body ) if response.status_code != 200: print(f"Vision API Hatası: {response.status_code}") return None return response.json() def extract_blocks(vision_response: dict, dpi: int = 200): """Vision yanıtından metin bloklarını çıkar""" if not vision_response or "responses" not in vision_response: return [] response = vision_response["responses"][0] if "fullTextAnnotation" not in response: return [] blocks = [] scale = 72 / dpi for page in response["fullTextAnnotation"].get("pages", []): for block in page.get("blocks", []): vertices = block.get("boundingBox", {}).get("vertices", []) if len(vertices) < 4: continue x0 = min(v.get("x", 0) for v in vertices) * scale y0 = min(v.get("y", 0) for v in vertices) * scale x1 = max(v.get("x", 0) for v in vertices) * scale y1 = max(v.get("y", 0) for v in vertices) * scale # Metni birleştir text_parts = [] for paragraph in block.get("paragraphs", []): for word in paragraph.get("words", []): word_text = "".join(s.get("text", "") for s in word.get("symbols", [])) text_parts.append(word_text) text = " ".join(text_parts) if text.strip() and len(text) > 1: blocks.append({ "text": text, "bbox": [x0, y0, x1, y1] }) return blocks def translate_texts(blocks: list) -> list: """Gemini ile toplu çeviri""" # Çevrilecek metinleri hazırla texts = [b["text"] for b in blocks] prompt = f"""Bu teknik doküman metinlerini Türkçe'ye çevir. KURALLAR: - Teknik terimler: engine=motor, pump=pompa, valve=valf, brake=fren, sensor=sensör, controller=kontrol ünitesi - Hata kodları (111000 gibi) ve referansları (T2-2, T/M) ÇEVİRME - min⁻¹ = dev/dk - SECTION, Group başlıklarını çevir METİNLER: {json.dumps(texts, ensure_ascii=False, indent=2)} ÇIKTI: Sadece JSON array döndür, her metin için çeviri: ["çeviri1", "çeviri2", ...] """ genai.configure(api_key=GEMINI_API_KEY) model = genai.GenerativeModel('gemini-2.0-flash') response = model.generate_content(prompt) text = response.text try: if "```json" in text: json_str = text.split("```json")[1].split("```")[0].strip() elif "```" in text: json_str = text.split("```")[1].split("```")[0].strip() else: json_str = text.strip() translations = json.loads(json_str) # Çevirileri bloklara ekle for i, block in enumerate(blocks): if i < len(translations): block["turkish"] = translations[i] else: block["turkish"] = block["text"] return blocks except Exception as e: print(f"Çeviri parse hatası: {e}") for block in blocks: block["turkish"] = block["text"] return blocks def apply_translations(pdf_path: str, blocks: list, output_path: str, page_num: int = 0): """Çevirileri PDF'e uygula""" doc = fitz.open(pdf_path) page = doc.load_page(page_num) font_path = "/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf" applied = 0 for block in blocks: text = block.get("text", "") turkish = block.get("turkish", text) bbox = block.get("bbox", []) if not turkish or turkish == text or len(bbox) != 4: continue # Hata kodları ve referansları atla if any(c.isdigit() for c in text) and len(text) < 20: continue x0, y0, x1, y1 = bbox rect = fitz.Rect(x0, y0, x1, y1) # Maskeleme page.draw_rect(rect, color=(1, 1, 1), fill=(1, 1, 1)) # Font boyutu box_height = y1 - y0 box_width = x1 - x0 chars_per_line = max(10, int(box_width / 5)) lines_needed = max(1, (len(turkish) // chars_per_line) + 1) fontsize = min(10, max(5, (box_height / lines_needed) - 1)) # Metni yaz text_rect = fitz.Rect(x0 + 1, y0 + 1, x1 - 1, y1 - 1) try: page.insert_textbox( text_rect, turkish, fontsize=fontsize, fontname="dejavu", fontfile=font_path, color=(0.1, 0.1, 0.1), align=fitz.TEXT_ALIGN_LEFT ) applied += 1 except: pass doc.save(output_path) doc.close() return applied def main(): input_pdf = "/var/www/html/PEPCVSON/public/katalog/api/output/ZW140-5B_sayfa_315.pdf" output_pdf = "/var/www/html/PEPCVSON/public/katalog/api/output/ZW140-5B_sayfa_315_VISION.pdf" print("📄 1/4 PDF görüntüye dönüştürülüyor...") img_bytes, page_width, page_height = pdf_page_to_image(input_pdf, 0, dpi=200) print("🔍 2/4 Vision API ile OCR yapılıyor...") vision_result = vision_ocr(img_bytes) if not vision_result: print("❌ Vision API başarısız") return print("📊 3/4 Metin blokları çıkarılıyor ve çevriliyor...") blocks = extract_blocks(vision_result, dpi=200) print(f" → {len(blocks)} blok bulundu") blocks = translate_texts(blocks) print("✍️ 4/4 PDF'e yazılıyor...") applied = apply_translations(input_pdf, blocks, output_pdf, 0) print(f"\n✅ Tamamlandı! {applied} blok çevrildi") print(f" Çıktı: {output_pdf}") # Örnek göster print("\n📋 Örnek çeviriler:") for block in blocks[:5]: print(f" EN: {block['text'][:40]}...") print(f" TR: {block.get('turkish', '')[:40]}...") print() if __name__ == "__main__": main()