Model Integration

Dockerfile

@@ -1,24 +1,23 @@
-# Gunakan dasar sistem operasi Linux dengan Python 3.9
-FROM python:3.9
+FROM python:3.9-slim
 
-# Instal dependensi sistem yang dibutuhkan oleh OpenCV
 RUN apt-get update && apt-get install -y \
     libgl1 \
-    libglib2.0-0
+    libglib2.0-0 \
+    libgomp1 \
+    libsm6 \
+    libxext6 \
+    && rm -rf /var/lib/apt/lists/*
 
-# Buat user baru (Syarat keamanan dari Hugging Face)
+# Buat user baru
 RUN useradd -m -u 1000 user
 USER user
 ENV PATH="/home/user/.local/bin:$PATH"
 
 WORKDIR /app
 
-# Salin file requirements dan instal
-COPY --chown=user requirements.txt requirements.txt
+COPY --chown=user requirements.txt .
 RUN pip install --no-cache-dir --upgrade -r requirements.txt
 
-# Salin semua kode dan model kamu ke dalam kontainer
-COPY --chown=user . /app
+COPY --chown=user . .
 
-# Jalankan server FastAPI di port 7860 (Port standar Hugging Face)
 CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

app.py

@@ -3,22 +3,23 @@ from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import JSONResponse
 from paddleocr import PaddleOCR
 from datetime import datetime
+from rapidfuzz import fuzz
 import cv2
 import re
 import numpy as np
-import random
+import requests
 
-
-app = FastAPI(title="OCR Service")
+app = FastAPI(title="OCR & Extraction Service")
 
 app.add_middleware(
     CORSMiddleware,
-    allow_origins=["*"], # ganti dengan URL Vercel
+    allow_origins=["*"], 
     allow_credentials=True,
     allow_methods=["*"],
     allow_headers=["*"],
 )
 
+# Inisialisasi PaddleOCR
 paddle_engine = PaddleOCR(
     lang='id',
     show_log=False,
@@ -26,10 +27,13 @@ paddle_engine = PaddleOCR(
     enable_mkldnn=True,
     cpu_threads=2,
     ocr_version="PP-OCRv4",
-    det_limit_side_len=960
+    det_limit_side_len=640
 )
 
-#FUNGSI EKSTRAKSI TANGGAL
+SPACE_B_URL = "https://struklyai-clasify.hf.space/classify/"
+
+# ENGINE EKSTRAKSI (RULE-BASED + REGEX + FUZZY)
+
 MONTH_MAP = {
     "jan": 1, "januari": 1,
     "feb": 2, "februari": 2,
@@ -47,10 +51,7 @@ MONTH_MAP = {
 }
 
 def extract_transaction_date(raw_text: str):
-    # OCR cleanup
-    raw_text = raw_text.replace("O", "0").replace("I", "1")
     raw_text = re.sub(r"\s+", " ", raw_text)
-
     patterns = [
         # dd/mm/yyyy | dd-mm-yyyy | dd.mm.yyyy
         r'\d{2}[\/\-.]\d{2}[\/\-.]\d{4}',
@@ -67,55 +68,226 @@ def extract_transaction_date(raw_text: str):
         # 31 Feb 25 / 31 Februari 2025
         r'\d{1,2}\s[a-zA-Z]+\s\d{2,4}'
     ]
-
     numeric_formats = [
         "%d/%m/%Y", "%d-%m-%Y", "%d.%m.%Y",
         "%d/%m/%y", "%d-%m-%y", "%d.%m.%y",
         "%Y/%m/%d", "%Y-%m-%d", "%Y.%m.%d",
         "%Y%m%d"
     ]
-
     candidates = []
 
     for pattern in patterns:
         matches = re.findall(pattern, raw_text, re.IGNORECASE)
-
         for text in matches:
             dt = None
-            # numeric formats
             for fmt in numeric_formats:
                 try:
                     dt = datetime.strptime(text, fmt)
                     break
                 except:
                     continue
-
-            # text month format
             if not dt:
-                match = re.match(
-                    r'(\d{1,2})\s([a-zA-Z]+)\s(\d{2,4})',
-                    text,
-                    re.IGNORECASE
-                )
+                match = re.match(r'(\d{1,2})\s([a-zA-Z]+)\s(\d{2,4})', text, re.IGNORECASE)
                 if match:
                     day, month_txt, year = match.groups()
                     month = MONTH_MAP.get(month_txt.lower())
-
                     if month:
                         year = int(year)
-                        if year < 100:
-                            year += 2000
+                        if year < 100: year += 2000
                         dt = datetime(year, month, int(day))
-
             if dt:
                 candidates.append(dt)
 
-    if not candidates:
-        return None
-
+    if not candidates: return None
     best_date = max(candidates)
     return best_date.strftime("%d-%m-%Y")
 
+def clean(text):
+    return text.replace("  ", " ")
+
+def extract_number(line):
+    match = re.search(r'\d{1,3}(?:[.,]\d{3})+', line)
+    return match.group() if match else None
+
+def has_money(line):
+    return bool(extract_number(line))
+
+def is_noise(line):
+    upper = line.upper().strip()
+
+    noise_keywords = [
+        "TOTAL", "SUBTOTAL", "CASH", "KEMBALI", "CHANGE", "BAYAR", "PAID",
+        "ITEM", "QTY", "TERIMA", "KASIH", "THANK", "PELANGGAN",
+        "STAFF", "KASSA", "NPWP", "TRANS", "DELIVERY",
+        "DISCOUNT", "DISC", "ROUNDED", "DEBIT", "CREDIT",
+        "PPN", "DPP", "TUNAI", "HEMAT",
+        "UNBOXING", "COMPLAIN", "VIDEO",
+        "WWW.", "HTTP",
+        "CASHIER", "MOHON"
+    ]
+
+    address_keywords = [
+        "JL", "JLN", "JALAN", "RAYA", "KEL", "KEC", "KAB",
+        "KOTA", "RT", "RW","BARAT", "TIMUR",
+        "SELATAN", "UTARA", "TENGAH"
+    ]
+
+    promo_keywords = [
+        "PT ", "/PT", "CV ", "PROMO", "VOUCHER", "MEMBER"
+    ]
+
+    if any(k in upper for k in noise_keywords):
+        return True
+
+    if any(k in upper for k in address_keywords):
+        return True
+
+    if any(k in upper for k in promo_keywords):
+        return True
+
+    if re.fullmatch(r'[\d.,:=-]+', upper):
+        return True
+
+    if re.search(r'\{?\d+\}?X\d+', upper.replace(" ", "")):
+        return True
+
+    if re.search(r'\d+\s*X\s*@?', upper):
+        return True
+
+    if re.search(r'\bRP\b', upper):
+        return True
+
+    return False
+
+def is_product_like(line):
+    return (
+        bool(re.search(r'\d+\s*(PCS|BKS|BTL|@)', line.upper())) or
+        bool(re.search(r'\d{1,3}[.,]\d{3}', line))
+    )
+
+def is_noise_merc(line):
+    upper = line.upper()
+    noise_keywords = [
+        "TOTAL", "CASH", "KEMBALI", "CHANGE", "BAYAR", "PAID", "ITEM", "QTY",
+        "TERIMA", "KASIH", "THANK", "PELANGGAN", "STAFF", "KASSA",
+        "NPWP", "HP", "TRANS", "BISA DELIVERY"
+    ]
+    if any(k in upper for k in noise_keywords): return True
+    if re.fullmatch(r'[\d.,:]+', line.strip()): return True
+    if re.fullmatch(r'\d+\w*@', upper.replace(" ", "")): return True
+    if "." in line and len(line.split()) == 1: return True
+    if re.search(r'^[A-Z0-9]{8,}$', upper): return True
+    return False
+
+
+def detect_merchant(lines):
+    header = lines[:8]
+    candidates = []
+    for line in header:
+        if not line: continue
+        line = line.strip("= ").strip()
+        if is_noise_merc(line) or is_product_like(line): continue
+        if len(line.split()) >= 8: continue
+        score = 0
+        if line.isupper(): score += 2
+        if 5 <= len(line) <= 35: score += 2
+        if not any(c.isdigit() for c in line): score += 2
+        candidates.append((score, line))
+    
+    if not candidates: return "Tidak Terdeteksi"
+    candidates.sort(reverse=True)
+    return candidates[0][1]
+
+def is_item(line):
+    if not line:
+        return False
+
+    if is_noise(line):
+        return False
+
+    if len(line.split()) < 2:
+        return False
+
+    letters = sum(c.isalpha() for c in line)
+    alpha_ratio = letters / max(len(line), 1)
+
+    if letters < 5:
+        return False
+
+    if alpha_ratio < 0.35:
+        return False
+
+    return True
+
+
+def find_item_zone(lines):
+    start = 0
+    for i, l in enumerate(lines):
+        if re.search(r'\d+\s*X@|@\d|\d{1,3}[.,]\d{3}', l.upper()):
+            start = max(0, i - 1)
+            break
+    return start
+
+def extract_items(lines):
+    result = []
+
+    for i, line in enumerate(lines):
+        line = line.strip()
+
+        m = re.search(r'^\{?\d+[.,]?\d*\}?[Xx]@?[\d.,]+(.+)$', line)
+        if m:
+            item = m.group(1).strip(" =-")
+            if is_item(item):
+                result.append(item)
+            continue
+
+        if is_item(line):
+            next_lines = lines[i+1:i+5]
+
+            numeric_count = sum(
+                bool(re.fullmatch(r'[\d.,X@x=-]+', x.strip()))
+                for x in next_lines
+            )
+
+            if numeric_count >= 2:
+                result.append(line)
+
+    return list(dict.fromkeys(result))
+
+def extract_receipt(text):
+    text = clean(text)
+    lines = [l.strip() for l in text.split("\n") if l.strip()]
+    return {
+        "items": extract_items(lines),
+    }
+
+def detect_total(lines):
+    bad = ["TUNAI", "CASH", "KEMBALI", "CHANGE", "BAYAR", "PAID"]
+    for i, line in enumerate(lines):
+        upper = line.upper().replace(" ", "")
+        if any(b in upper for b in bad): continue
+        if "TOTAL" in upper:
+            for j in range(i, min(i + 6, len(lines))):
+                num = extract_number(lines[j])
+                if num: return num
+
+    for line in reversed(lines):
+        if not any(b in line.upper() for b in bad):
+            num = extract_number(line)
+            if num: return num
+    return "0"
+
+def extract_receipt(text):
+    text = clean(text)
+    lines = [l.strip() for l in text.split("\n") if l.strip()]
+    return {
+        "merchant": detect_merchant(lines),
+        "items": extract_items(lines),
+        "total": detect_total(lines)
+    }
+
+# API ENDPOINTS
+
 @app.get("/")
 def home():
     return {"status": "Optima Team - OCR API Aktif :D !"}
@@ -132,15 +304,44 @@ async def proses_gambar_paddle(file: UploadFile = File(...)):
         teks_terbaca = []
         if result and result[0]:
             for baris in result[0]:
-                teks = baris[1][0]
-                teks_terbaca.append(teks)
+                teks_terbaca.append(baris[1][0])
                 
         ocr_mentah = "\n".join(teks_terbaca)
+        
         date = extract_transaction_date(ocr_mentah)
-        items = "Ini Daftar Item ✅"
-        merchant = "Ini Nama Merchant ✅"
-        total = random.randint(10, 10000)*100
-        classes = 2
-        return {"sukses": True, "ocr_mentah": ocr_mentah, "date": date, "items": items, "merchant": merchant, "total": total, "classes": classes}
+        hasil_ekstraksi = extract_receipt(ocr_mentah)
+        
+        merchant = hasil_ekstraksi["merchant"]
+        items_list = hasil_ekstraksi["items"]
+        
+        total_str = re.sub(r'[^\d]', '', hasil_ekstraksi["total"])
+        total = int(total_str) if total_str else 0
+        
+        class_id = None
+        gabungan_items = " ".join(items_list)
+        
+        if gabungan_items.strip():
+            try:
+                response = requests.post(SPACE_B_URL, json={"items_text": gabungan_items}, timeout=15)
+                
+                if response.status_code == 200:
+                    data = response.json()
+                    class_id = data.get("class_id")
+                else:
+                    print(f"Gagal mendapat respons Klasifikasi, status: {response.status_code}")
+                    
+            except requests.exceptions.RequestException as e:
+                print(f"Koneksi ke Space B gagal: {e}")
+        
+        return {
+            "sukses": True, 
+            "ocr_mentah": ocr_mentah, 
+            "date": date, 
+            "items": items_list,
+            "merchant": merchant, 
+            "total": total, 
+            "classes": class_id
+        }
+        
     except Exception as e:
-        return JSONResponse(status_code=500, content={"sukses": False, "pesan": str(e)})
+        return JSONResponse(status_code=500, content={"sukses": False, "pesan": str(e)})

packages.txt

@@ -1,2 +0,0 @@
-tesseract-ocr
-tesseract-ocr-ind

requirements.txt

@@ -1,10 +1,7 @@
 fastapi
 uvicorn
 python-multipart
-#pytesseract
-#Pillow
 opencv-python-headless
 paddlepaddle
 paddleocr==2.7.3
-numpy<2.0.0
-onnxruntime
+numpy<2.0.0