bkaan99 commited on
Commit
14d09fe
·
1 Parent(s): 20977ea
Files changed (8) hide show
  1. Dockerfile +22 -0
  2. README.md +4 -5
  3. api.py +111 -0
  4. deskew.py +55 -0
  5. detect.py +41 -0
  6. read_data.py +14 -0
  7. requirements.txt +14 -0
  8. tableocr.py +85 -0
Dockerfile ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Ana imaj olarak Python imajını kullanın
2
+ FROM python:3.9-slim
3
+
4
+ # Uygulamanın çalışacağı dizini oluşturun
5
+ WORKDIR /app
6
+
7
+ # Gerekli bağımlılıkları kopyalayın ve yükleyin
8
+ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install\
9
+ libgl1\
10
+ libgl1-mesa-glx \
11
+ libglib2.0-0 -y\
12
+ libgomp1
13
+ COPY requirements.txt .
14
+ RUN pip install python-multipart
15
+ RUN pip install --no-cache-dir -r requirements.txt
16
+
17
+ # Uygulama dosyalarını kopyalayın
18
+ COPY . .
19
+
20
+ # Docker içinde FastAPI uygulamasını çalıştırın
21
+ CMD ["uvicorn", "api:app", "--host", "0.0.0.0", "--port", "7860"]
22
+
README.md CHANGED
@@ -1,11 +1,10 @@
1
  ---
2
- title: Invoice Api
3
- emoji: 📈
4
- colorFrom: green
5
- colorTo: purple
6
  sdk: docker
7
  pinned: false
8
- license: mit
9
  ---
10
 
11
  Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
 
1
  ---
2
+ title: Invoices Api
3
+ emoji: 📚
4
+ colorFrom: pink
5
+ colorTo: red
6
  sdk: docker
7
  pinned: false
 
8
  ---
9
 
10
  Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
api.py ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from fastapi import FastAPI, HTTPException, Request
2
+ from fastapi.responses import JSONResponse
3
+ import cv2
4
+ import os
5
+ import pandas as pd
6
+ import numpy as np
7
+ from typing import List, Any
8
+ from pydantic import BaseModel
9
+ from pymongo import MongoClient
10
+ import dns.resolver
11
+ from pymongo.server_api import ServerApi
12
+ from datetime import datetime
13
+ import base64
14
+ from detect import detect_with_yolo
15
+ from tableocr import get_list
16
+ from deskew import deskew
17
+ from read_data import get_text
18
+
19
+
20
+ os.environ['KMP_DUPLICATE_LIB_OK'] = 'True'
21
+
22
+ app = FastAPI()
23
+
24
+ dns.resolver.default_resolver=dns.resolver.Resolver(configure=False)
25
+ dns.resolver.default_resolver.nameservers=['8.8.8.8']
26
+
27
+ uri = "mongodb+srv://senel1806:dEWqI0FVFMDRpiFl@cluster0.f2kg7rn.mongodb.net/?retryWrites=true&w=majority&appName=Cluster0"
28
+ client = MongoClient(uri, server_api=ServerApi('1'))
29
+ db = client['invoices']
30
+ collection = db['invoices']
31
+
32
+
33
+ class ProcessedData(BaseModel):
34
+ company_name: str
35
+ invoice_number: str
36
+ invoice_date: str
37
+ table_data: List[Any]
38
+
39
+ class ImageData(BaseModel):
40
+ image: str
41
+
42
+ class BatchProcessedData(BaseModel):
43
+ results: List[ProcessedData]
44
+
45
+ class Invoice(BaseModel):
46
+ code: str
47
+ timestamp: str
48
+ date: str
49
+ isTransferred: bool
50
+ images: List[Any]
51
+ data: List[Any]
52
+
53
+ def extractData(input_img):
54
+ result_img, table_img, company_img, number_img, date_img = detect_with_yolo(input_img)
55
+
56
+ company_name = "" # Şirket adı
57
+ if company_img is not None and company_img.size > 0:
58
+ company_name = get_text(company_img)
59
+
60
+ invoice_number = "" # Fatura numarası
61
+ if number_img is not None and number_img.size > 0:
62
+ invoice_number = get_text(number_img)
63
+
64
+ invoice_date = "" # Fatura tarihi
65
+ if date_img is not None and date_img.size > 0:
66
+ invoice_date = get_text(date_img)
67
+
68
+ table_img = deskew(table_img)
69
+ data = get_list(table_img)
70
+ max_columns = max(len(row) for row in data)
71
+ column_names = [f"Column{i+1}" for i in range(max_columns)] # Column1, Column2, ... şeklinde sütun adları oluştur
72
+ df = pd.DataFrame(data, columns=column_names)
73
+
74
+ return df, company_name, invoice_number, invoice_date
75
+
76
+ @app.post("/process-images/", response_model=BatchProcessedData)
77
+ async def process_images(image_data_list: List[ImageData]):
78
+ if not image_data_list:
79
+ raise HTTPException(status_code=400, detail="Image data list cannot be empty.")
80
+ results = []
81
+ for image_data in image_data_list:
82
+ image_bytes = base64.b64decode(image_data.image)
83
+ nparr = np.frombuffer(image_bytes, np.uint8)
84
+ img = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
85
+
86
+ df, company_name, invoice_number, invoice_date = extractData(img)
87
+
88
+ table_data = df.to_dict(orient='records')
89
+
90
+ response_data = ProcessedData(
91
+ company_name=company_name,
92
+ invoice_number=invoice_number,
93
+ invoice_date=invoice_date,
94
+ table_data=table_data
95
+ )
96
+ results.append(response_data.model_dump()) # Use model_dump instead of dict
97
+
98
+ return JSONResponse(content=results)
99
+
100
+ @app.post("/save-invoice")
101
+ async def save_invoice(invoice: Invoice, request: Request):
102
+ try:
103
+ invoice_data = await request.json()
104
+ print("Received payload:", invoice_data) # Log the payload
105
+ invoice = Invoice(**invoice_data) # Validate the payload using Pydantic
106
+ collection.insert_one(invoice.model_dump())
107
+ return {"message": "Invoice saved successfully"}
108
+ except ValidationError as e:
109
+ raise HTTPException(status_code=422, detail=e.errors())
110
+ except Exception as e:
111
+ raise HTTPException(status_code=500, detail=str(e))
deskew.py ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import cv2
2
+ import numpy as np
3
+
4
+ def getSkewAngle(cvImage) -> float:
5
+
6
+
7
+ # Prep image, copy, convert to gray scale, blur, and threshold
8
+ newImage = cvImage.copy()
9
+ gray = cv2.cvtColor(newImage, cv2.COLOR_BGR2GRAY)
10
+ blur = cv2.GaussianBlur(gray, (9, 9), 0)
11
+ thresh = cv2.threshold(blur, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]
12
+
13
+ # Apply dilate to merge text into meaningful lines/paragraphs.
14
+ # Use larger kernel on X axis to merge characters into single line, cancelling out any spaces.
15
+ # But use smaller kernel on Y axis to separate between different blocks of text
16
+ kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (10, 5))
17
+ dilate = cv2.dilate(thresh, kernel, iterations=2)
18
+
19
+ # Find all contours
20
+ contours, hierarchy = cv2.findContours(dilate, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
21
+ contours = [cnt for cnt in contours if cv2.contourArea(cnt) > 1000]
22
+ contours = sorted(contours, key = cv2.contourArea, reverse = True)
23
+ for c in contours:
24
+ rect = cv2.boundingRect(c)
25
+ x,y,w,h = rect
26
+ cv2.rectangle(newImage,(x,y),(x+w,y+h),(0,255,0),2)
27
+
28
+ # Find largest contour and surround in min area box
29
+ largestContour = contours[0]
30
+ minAreaRect = cv2.minAreaRect(largestContour)
31
+ # Determine the angle. Convert it to the value that was originally used to obtain skewed image
32
+ angle = minAreaRect[-1]
33
+ print(angle)
34
+ if angle < -45:
35
+ angle = 90 + angle
36
+
37
+ return -1.0 * angle
38
+ # Rotate the image around its center
39
+ def rotateImage(cvImage, angle: float):
40
+ newImage = cvImage.copy()
41
+ (h, w) = newImage.shape[:2]
42
+ center = (w // 2, h // 2)
43
+ M = cv2.getRotationMatrix2D(center, angle, 1.0)
44
+ newImage = cv2.warpAffine(newImage, M, (w, h), flags=cv2.INTER_CUBIC, borderMode=cv2.BORDER_REPLICATE)
45
+ return newImage
46
+
47
+ def deskew(cvImage):
48
+ angle = getSkewAngle(cvImage)
49
+ check_angle = angle * -1.0
50
+ if check_angle > -60 and check_angle < 60:
51
+ rotated_image = rotateImage(cvImage, -1.0 * angle)
52
+ else:
53
+ rotated_image = cvImage
54
+ return rotated_image
55
+
detect.py ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from ultralytics import YOLO
2
+ import cv2
3
+
4
+ model = YOLO('last (1).pt')
5
+
6
+ def detect_with_yolo(img):
7
+ color_palette = [(0, 255, 0), (0, 0, 255), (255, 0, 0), (0, 165, 255), (128, 0, 128), (255, 255, 0), (255, 128, 0)]
8
+
9
+ results = model.predict(img, stream=True)
10
+
11
+ detected_img = img.copy() # Copy of the original image
12
+ table_roi = None
13
+ company_roi = None
14
+ number_roi = None
15
+ date_roi = None
16
+
17
+ for result in results:
18
+ boxes = result.boxes.cpu().numpy() # Get boxes on CPU in numpy format
19
+ for box in boxes: # Iterate over boxes
20
+ r = box.xyxy[0].astype(int) # Get corner points as int
21
+ class_id = int(box.cls[0]) # Get class ID
22
+ class_name = model.names[class_id] # Get class name using the class ID
23
+ print(f"Class: {class_name}, Box: {r}") # Print class name and box coordinates
24
+ # Draw boxes on the image
25
+ color = color_palette[class_id]
26
+ cv2.rectangle(detected_img, (r[0], r[1]), (r[2], r[3]), color, 2)
27
+
28
+ # Check if the detected object is a table
29
+ if class_name == 'table':
30
+ table_roi = img[r[1]:r[3], r[0]:r[2]] # Extract ROI for the table
31
+ if class_name == 'company':
32
+ company_roi = img[r[1]:r[3], r[0]:r[2]] # Extract ROI for the table
33
+ if class_name == 'number':
34
+ number_roi = img[r[1]:r[3], r[0]:r[2]] # Extract ROI for the table
35
+ if class_name == 'date':
36
+ date_roi = img[r[1]:r[3], r[0]:r[2]] # Extract ROI for the table
37
+ print("TABLE ROI ",table_roi)
38
+ print("COMPANY ROI ",company_roi)
39
+ print("NUMBER ROI ",number_roi)
40
+ print("DATE ROI ",date_roi)
41
+ return detected_img, table_roi, company_roi, number_roi, date_roi
read_data.py ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from paddleocr import PaddleOCR
2
+ import os
3
+
4
+ current_directory = os.getcwd()
5
+
6
+ model = PaddleOCR(use_angle_cls=True, lang='en', use_gpu=True, det_model_dir=f"{current_directory}/.paddleocr/whl/det/en/en_PP-OCRv3_det_infer", rec_model_dir=f"{current_directory}/.paddleocr/whl/rec/en/en_PP-OCRv4_rec_infer", cls_model_dir=f"{current_directory}/.paddleocr/whl/cls/ch_ppocr_mobile_v2.0_cls_infer")
7
+
8
+ def get_text(img) -> str:
9
+ result = model.ocr(img, cls=False)
10
+ # OCR sonuçlarını ayıkla
11
+ for iResult in result:
12
+ for res in iResult:
13
+ line_text = res[1][0]
14
+ return line_text
requirements.txt ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ fastapi
2
+ gunicorn
3
+ shutil-extra
4
+ PyMuPDF
5
+ opencv-python==4.6.0.66
6
+ pillow==10.2.0
7
+ uvicorn
8
+ paddleocr==2.7.0
9
+ paddlepaddle
10
+ python-dotenv
11
+ ultralytics
12
+ dill
13
+ scikit-learn
14
+ pymongo==4.8.0
tableocr.py ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from paddleocr import PaddleOCR
2
+ import cv2
3
+ import pandas as pd
4
+ import numpy as np
5
+ from sklearn.cluster import DBSCAN
6
+ import os
7
+
8
+ current_directory = os.getcwd()
9
+ # OCR modelini başlat
10
+ model = PaddleOCR(use_angle_cls=True, lang='en', use_gpu=True, det_model_dir=f"{current_directory}/.paddleocr/whl/det/en/en_PP-OCRv3_det_infer", rec_model_dir=f"{current_directory}/.paddleocr/whl/rec/en/en_PP-OCRv4_rec_infer", cls_model_dir=f"{current_directory}/.paddleocr/whl/cls/ch_ppocr_mobile_v2.0_cls_infer")
11
+
12
+ # Resmi oku ve RGB formatına dönüştür
13
+ # image_path = "test_images/3.jpg"
14
+ # image = cv2.imread(image_path)
15
+ # image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
16
+
17
+ def get_list(img) -> list:
18
+ result = model.ocr(img, cls=False)
19
+ sorted_text = ocr_result_to_dataframe(result)
20
+ df = pd.DataFrame(sorted_text)
21
+ return sorted_text
22
+
23
+ def ocr_result_to_dataframe(result):
24
+ all_texts = []
25
+ all_coords = []
26
+
27
+ # OCR sonuçlarını ayıkla
28
+ for iResult in result:
29
+ for res in iResult:
30
+ line_text = res[1][0]
31
+ box_coordinates = res[0]
32
+ all_texts.append(line_text)
33
+ all_coords.append(box_coordinates)
34
+
35
+ # Her bir kutunun merkez y koordinatını hesapla
36
+ centers_y = [np.mean([coord[1] for coord in box]) for box in all_coords]
37
+ heights = [max(coord[1] for coord in box) - min(coord[1] for coord in box) for box in all_coords]
38
+
39
+ # Ortalama yüksekliği hesapla ve eşik değeri olarak kullan
40
+ avg_height = np.mean(heights)
41
+ y_threshold = avg_height * 0.5 # Y ekseni farkı için eşik
42
+
43
+ # DBSCAN kullanarak satırları gruplandır
44
+ dbscan = DBSCAN(eps=y_threshold, min_samples=1)
45
+ labels = dbscan.fit_predict(np.array(centers_y).reshape(-1, 1))
46
+
47
+ n_clusters = len(set(labels)) - (1 if -1 in labels else 0)
48
+ grouped_texts = [[] for _ in range(n_clusters)]
49
+ grouped_coords = [[] for _ in range(n_clusters)]
50
+ for text, coords, label in zip(all_texts, all_coords, labels):
51
+ if label != -1:
52
+ grouped_texts[label].append(text)
53
+ grouped_coords[label].append(coords)
54
+
55
+ # Grupları y koordinatlarına göre sırala
56
+ grouped_texts_coords = sorted(zip(grouped_texts, grouped_coords), key=lambda x: np.mean([np.mean([coord[1] for coord in box]) for box in x[1]]))
57
+
58
+ sorted_texts = []
59
+ for group_texts, group_coords in grouped_texts_coords:
60
+ # Her grubu x koordinatına göre sırala (soldan sağa)
61
+ sorted_group = sorted(zip(group_texts, group_coords), key=lambda x: x[1][0][0])
62
+ if len(sorted_group) > 3:
63
+ first_text, first_coord = sorted_group[0] # Listenin ilk elemanını al
64
+ if any(char.isalpha() for char in first_text): # Harf kontrolü
65
+ # Harf içeren bölümü ayır
66
+ split_text = first_text.split()
67
+ index = len(split_text[0])
68
+ if index < len(first_text): # İlk kelime sonrasında bir şey varsa
69
+ remaining_text = first_text[index:]
70
+ sorted_group[0] = (split_text[0], first_coord) # İlk elemanı güncelle
71
+ # Yeni elemanı oluşturarak listeye ekle
72
+ sorted_group.insert(1, (remaining_text, first_coord))
73
+ sorted_texts.append([text for text, _ in sorted_group])
74
+
75
+ return sorted_texts
76
+
77
+ # # OCR sonuçlarını satırlara dönüştür
78
+ # grouped_text = ocr_result_to_dataframe(result)
79
+
80
+ # DataFrame'e dönüştür ve Excel dosyasına yaz
81
+ # df = pd.DataFrame(grouped_text)
82
+ # df.to_excel('ocr_results.xlsx', index=False, header=False)
83
+
84
+
85
+ # get_list(image)