from fastapi import FastAPI, File, UploadFile from fastapi.responses import JSONResponse from pdf2image import convert_from_bytes import pytesseract import cv2 import numpy as np import re import os from tempfile import TemporaryDirectory app = FastAPI() ownership_labels = { "নিজস্ব জমি": "own", "লিজ নেওয়া জমি": "lease", "যৌথ মালিকানাধীন জমি": "joint" } section_headers = { "own": "কৃষকের নিজস্ব জমির জন্য প্রযোজ্য অংশ", "lease": "কৃষকের লিজ নেওয়া জমির জন্য প্রযোজ্য অংশ", "joint": "যৌথ মালিকানাধীন জমির জন্য প্রযোজ্য অংশ" } name_labels = { "own": ["কৃষকের নাম"], "lease": ["জমির মালিকের নাম", "চাষাবাদকারী কৃষকের নাম"], "joint": ["পরিবারের সদস্যের নাম"] } signature_labels = { "own": ["কৃষকের স্বাক্ষর"], "lease": ["জমির মালিকের স্বাক্ষর", "চাষাবাদকারী কৃষকের স্বাক্ষর"], "joint": ["পরিবারের সদস্যের স্বাক্ষর"] } def detect_ownership_from_text(text): line = next((l for l in text.splitlines() if "জমি মালিকানার ধরণ" in l), "") if not line: return None line = line.replace(":", "").replace("_", " ").replace("চর", "").strip() if "নিজ" in line and "লিজ" not in line and "যৌথ" not in line: return "own" elif "লিজ" in line: return "lease" elif "যৌথ" in line: return "joint" return None def find_section(image, header_text): data = pytesseract.image_to_data(image, lang='ben', output_type=pytesseract.Output.DICT) for i, word in enumerate(data['text']): if header_text in word: x, y, w, h = data['left'][i], data['top'][i], data['width'][i], data['height'][i] return image[y-50:y+600, x-50:x+600] return None def extract_name(segment_image, label_text): text = pytesseract.image_to_string(segment_image, lang='ben') match = re.search(fr"{label_text}[:\-]?\s*(\S+)", text) return match.group(1) if match else "Not found" def extract_signature(segment_image, label, output_dir): gray = cv2.cvtColor(segment_image, cv2.COLOR_BGR2GRAY) _, thresh = cv2.threshold(gray, 150, 255, cv2.THRESH_BINARY_INV) contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) count = 0 paths = [] for cnt in contours: x, y, w, h = cv2.boundingRect(cnt) if w > 100 and h < 100: sig = segment_image[y:y+h, x:x+w] path = os.path.join(output_dir, f"{label}_signature_{count}.jpg") cv2.imwrite(path, sig) paths.append(path) count += 1 return paths @app.post("/extract") async def extract_fields(file: UploadFile = File(...)): with TemporaryDirectory() as tmpdir: pdf_bytes = await file.read() images = convert_from_bytes(pdf_bytes, dpi=300) result = { "ownership_types": [], "names": {}, "signatures": {} } for img in images: img_cv = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR) text = pytesseract.image_to_string(img_cv, lang='ben+eng') ownership = detect_ownership_from_text(text) if ownership and ownership not in result["ownership_types"]: result["ownership_types"].append(ownership) section = find_section(img_cv, section_headers.get(ownership, "")) if ownership else None target = section if section is not None else img_cv for label in name_labels.get(ownership, []): if label not in result["names"]: name = extract_name(target, label) result["names"][label] = name for label in signature_labels.get(ownership, []): sig_paths = extract_signature(target, label, tmpdir) if label in result["signatures"]: result["signatures"][label].extend(sig_paths) else: result["signatures"][label] = sig_paths return result # from fastapi import FastAPI, File, UploadFile # from fastapi.responses import JSONResponse # from pdf2image import convert_from_bytes # import pytesseract # import cv2 # import numpy as np # import re # import os # from tempfile import TemporaryDirectory # app = FastAPI() # ownership_labels = { # "নিজস্ব জমি": "own", # "লিজ নেওয়া জমি": "lease", # "যৌথ মালিকানাধীন জমি": "joint" # } # section_headers = { # "own": "কৃষকের নিজস্ব জমির জন্য প্রযোজ্য অংশ", # "lease": "কৃষকের লিজ নেওয়া জমির জন্য প্রযোজ্য অংশ", # "joint": "যৌথ মালিকানাধীন জমির জন্য প্রযোজ্য অংশ" # } # name_labels = { # "own": ["কৃষকের নাম"], # "lease": ["জমির মালিকের নাম", "চাষাবাদকারী কৃষকের নাম"], # "joint": ["পরিবারের সদস্যের নাম"] # } # signature_labels = { # "own": ["কৃষকের স্বাক্ষর"], # "lease": ["জমির মালিকের স্বাক্ষর", "চাষাবাদকারী কৃষকের স্বাক্ষর"], # "joint": ["পরিবারের সদস্যের স্বাক্ষর"] # } # def detect_ownership_from_text(text): # line = next((l for l in text.splitlines() if "জমি মালিকানার ধরণ" in l), "") # if not line: # return None # line = line.replace(":", "").replace("_", " ").replace("চর", "").strip() # if "নিজ" in line and "লিজ" not in line and "যৌথ" not in line: # return "own" # elif "লিজ" in line: # return "lease" # elif "যৌথ" in line: # return "joint" # return None # def find_section(image, header_text): # data = pytesseract.image_to_data(image, lang='ben', output_type=pytesseract.Output.DICT) # for i, word in enumerate(data['text']): # if header_text in word: # x, y, w, h = data['left'][i], data['top'][i], data['width'][i], data['height'][i] # return image[y-50:y+600, x-50:x+600] # return None # def extract_name(segment_image, label_text): # text = pytesseract.image_to_string(segment_image, lang='ben') # match = re.search(fr"{label_text}[:\-]?\s*(\S+)", text) # return match.group(1) if match else "Not found" # def extract_signature(segment_image, label, output_dir): # gray = cv2.cvtColor(segment_image, cv2.COLOR_BGR2GRAY) # _, thresh = cv2.threshold(gray, 150, 255, cv2.THRESH_BINARY_INV) # contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # count = 0 # paths = [] # for cnt in contours: # x, y, w, h = cv2.boundingRect(cnt) # if w > 100 and h < 100: # sig = segment_image[y:y+h, x:x+w] # path = os.path.join(output_dir, f"{label}_signature_{count}.jpg") # cv2.imwrite(path, sig) # paths.append(path) # count += 1 # return paths # @app.post("/extract") # async def extract_fields(file: UploadFile = File(...)): # with TemporaryDirectory() as tmpdir: # pdf_bytes = await file.read() # images = convert_from_bytes(pdf_bytes, dpi=300) # result = {"ownership_type": None, "names": {}, "signatures": {}} # for img in images: # img_cv = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR) # text = pytesseract.image_to_string(img_cv, lang='ben+eng') # ownership = detect_ownership_from_text(text) # result["ownership_type"] = ownership # section = find_section(img_cv, section_headers.get(ownership, "")) if ownership else None # target = section if section is not None else img_cv # for label in name_labels.get(ownership, []): # name = extract_name(target, label) # result["names"][label] = name # for label in signature_labels.get(ownership, []): # sig_paths = extract_signature(target, label, tmpdir) # result["signatures"][label] = sig_paths # break # process only first matching page # return result # def find_ownership_segment(image): # data = pytesseract.image_to_data(image, lang='ben', output_type=pytesseract.Output.DICT) # for i, word in enumerate(data['text']): # if "জমি মালিকানার ধরণ:" in word: # x, y, w, h = data['left'][i], data['top'][i], data['width'][i], data['height'][i] # return image[y-50:y+300, x-50:x+600] # return None # def detect_ownership_from_segment(segment): # gray = cv2.cvtColor(segment, cv2.COLOR_BGR2GRAY) # _, thresh = cv2.threshold(gray, 180, 255, cv2.THRESH_BINARY_INV) # contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # for cnt in contours: # x, y, w, h = cv2.boundingRect(cnt) # if 10 < w < 50 and 10 < h < 50: # roi = segment[y:y+h, x+50:x+300] # text = pytesseract.image_to_string(roi, lang='ben') # for label in ownership_labels: # if label in text: # return ownership_labels[label] # return None # @app.post("/extract") # async def extract_fields(file: UploadFile = File(...)): # with TemporaryDirectory() as tmpdir: # pdf_bytes = await file.read() # images = convert_from_bytes(pdf_bytes, dpi=300) # result = {"ownership_type": None, "names": {}, "signatures": {}} # found = False # for img in images: # img_cv = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR) # segment = find_ownership_segment(img_cv) # if segment is not None: # ownership = detect_ownership_from_segment(segment) # if ownership: # section = find_section(img_cv, section_headers[ownership]) # if section is not None: # result["ownership_type"] = ownership # for label in name_labels[ownership]: # name = extract_name(section, label) # result["names"][label] = name # for label in signature_labels[ownership]: # sig_paths = extract_signature(section, label, tmpdir) # result["signatures"][label] = sig_paths # found = True # break # if not found: # return JSONResponse(content={"error": "Ownership type or section not found"}, status_code=400) # return result ################################################################# # from fastapi import FastAPI, File, UploadFile # from fastapi.responses import JSONResponse # from pdf2image import convert_from_bytes # import pytesseract # import cv2 # import numpy as np # import re # import os # from tempfile import TemporaryDirectory # app = FastAPI() # ownership_labels = { # "নিজস্ব জমি": "own", # "লিজ নেওয়া জমি": "lease", # "যৌথ মালিকানাধীন জমি": "joint" # } # section_headers = { # "own": "কৃষকের নিজস্ব জমির জন্য প্রযোজ্য অংশ", # "lease": "কৃষকের লিজ নেওয়া জমির জন্য প্রযোজ্য অংশ", # "joint": "যৌথ মালিকানাধীন জমির জন্য প্রযোজ্য অংশ" # } # name_labels = { # "own": ["কৃষকের নাম"], # "lease": ["জমির মালিকের নাম", "চাষাবাদকারী কৃষকের নাম"], # "joint": ["পরিবারের সদস্যের নাম"] # } # signature_labels = { # "own": ["কৃষকের স্বাক্ষর"], # "lease": ["জমির মালিকের স্বাক্ষর", "চাষাবাদকারী কৃষকের স্বাক্ষর"], # "joint": ["পরিবারের সদস্যের স্বাক্ষর"] # } # def detect_ownership(image): # gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) # _, thresh = cv2.threshold(gray, 180, 255, cv2.THRESH_BINARY_INV) # contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # for cnt in contours: # x, y, w, h = cv2.boundingRect(cnt) # if 10 < w < 50 and 10 < h < 50: # roi = image[y:y+h, x+50:x+300] # text = pytesseract.image_to_string(roi, lang='ben') # for label in ownership_labels: # if label in text: # return ownership_labels[label] # return None # def find_section(image, header_text): # data = pytesseract.image_to_data(image, lang='ben', output_type=pytesseract.Output.DICT) # for i, word in enumerate(data['text']): # if header_text in word: # x, y, w, h = data['left'][i], data['top'][i], data['width'][i], data['height'][i] # return image[y-50:y+600, x-50:x+600] # return None # def extract_name(segment_image, label_text): # text = pytesseract.image_to_string(segment_image, lang='ben') # match = re.search(fr"{label_text}[:\-]?\s*(\S+)", text) # return match.group(1) if match else "Not found" # def extract_signature(segment_image, label, output_dir): # gray = cv2.cvtColor(segment_image, cv2.COLOR_BGR2GRAY) # _, thresh = cv2.threshold(gray, 150, 255, cv2.THRESH_BINARY_INV) # contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # count = 0 # paths = [] # for cnt in contours: # x, y, w, h = cv2.boundingRect(cnt) # if w > 100 and h < 100: # sig = segment_image[y:y+h, x:x+w] # path = os.path.join(output_dir, f"{label}_signature_{count}.jpg") # cv2.imwrite(path, sig) # paths.append(path) # count += 1 # return paths # @app.post("/extract") # async def extract_fields(file: UploadFile = File(...)): # with TemporaryDirectory() as tmpdir: # pdf_bytes = await file.read() # images = convert_from_bytes(pdf_bytes, dpi=300) # result = {"ownership_type": None, "names": {}, "signatures": {}} # found = False # for img in images: # img_cv = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR) # ownership = detect_ownership(img_cv) # if ownership: # section = find_section(img_cv, section_headers[ownership]) # if section is not None: # result["ownership_type"] = ownership # for label in name_labels[ownership]: # name = extract_name(section, label) # result["names"][label] = name # for label in signature_labels[ownership]: # sig_paths = extract_signature(section, label, tmpdir) # result["signatures"][label] = sig_paths # found = True # break # if not found: # return JSONResponse(content={"error": "Ownership type or section not found"}, status_code=400) # return result