Mohanrajdeena/Intelligent-document-processing
0
1import gradio as gr2import spaces3import cv24import numpy as np5import easyocr6import torch7import re8import json9import time10import tempfile11 12from pathlib import Path13from datetime import datetime14 15 16# ============================================================17# CONFIGURATION18# ============================================================19 20SUPPORTED_TYPES = [21 "Invoice",22 "Resume",23 "ID Card"24]25 26 27SKILL_DICTIONARY = [28 "Python",29 "SQL",30 "MySQL",31 "PostgreSQL",32 "Power BI",33 "Excel",34 "Power Query",35 "Tableau",36 "Machine Learning",37 "Deep Learning",38 "Pandas",39 "NumPy",40 "Matplotlib",41 "Seaborn",42 "Streamlit",43 "TensorFlow",44 "PyTorch",45 "Scikit-learn",46 "Java",47 "C++",48 "C",49 "AWS",50 "Azure",51 "Git",52 "GitHub",53 "SAP",54 "Oracle"55]56 57 58# ============================================================59# IMAGE UTILITIES60# ============================================================61 62def read_image(image_path):63 64 if not image_path:65 raise ValueError(66 "Please upload a document image."67 )68 69 image = cv2.imread(70 str(image_path)71 )72 73 if image is None:74 raise ValueError(75 "Unable to read the uploaded image."76 )77 78 return image79 80 81def resize_image(82 image,83 max_width=160084):85 86 height, width = image.shape[:2]87 88 if width <= max_width:89 return image90 91 scale = max_width / width92 93 new_height = int(94 height * scale95 )96 97 return cv2.resize(98 image,99 (100 max_width,101 new_height102 ),103 interpolation=cv2.INTER_AREA104 )105 106 107# ============================================================108# DOCUMENT PREPROCESSING109# ============================================================110 111def preprocess_document(image):112 113 image = resize_image(114 image,115 max_width=1600116 )117 118 gray = cv2.cvtColor(119 image,120 cv2.COLOR_BGR2GRAY121 )122 123 # CLAHE enhancement124 clahe = cv2.createCLAHE(125 clipLimit=2.0,126 tileGridSize=(8, 8)127 )128 129 enhanced = clahe.apply(130 gray131 )132 133 # Gaussian denoising134 denoised = cv2.GaussianBlur(135 enhanced,136 (3, 3),137 0138 )139 140 # Otsu thresholding141 thresholded = cv2.threshold(142 denoised,143 0,144 255,145 cv2.THRESH_BINARY +146 cv2.THRESH_OTSU147 )[1]148 149 return thresholded150 151 152# ============================================================153# OCR154# ============================================================155 156def create_ocr_reader():157 158 # ZeroGPU provides CUDA while the159 # @spaces.GPU function is executing.160 161 gpu_available = torch.cuda.is_available()162 163 print(164 f"Initializing EasyOCR | GPU: {gpu_available}"165 )166 167 reader = easyocr.Reader(168 ["en"],169 gpu=gpu_available,170 verbose=False171 )172 173 return reader174 175 176def perform_ocr(177 image,178 reader179):180 181 results = reader.readtext(182 image,183 detail=1,184 paragraph=False185 )186 187 text_lines = []188 confidence_scores = []189 190 for result in results:191 192 if len(result) >= 3:193 194 text_lines.append(195 str(result[1])196 )197 198 confidence_scores.append(199 float(result[2])200 )201 202 full_text = "\n".join(203 text_lines204 )205 206 average_confidence = (207 sum(confidence_scores)208 /209 len(confidence_scores)210 if confidence_scores211 else 0.0212 )213 214 return {215 "text": full_text,216 "confidence": average_confidence,217 "results": results218 }219 220 221# ============================================================222# TEXT CLEANING223# ============================================================224 225def clean_text(text):226 227 text = re.sub(228 r"[ \t]+",229 " ",230 text231 )232 233 text = re.sub(234 r"\n{2,}",235 "\n",236 text237 )238 239 return text.strip()240 241 242# ============================================================243# GENERIC HELPERS244# ============================================================245 246def unique(values):247 248 return list(249 dict.fromkeys(values)250 )251 252 253def extract_emails(text):254 255 pattern = (256 r"\b[A-Za-z0-9._%+-]+"257 r"@[A-Za-z0-9.-]+\."258 r"[A-Za-z]{2,}\b"259 )260 261 return unique(262 re.findall(263 pattern,264 text265 )266 )267 268 269def extract_phone_numbers(text):270 271 pattern = (272 r"(?<!\d)"273 r"(?:\+91[\s-]?)?"274 r"[6-9]\d{9}"275 r"(?!\d)"276 )277 278 return unique(279 re.findall(280 pattern,281 text282 )283 )284 285 286def extract_dates(text):287 288 pattern = (289 r"\b(?:"290 r"\d{1,2}[/-]\d{1,2}[/-]\d{2,4}"291 r"|"292 r"\d{1,2}[/-]"293 r"[A-Za-z]{3,9}[/-]?"294 r"\d{2,4}"295 r"|"296 r"[A-Za-z]{3,9}\s+"297 r"\d{1,2},?\s+\d{2,4}"298 r")\b"299 )300 301 return unique(302 re.findall(303 pattern,304 text,305 flags=re.IGNORECASE306 )307 )308 309 310def extract_amounts(text):311 312 pattern = (313 r"(?:₹|Rs\.?|INR|\$|€|£)\s*"314 r"\d+(?:,\d{3})*(?:\.\d{1,2})?"315 )316 317 return unique(318 re.findall(319 pattern,320 text,321 flags=re.IGNORECASE322 )323 )324 325 326# ============================================================327# INVOICE EXTRACTION328# ============================================================329 330def extract_invoice_number(text):331 332 patterns = [333 334 (335 r"(?:invoice\s*"336 r"(?:no|number|#)"337 r"\s*[:\-]?\s*)"338 r"([A-Z0-9\/\-_]+)"339 ),340 341 (342 r"(?:inv\s*"343 r"(?:no|number|#)"344 r"\s*[:\-]?\s*)"345 r"([A-Z0-9\/\-_]+)"346 )347 ]348 349 for pattern in patterns:350 351 match = re.search(352 pattern,353 text,354 flags=re.IGNORECASE355 )356 357 if match:358 359 return (360 match.group(1)361 .strip()362 )363 364 return None365 366 367def extract_gstin(text):368 369 pattern = (370 r"\b"371 r"\d{2}"372 r"[A-Z]{5}"373 r"\d{4}"374 r"[A-Z]"375 r"\d"376 r"[A-Z0-9]"377 r"\b"378 )379 380 return unique(381 re.findall(382 pattern,383 text.upper()384 )385 )386 387 388def extract_invoice_entities(text):389 390 return {391 392 "Invoice Number":393 extract_invoice_number(text),394 395 "Invoice Date":396 extract_dates(text),397 398 "GSTIN":399 extract_gstin(text),400 401 "Amounts":402 extract_amounts(text),403 404 "Email":405 extract_emails(text),406 407 "Phone Number":408 extract_phone_numbers(text)409 }410 411 412# ============================================================413# RESUME EXTRACTION414# ============================================================415 416def extract_name_from_resume(text):417 418 lines = [419 420 line.strip()421 422 for line in text.split("\n")423 424 if line.strip()425 ]426 427 if lines:428 429 candidate = lines[0]430 431 if (432 len(candidate.split()) <= 5433 and len(candidate) <= 60434 ):435 436 return candidate437 438 return None439 440 441def extract_skills(text):442 443 text_lower = text.lower()444 445 return [446 447 skill448 449 for skill in SKILL_DICTIONARY450 451 if skill.lower()452 in text_lower453 ]454 455 456def extract_resume_entities(text):457 458 return {459 460 "Name":461 extract_name_from_resume(text),462 463 "Email":464 extract_emails(text),465 466 "Phone Number":467 extract_phone_numbers(text),468 469 "Skills":470 extract_skills(text),471 472 "Dates":473 extract_dates(text)474 }475 476 477# ============================================================478# ID CARD EXTRACTION479# ============================================================480 481def extract_aadhaar_numbers(text):482 483 pattern = (484 r"\b\d{4}\s\d{4}\s\d{4}\b"485 )486 487 return unique(488 re.findall(489 pattern,490 text491 )492 )493 494 495def extract_id_card_entities(text):496 497 return {498 499 "Name":500 extract_name_from_resume(text),501 502 "ID Number":503 extract_aadhaar_numbers(text),504 505 "Date":506 extract_dates(text),507 508 "Phone Number":509 extract_phone_numbers(text)510 }511 512 513# ============================================================514# DOCUMENT ROUTER515# ============================================================516 517def extract_entities(518 text,519 document_type520):521 522 if document_type == "Invoice":523 524 return extract_invoice_entities(525 text526 )527 528 if document_type == "Resume":529 530 return extract_resume_entities(531 text532 )533 534 if document_type == "ID Card":535 536 return extract_id_card_entities(537 text538 )539 540 return {}541 542 543# ============================================================544# OUTPUT HELPERS545# ============================================================546 547def count_detected_entities(548 entities549):550 551 count = 0552 553 for value in entities.values():554 555 if value is None:556 continue557 558 if isinstance(559 value,560 list561 ):562 563 count += len(value)564 565 elif str(value).strip():566 567 count += 1568 569 return count570 571 572def format_entities(573 entities574):575 576 formatted = {}577 578 for key, value in entities.items():579 580 if value is None:581 582 formatted[key] = (583 "Not detected"584 )585 586 elif isinstance(587 value,588 list589 ):590 591 formatted[key] = (592 593 ", ".join(594 str(x)595 for x in value596 )597 598 if value599 600 else "Not detected"601 )602 603 else:604 605 formatted[key] = (606 str(value)607 )608 609 return formatted610 611 612def create_json_file(data):613 614 temp = tempfile.NamedTemporaryFile(615 mode="w",616 suffix=".json",617 prefix="document_results_",618 delete=False,619 encoding="utf-8"620 )621 622 json.dump(623 data,624 temp,625 indent=4,626 ensure_ascii=False627 )628 629 temp.close()630 631 return temp.name632 633 634# ============================================================635# MAIN PROCESSING FUNCTION636# ============================================================637 638@spaces.GPU(duration=120)639def process_document(640 document_type,641 image_path642):643 644 start_time = time.perf_counter()645 646 if not image_path:647 648 raise gr.Error(649 "Please upload an invoice, "650 "resume, or ID card image."651 )652 653 if document_type not in SUPPORTED_TYPES:654 655 raise gr.Error(656 "Please select a valid "657 "document type."658 )659 660 try:661 662 # ----------------------------------------------------663 # STEP 1: LOAD IMAGE664 # ----------------------------------------------------665 666 original_bgr = read_image(667 image_path668 )669 670 original_rgb = cv2.cvtColor(671 original_bgr,672 cv2.COLOR_BGR2RGB673 )674 675 # ----------------------------------------------------676 # STEP 2: PREPROCESS677 # ----------------------------------------------------678 679 processed_image = (680 preprocess_document(681 original_bgr682 )683 )684 685 # ----------------------------------------------------686 # STEP 3: LOAD EASYOCR687 # ----------------------------------------------------688 689 # IMPORTANT:690 # Reader is created inside the GPU-decorated691 # function so ZeroGPU can provide CUDA.692 693 reader = create_ocr_reader()694 695 # ----------------------------------------------------696 # STEP 4: OCR697 # ----------------------------------------------------698 699 ocr_output = perform_ocr(700 processed_image,701 reader702 )703 704 raw_text = (705 ocr_output["text"]706 )707 708 confidence = float(709 ocr_output["confidence"]710 )711 712 # ----------------------------------------------------713 # STEP 5: CLEAN TEXT714 # ----------------------------------------------------715 716 cleaned_text = clean_text(717 raw_text718 )719 720 # ----------------------------------------------------721 # STEP 6: ENTITY EXTRACTION722 # ----------------------------------------------------723 724 entities = extract_entities(725 cleaned_text,726 document_type727 )728 729 formatted_entities = (730 format_entities(731 entities732 )733 )734 735 detected_fields = (736 count_detected_entities(737 entities738 )739 )740 741 # ----------------------------------------------------742 # STEP 7: PROCESSING TIME743 # ----------------------------------------------------744 745 processing_time = (746 time.perf_counter()747 - start_time748 )749 750 # ----------------------------------------------------751 # STEP 8: FINAL JSON752 # ----------------------------------------------------753 754 final_output = {755 756 "document_type":757 document_type,758 759 "file_name":760 Path(761 image_path762 ).name,763 764 "processing_timestamp":765 datetime.now().isoformat(),766 767 "ocr_confidence":768 round(769 confidence,770 4771 ),772 773 "ocr_confidence_percent":774 round(775 confidence * 100,776 2777 ),778 779 "processing_time_seconds":780 round(781 processing_time,782 3783 ),784 785 "detected_fields":786 detected_fields,787 788 "entities":789 formatted_entities,790 791 "ocr_text":792 raw_text,793 794 "cleaned_text":795 cleaned_text796 }797 798 # ----------------------------------------------------799 # STEP 9: CREATE DOWNLOADABLE JSON800 # ----------------------------------------------------801 802 json_file = create_json_file(803 final_output804 )805 806 # ----------------------------------------------------807 # STEP 10: SUMMARY808 # ----------------------------------------------------809 810 confidence_text = (811 f"{confidence * 100:.1f}%"812 )813 814 summary = f"""815### ✅ Processing Complete816 817| Metric | Result |818|---|---|819| 📄 Document Type | **{document_type}** |820| 🎯 OCR Confidence | **{confidence_text}** |821| 🔎 Fields Detected | **{detected_fields}** |822| 📝 OCR Characters | **{len(raw_text)}** |823| ⚡ Processing Time | **{processing_time:.2f} seconds** |824 825**Pipeline:** 826Preprocessing → EasyOCR → Text Cleaning → Entity Extraction827"""828 829 return (830 831 original_rgb,832 833 processed_image,834 835 formatted_entities,836 837 raw_text,838 839 cleaned_text,840 841 summary,842 843 json.dumps(844 final_output,845 indent=4,846 ensure_ascii=False847 ),848 849 json_file850 )851 852 except gr.Error:853 854 raise855 856 except Exception as error:857 858 print(859 "Processing error:",860 repr(error)861 )862 863 raise gr.Error(864 f"Processing failed: {error}"865 )866 867 868# ============================================================869# PROFESSIONAL UI CSS870# ============================================================871 872CSS = """873 874/* ================================875 PAGE876================================ */877 878body {879 880 background:881 linear-gradient(882 135deg,883 #f8faff 0%,884 #f3f0ff 50%,885 #eef7ff 100%886 ) !important;887}888 889 890/* ================================891 MAIN CONTAINER892================================ */893 894.gradio-container {895 896 max-width:897 1450px !important;898 899 margin:900 auto !important;901 902 padding:903 20px !important;904}905 906 907/* ================================908 HERO909================================ */910 911.hero {912 913 background:914 linear-gradient(915 135deg,916 #2563eb,917 #7c3aed918 );919 920 color:921 white;922 923 border-radius:924 20px;925 926 padding:927 30px 35px;928 929 margin-bottom:930 20px;931 932 box-shadow:933 0 12px 35px934 rgba(935 37,936 99,937 235,938 0.20939 );940}941 942.hero h1 {943 944 color:945 white !important;946 947 margin:948 0 !important;949 950 font-size:951 2.3rem !important;952 953 font-weight:954 800 !important;955}956 957.hero p {958 959 color:960 rgba(961 255,962 255,963 255,964 0.95965 ) !important;966 967 margin-top:968 10px !important;969 970 font-size:971 1.05rem !important;972}973 974 975/* ================================976 SECTION HEADINGS977================================ */978 979.section-title {980 981 font-size:982 1.2rem;983 984 font-weight:985 800;986 987 color:988 #1e293b !important;989}990 991 992/* ================================993 CARDS994================================ */995 996.card {997 998 background:999 white;1000 1001 border:1002 1px solid #e5e7eb;1003 1004 border-radius:1005 16px;1006 1007 padding:1008 18px;1009 1010 box-shadow:1011 0 6px 20px1012 rgba(1013 15,1014 23,1015 42,1016 0.061017 );1018}1019 1020 1021/* ================================1022 LABELS1023================================ */1024 1025label {1026 1027 color:1028 #1e293b !important;1029 1030 font-weight:1031 700 !important;1032}1033 1034 1035/* ================================1036 INPUTS1037================================ */1038 1039textarea,1040input {1041 1042 color:1043 #111827 !important;1044 1045 background:1046 #ffffff !important;1047}1048 1049 1050/* ================================1051 BUTTON1052================================ */1053 1054button.primary {1055 1056 background:1057 linear-gradient(1058 135deg,1059 #2563eb,1060 #7c3aed1061 ) !important;1062 1063 color:1064 white !important;1065 1066 border:1067 none !important;1068 1069 font-weight:1070 800 !important;1071 1072 border-radius:1073 10px !important;1074}1075 1076 1077/* ================================1078 OUTPUT1079================================ */1080 1081.output-markdown {1082 1083 color:1084 #1e293b !important;1085}1086 1087 1088/* ================================1089 FOOTER1090================================ */1091 1092footer {1093 1094 display:1095 none !important;1096}1097 1098"""1099 1100 1101# ============================================================1102# GRADIO APPLICATION1103# ============================================================1104 1105with gr.Blocks(1106 title="Intelligent Document Processing"1107) as demo:1108 1109 # --------------------------------------------------------1110 # HERO1111 # --------------------------------------------------------1112 1113 gr.HTML(1114 """1115 <div class="hero">1116 1117 <h1>1118 📄 Intelligent Document Processing1119 </h1>1120 1121 <p>1122 AI-powered OCR, text cleaning and1123 structured information extraction1124 for invoices, resumes and ID cards.1125 </p>1126 1127 </div>1128 """1129 )1130 1131 1132 # --------------------------------------------------------1133 # INPUT / PREVIEW1134 # --------------------------------------------------------1135 1136 with gr.Row():1137 1138 # ====================================================1139 # LEFT COLUMN1140 # ====================================================1141 1142 with gr.Column(1143 scale=1,1144 elem_classes=["card"]1145 ):1146 1147 gr.Markdown(1148 "### ⚙️ Document Configuration",1149 elem_classes=["section-title"]1150 )1151 1152 document_type = gr.Dropdown(1153 1154 choices=SUPPORTED_TYPES,1155 1156 value="Invoice",1157 1158 label="Document Type",1159 1160 info=(1161 "Select the type of "1162 "document you are uploading."1163 )1164 )1165 1166 1167 image_input = gr.Image(1168 1169 label="📤 Upload Document",1170 1171 type="filepath",1172 1173 sources=["upload"],1174 1175 height=3501176 )1177 1178 1179 process_button = gr.Button(1180 1181 "🚀 Extract Information",1182 1183 variant="primary",1184 1185 size="lg"1186 )1187 1188 1189 gr.Markdown(1190 """1191**Supported formats:** JPG, JPEG, PNG1192 1193**Supported documents:**1194 1195🧾 Invoice 1196📄 Resume 1197🪪 ID Card1198 1199**Processing pipeline:**1200