从原理到实战详解Python实现PDF图片OCR识别全流程

# Ubuntu 24.04安装命令
sudo apt install tesseract-ocr tesseract-ocr-chi-sim libtesseract-dev
pip install pytesseract pdf2image opencv-python numpy

from pdf2image import convert_from_path
 
def pdf_to_images(pdf_path, dpi=300):
    images = convert_from_path(
        pdf_path,
        dpi=dpi,
        output_folder="temp_images",
        fmt="png",
        thread_count=4  # 启用多线程加速
    )
    return images

import cv2
import numpy as np
 
def preprocess_image(image_path):
    img = cv2.imread(image_path)
    # 灰度化+二值化
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    _, binary = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
    
    # 去噪处理
    denoised = cv2.fastNlMeansDenoising(binary, h=10)
    
    # 倾斜校正（2025年新增算法）
    coords = np.column_stack(np.where(denoised > 0))
    angle = cv2.minAreaRect(coords)[-1]
    if angle < -45:
        angle = -(90 + angle)
    else:
        angle = -angle
    (h, w) = denoised.shape[:2]
    center = (w // 2, h // 2)
    M = cv2.getRotationMatrix2D(center, angle, 1.0)
    rotated = cv2.warpAffine(denoised, M, (w, h), flags=cv2.INTER_CUBIC, borderMode=cv2.BORDER_REPLICATE)
    
    return rotated

import pytesseract
 
def ocr_recognition(image, lang='chi_sim+eng'):
    custom_config = r'--oem 3 --psm 6'  # LSTM+自动分页模式
    text = pytesseract.image_to_string(
        image,
        lang=lang,
        config=custom_config,
        output_type=pytesseract.Output.DICT  # 返回结构化数据
    )
    return text

import re
 
def post_process(raw_text):
    # 去除OCR常见错误
    cleaned = re.sub(r'\s+', ' ', raw_text)  # 合并多余空格
    cleaned = re.sub(r'[^\w\s\u4e00-\u9fff]', '', cleaned)  # 过滤特殊字符
    
    # 段落重建（基于2025年改进的NLP算法）
    sentences = re.split(r'(?<=[。！？])', cleaned)
    paragraphs = []
    current_para = []
    
    for sent in sentences:
        if len(current_para) > 0 and len(current_para[-1]) + len(sent) > 120:
            paragraphs.append(''.join(current_para))
            current_para = [sent]
        else:
            current_para.append(sent)
    
    if current_para:
        paragraphs.append(''.join(current_para))
    
    return '\n\n'.join(paragraphs)

def ensemble_ocr(image_path):
    # Tesseract识别
    tess_text = pytesseract.image_to_string(image_path, lang='chi_sim+eng')
    
    # EasyOCR识别
    import easyocr
    reader = easyocr.Reader(['ch_sim', 'en'])
    easy_result = reader.readtext(image_path, detail=0)
    easy_text = ' '.join(easy_result)
    
    # 相似度融合（2025年新增算法）
    from difflib import SequenceMatcher
    similarity = SequenceMatcher(None, tess_text, easy_text).ratio()
    
    if similarity > 0.85:
        return max(tess_text, easy_text, key=len)  # 取较长结果
    else:
        return f"Tesseract:\n{tess_text}\n\nEasyOCR:\n{easy_text}"

def table_recognition(image_path):
    import layoutparser as lp
    
    # 加载预训练模型（2025年新增表格专用模型）
    model = lp.Detectron2LayoutModel('lp://PubLayNet/mask_rcnn_R_50_FPN_3x/config')
    image = lp.load_image(image_path)
    layout = model.detect(image)
    
    table_blocks = [b for b in layout if b.type == 'Table']
    results = []
    
    for block in table_blocks:
        # 提取表格区域
        table_img = block.pad(10).crop_image(image)
        
        # 使用PaddleOCR表格识别
        from paddleocr import PaddleOCR
        ocr = PaddleOCR(use_angle_cls=True, lang='ch')
        table_result = ocr.ocr(table_img, cls=True, table=True)
        
        # 转换为CSV格式
        csv_lines = []
        for line in table_result[0]['data']:
            csv_lines.append(','.join([cell[1][0] for cell in line]))
        results.append('\n'.join(csv_lines))
    
    return results

def invoice_parser(image_path):
    # 使用PaddleOCR的票据识别模型
    ocr = PaddleOCR(rec_model_dir='ch_PP-OCRv4_rec_infer',
                   det_db_thresh=0.3,
                   use_dilation=True)
    
    result = ocr.ocr(image_path, cls=True)
    
    # 字段映射规则
    field_map = {
        '发票号码': ['INVOICE NO.', '发票号'],
        '开票日期': ['DATE', '开票日期'],
        '金额': ['AMOUNT', '金额（小写）']
    }
    
    extracted_data = {}
    for line in result:
        for field, keywords in field_map.items():
            if any(kw in line[1][0] for kw in keywords):
                extracted_data[field] = line[1][0].replace(keywords[0], '').strip()
    
    return extracted_data

def ancient_book_ocr(image_path):
    # 自定义预处理
    img = preprocess_image(image_path)
    
    # 使用Tesseract的古籍专用配置
    custom_config = r'--oem 3 --psm 6 -c tessedit_char_whitelist=零壹贰叁肆伍陆柒捌玖拾佰仟万亿'
    
    # 竖排文字处理
    from PIL import Image
    img_pil = Image.fromarray(img)
    rotated_img = img_pil.rotate(90, expand=True)
    
    text = pytesseract.image_to_string(
        rotated_img,
        config=custom_config,
        lang='chi_tra'  # 繁体中文
    )
    
    # 繁简转换（使用2025年更新的opencc库）
    import opencc
    cc = opencc.OpenCC('t2s.json')  # 繁体转简体
    return cc.convert(text)

# 检查语言包是否安装
ls /usr/share/tesseract-ocr/5/tessdata/ | grep chi_sim
 
# 验证Tesseract路径配置
import pytesseract
print(pytesseract.get_tesseract_version())

from concurrent.futures import ThreadPoolExecutor
 
def parallel_ocr(image_list):
    with ThreadPoolExecutor(max_workers=8) as executor:
        results = list(executor.map(ocr_recognition, image_list))
    return results

def adaptive_ocr(image_path):
    layout = analyze_layout(image_path)  # 自定义版面分析函数
    
    full_text = []
    for region in layout:
        if region['type'] == 'text':
            text = ocr_recognition(region['image'])
        elif region['type'] == 'table':
            text = table_to_markdown(region['image'])
        full_text.append(text)
    
    return '\n'.join(full_text)