如何用Python及图像库拆分双栏扫描文档段落与单个问题？

一、前置准备 #

先安装所需依赖库：

pip install opencv-python pillow pytesseract

注意：需提前在系统中安装Tesseract OCR引擎，确保pytesseract能正常调用。

二、扫描图像单问题拆分实现 #

核心思路 #

先对图像做预处理提升OCR识别精度，再提取文本，最后根据问题的编号特征（如1.、2.这类前缀）拆分出单个问题。

代码示例 #

import cv2
import pytesseract
import re

def preprocess_image(image_path):
    # 读取图像
    img = cv2.imread(image_path)
    # 灰度化
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    # 二值化（自适应阈值，处理扫描阴影）
    thresh = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 11, 2)
    # 去噪
    denoised = cv2.medianBlur(thresh, 3)
    return denoised

def split_single_questions(image_path):
    processed_img = preprocess_image(image_path)
    # OCR提取文本
    text = pytesseract.image_to_string(processed_img, lang='eng')
    # 按问题编号拆分（匹配类似"1. "、"2. "的前缀）
    question_pattern = re.compile(r'(^\d+\.\s)', re.MULTILINE)
    questions = re.split(question_pattern, text)
    # 整理成单个问题列表
    formatted_questions = []
    for i in range(1, len(questions), 2):
        formatted_questions.append(f"{questions[i]}{questions[i+1]}".strip())
    return formatted_questions

# 调用示例
# questions = split_single_questions("your_image_path.jpg")
# for q in questions:
#     print("---问题分割线---")
#     print(q)

三、双栏扫描文档段落拆分实现 #

核心思路 #

通过分析OCR识别出的文本行坐标，聚类区分左右栏，再按栏内的行顺序合并为段落，同时处理可能的跨栏段落。

代码示例 #

import cv2
import pytesseract
from sklearn.cluster import KMeans
import numpy as np

def get_text_lines_with_coords(image_path):
    processed_img = preprocess_image(image_path)
    # 获取带坐标的OCR结果
    d = pytesseract.image_to_data(processed_img, lang='eng', output_type=pytesseract.Output.DICT)
    n_boxes = len(d['text'])
    text_lines = []
    # 按行分组（同一y坐标区间视为一行）
    current_line = {"text": "", "x": [], "y": d['top'][0]}
    for i in range(n_boxes):
        if int(d['conf'][i]) > 60:  # 过滤低置信度文本
            # 判断是否为同一行（y坐标差小于10像素）
            if abs(d['top'][i] - current_line["y"]) < 10:
                current_line["text"] += d['text'][i] + " "
                current_line["x"].append(d['left'][i])
            else:
                text_lines.append(current_line)
                current_line = {"text": d['text'][i] + " ", "x": [d['left'][i]], "y": d['top'][i]}
    text_lines.append(current_line)
    return text_lines

def split_two_column_paragraphs(image_path):
    text_lines = get_text_lines_with_coords(image_path)
    # 提取每行的平均x坐标，用于聚类
    x_coords = np.array([np.mean(line["x"]) for line in text_lines]).reshape(-1, 1)
    # KMeans聚类分为2类（左右栏）
    kmeans = KMeans(n_clusters=2, random_state=0).fit(x_coords)
    labels = kmeans.labels_
    # 区分左右栏（按聚类中心的x坐标排序）
    cluster_centers = kmeans.cluster_centers_.flatten()
    left_col_idx = 0 if cluster_centers[0] < cluster_centers[1] else 1
    right_col_idx = 1 - left_col_idx

    # 按栏分组
    left_lines = [line["text"].strip() for line, label in zip(text_lines, labels) if label == left_col_idx]
    right_lines = [line["text"].strip() for line, label in zip(text_lines, labels) if label == right_col_idx]

    # 合并段落（按行顺序，空行分隔段落）
    def lines_to_paragraphs(lines):
        paragraphs = []
        current_paragraph = []
        for line in lines:
            if line.strip() == "":
                if current_paragraph:
                    paragraphs.append(" ".join(current_paragraph))
                    current_paragraph = []
            else:
                current_paragraph.append(line)
        if current_paragraph:
            paragraphs.append(" ".join(current_paragraph))
        return paragraphs

    left_paragraphs = lines_to_paragraphs(left_lines)
    right_paragraphs = lines_to_paragraphs(right_lines)
    # 合并左右栏段落（先左后右，或按实际排版调整顺序）
    all_paragraphs = left_paragraphs + right_paragraphs
    return all_paragraphs

# 调用示例
# paragraphs = split_two_column_paragraphs("your_image_path.jpg")
# for p in paragraphs:
#     print("---段落分割线---")
#     print(p)

内容的提问来源于stack exchange，提问作者sibi kanagaraj