logo

开发者热搜

基于Python与OpenCV的票据识别系统开发指南

作者：很酷cat2025.09.19 17:59浏览量：1

简介：本文详细介绍如何利用Python和OpenCV实现票据识别功能，涵盖图像预处理、边缘检测、轮廓提取、文字定位与识别等关键步骤，并提供完整的代码实现。

基于Python与OpenCV的票据识别系统开发指南

一、票据识别技术背景与意义

票据识别是OCR（光学字符识别）技术的重要应用场景，广泛应用于财务报销、银行票据处理、税务申报等领域。传统票据识别系统多依赖商业OCR引擎，存在成本高、定制性差等问题。基于Python和OpenCV的开源方案，不仅能显著降低开发成本，还能通过算法优化提升识别准确率。

OpenCV作为计算机视觉领域的标准库，提供了丰富的图像处理函数，结合Python的简洁语法，可快速构建票据识别系统。本方案通过图像预处理、边缘检测、轮廓提取等步骤，实现票据区域的精准定位，为后续文字识别奠定基础。

二、票据识别系统核心流程

1. 图像预处理

票据图像常存在倾斜、光照不均、背景干扰等问题，预处理环节至关重要。主要步骤包括：

  • 灰度化转换：将彩色图像转为灰度图，减少计算量

    1. import cv2
    2. def rgb2gray(img_path):
    3.   img = cv2.imread(img_path)
    4.   gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    5.   return gray

  • 高斯模糊：消除高频噪声

    1. def gaussian_blur(img):
    2.   blurred = cv2.GaussianBlur(img, (5,5), 0)
    3.   return blurred

  • 自适应阈值处理：解决光照不均问题

    1. def adaptive_threshold(img):
    2.   thresh = cv2.adaptiveThreshold(img, 255, 
    3.                                 cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
    4.                                 cv2.THRESH_BINARY_INV, 11, 2)
    5.   return thresh

2. 边缘检测与轮廓提取

Canny边缘检测算法能有效提取票据边缘：

  1. def canny_edge(img):
  2.     edges = cv2.Canny(img, 50, 150)
  3.     return edges

结合形态学操作优化边缘：

  1. def morph_operations(img):
  2.     kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5,5))
  3.     dilated = cv2.dilate(img, kernel, iterations=1)
  4.     eroded = cv2.erode(dilated, kernel, iterations=1)
  5.     return eroded

轮廓提取与筛选：

  1. def find_contours(img):
  2.     contours, _ = cv2.findContours(img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
  3.     # 筛选面积大于阈值的轮廓
  4.     min_area = 1000
  5.     valid_contours = [cnt for cnt in contours if cv2.contourArea(cnt) > min_area]
  6.     return valid_contours

3. 票据区域定位与矫正

通过轮廓分析确定票据位置：

  1. def locate_receipt(contours):
  2.     # 假设最大轮廓为票据区域
  3.     if not contours:
  4.         return None
  5.     target_cnt = max(contours, key=cv2.contourArea)
  7.     # 获取边界矩形
  8.     x,y,w,h = cv2.boundingRect(target_cnt)
  9.     return (x,y,w,h)

透视变换矫正倾斜票据：

  1. def perspective_transform(img, pts):
  2.     # pts为票据四个顶点坐标
  3.     rect = np.array(pts, dtype="float32")
  4.     (tl, tr, br, bl) = rect
  6.     # 计算新图像尺寸
  7.     widthA = np.sqrt(((br[0] - bl[0]) ** 2) + ((br[1] - bl[1]) ** 2))
  8.     widthB = np.sqrt(((tr[0] - tl[0]) ** 2) + ((tr[1] - tl[1]) ** 2))
  9.     maxWidth = max(int(widthA), int(widthB))
  11.     heightA = np.sqrt(((tr[0] - br[0]) ** 2) + ((tr[1] - br[1]) ** 2))
  12.     heightB = np.sqrt(((tl[0] - bl[0]) ** 2) + ((tl[1] - bl[1]) ** 2))
  13.     maxHeight = max(int(heightA), int(heightB))
  15.     dst = np.array([
  16.         [0, 0],
  17.         [maxWidth - 1, 0],
  18.         [maxWidth - 1, maxHeight - 1],
  19.         [0, maxHeight - 1]], dtype="float32")
  21.     M = cv2.getPerspectiveTransform(rect, dst)
  22.     warped = cv2.warpPerspective(img, M, (maxWidth, maxHeight))
  23.     return warped

4. 文字区域提取与识别

使用投影法分割文字行：

  1. def extract_text_lines(img):
  2.     # 计算垂直投影
  3.     hist = cv2.reduce(img, 1, cv2.REDUCE_AVG).reshape(-1)
  5.     # 寻找分割点
  6.     threshold = hist.max() * 0.1
  7.     lines = []
  8.     start = 0
  9.     for i in range(len(hist)):
  10.         if hist[i] > threshold and (i == 0 or hist[i-1] <= threshold):
  11.             start = i
  12.         elif hist[i] <= threshold and i > 0 and hist[i-1] > threshold:
  13.             lines.append((start, i))
  15.     # 提取文字行
  16.     text_lines = []
  17.     for (start, end) in lines:
  18.         roi = img[start:end, :]
  19.         text_lines.append(roi)
  20.     return text_lines

三、完整代码实现

  1. import cv2
  2. import numpy as np
  4. class ReceiptRecognizer:
  5.     def __init__(self):
  6.         pass
  8.     def preprocess(self, img_path):
  9.         img = cv2.imread(img_path)
  10.         gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
  11.         blurred = cv2.GaussianBlur(gray, (5,5), 0)
  12.         thresh = cv2.adaptiveThreshold(blurred, 255, 
  13.                                       cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
  14.                                       cv2.THRESH_BINARY_INV, 11, 2)
  15.         return thresh
  17.     def detect_edges(self, img):
  18.         edges = cv2.Canny(img, 50, 150)
  19.         kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5,5))
  20.         dilated = cv2.dilate(edges, kernel, iterations=1)
  21.         return dilated
  23.     def find_receipt_contour(self, edges):
  24.         contours, _ = cv2.findContours(edges, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
  25.         if not contours:
  26.             return None
  27.         target_cnt = max(contours, key=cv2.contourArea)
  28.         return target_cnt
  30.     def perspective_correction(self, img, contour):
  31.         epsilon = 0.02 * cv2.arcLength(contour, True)
  32.         approx = cv2.approxPolyDP(contour, epsilon, True)
  33.         approx = np.array(approx, dtype="float32")
  35.         # 排序四个顶点 (tl, tr, br, bl)
  36.         rect = order_points(approx)
  38.         # 透视变换
  39.         (tl, tr, br, bl) = rect
  40.         widthA = np.sqrt(((br[0] - bl[0]) ** 2) + ((br[1] - bl[1]) ** 2))
  41.         widthB = np.sqrt(((tr[0] - tl[0]) ** 2) + ((tr[1] - tl[1]) ** 2))
  42.         maxWidth = max(int(widthA), int(widthB))
  44.         heightA = np.sqrt(((tr[0] - br[0]) ** 2) + ((tr[1] - br[1]) ** 2))
  45.         heightB = np.sqrt(((tl[0] - bl[0]) ** 2) + ((tl[1] - bl[1]) ** 2))
  46.         maxHeight = max(int(heightA), int(heightB))
  48.         dst = np.array([
  49.             [0, 0],
  50.             [maxWidth - 1, 0],
  51.             [maxWidth - 1, maxHeight - 1],
  52.             [0, maxHeight - 1]], dtype="float32")
  54.         M = cv2.getPerspectiveTransform(rect, dst)
  55.         warped = cv2.warpPerspective(img, M, (maxWidth, maxHeight))
  56.         return warped
  58.     def extract_text_regions(self, warped_img):
  59.         # 计算垂直投影
  60.         hist = cv2.reduce(warped_img, 1, cv2.REDUCE_AVG).reshape(-1)
  62.         # 寻找分割点
  63.         threshold = hist.max() * 0.1
  64.         lines = []
  65.         start = 0
  66.         for i in range(len(hist)):
  67.             if hist[i] > threshold and (i == 0 or hist[i-1] <= threshold):
  68.                 start = i
  69.             elif hist[i] <= threshold and i > 0 and hist[i-1] > threshold:
  70.                 lines.append((start, i))
  72.         # 提取文字行
  73.         text_lines = []
  74.         for (start, end) in lines:
  75.             roi = warped_img[start:end, :]
  76.             text_lines.append(roi)
  77.         return text_lines
  79.     def recognize(self, img_path):
  80.         # 1. 预处理
  81.         processed = self.preprocess(img_path)
  83.         # 2. 边缘检测
  84.         edges = self.detect_edges(processed)
  86.         # 3. 轮廓提取
  87.         contour = self.find_receipt_contour(edges)
  88.         if contour is None:
  89.             return None
  91.         # 4. 透视矫正
  92.         img = cv2.imread(img_path)
  93.         warped = self.perspective_correction(img, contour)
  95.         # 5. 文字区域提取
  96.         gray_warped = cv2.cvtColor(warped, cv2.COLOR_BGR2GRAY)
  97.         _, binary = cv2.threshold(gray_warped, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
  98.         text_lines = self.extract_text_regions(binary)
  100.         return text_lines
  102. def order_points(pts):
  103.     # 初始化坐标点
  104.     rect = np.zeros((4, 2), dtype="float32")
  106.     # 计算四个点的总和与差值
  107.     s = pts.sum(axis=1)
  108.     rect[0] = pts[np.argmin(s)]  # 左上角
  109.     rect[2] = pts[np.argmax(s)]  # 右下角
  111.     diff = np.diff(pts, axis=1)
  112.     rect[1] = pts[np.argmin(diff)]  # 右上角
  113.     rect[3] = pts[np.argmax(diff)]  # 左下角
  115.     return rect
  117. # 使用示例
  118. if __name__ == "__main__":
  119.     recognizer = ReceiptRecognizer()
  120.     text_lines = recognizer.recognize("receipt.jpg")
  122.     if text_lines:
  123.         for i, line in enumerate(text_lines):
  124.             cv2.imwrite(f"line_{i}.png", line)

四、性能优化建议

  1. 多尺度处理：对图像进行不同尺度的处理，提高小票据的识别率
  2. 并行计算：利用多线程处理多个票据图像
  3. 深度学习融合：结合CNN网络进行更精准的文字定位
  4. 后处理优化：添加文字方向校正、字符分割等后处理步骤

五、应用场景扩展

  1. 财务报销系统：自动识别发票信息并填充报销单
  2. 银行票据处理：识别支票、汇款单等金融票据
  3. 物流单据管理：自动提取运单号、收发货人信息
  4. 医疗票据处理：识别医保报销所需的各种票据

本方案通过Python和OpenCV实现了票据识别的核心功能，开发者可根据实际需求进行功能扩展和性能优化。随着计算机视觉技术的不断发展，票据识别系统的准确率和效率将持续提升，为各行业的数字化转型提供有力支持。

相关文章推荐

发表评论

最热文章

关于作者

  • 被阅读数
  • 被赞数
  • 被收藏数
活动
咨询