以下内容涉及opencv的膨胀腐蚀等形态学处理，仿射透视矫正，如需要请查详情： 形态学处理、 仿射透视矫正 、表格提取的OpenCV-python实现，思路参考
import cv2
import os
import numpy as np
from imutils.perspective import four_point_transform
def FindContours(img_path):
src_img = cv2.imread(img_path)
src_img0 = cv2.cvtColor(src_img, cv2.COLOR_BGR2GRAY)
src_img0 = cv2.GaussianBlur(src_img0,(3,3),0)
src_img1 = cv2.bitwise_not(src_img0)
AdaptiveThreshold = cv2.adaptiveThreshold(src_img1, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 15, -2)
horizontal = AdaptiveThreshold.copy()
vertical = AdaptiveThreshold.copy()
scale = 20
horizontalSize = int(horizontal.shape[1]/scale)
horizontalStructure = cv2.getStructuringElement(cv2.MORPH_RECT, (horizontalSize, 1))
horizontal = cv2.erode(horizontal, horizontalStructure)
horizontal = cv2.dilate(horizontal, horizontalStructure)
cv2.imshow("horizontal", horizontal)
cv2.waitKey(0)
verticalsize = int(vertical.shape[1]/scale)
verticalStructure = cv2.getStructuringElement(cv2.MORPH_RECT, (1, verticalsize))
vertical = cv2.erode(vertical, verticalStructure, (-1, -1))
vertical = cv2.dilate(vertical, verticalStructure, (-1, -1))
cv2.imshow("verticalsize", vertical)
cv2.waitKey(0)
mask = horizontal + vertical
#np.count_nonzero(mask, axis=0) #对列统计白色（非零值）
#np.count_nonzero(mask, axis=1) #对行统计白色（非零值）
cv2.imshow("mask", mask)
cv2.waitKey(0)
Net_img = cv2.bitwise_and(horizontal, vertical)
cv2.imshow("Net_img", Net_img)
cv2.waitKey(0)
cv2.destroyAllWindows()
_, contours, hierarchy = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
#=========================绘出所有轮廓=========================
# IMG = cv2.drawContours(src_img0, contours, -1, (0, 255, 255), 2)
# cv2.imshow('IMG', IMG)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
#=============================================================
return src_img,Net_img,contours
def get_Affine_Location(src_img,Net_img,contours,cutImg_name,cutImg_path):
contours = sorted(contours, key=cv2.contourArea, reverse=True)
for i in range(len(contours)):
area0 = cv2.contourArea(contours[i])
if area0<20:continue
# =======================查找每个表的关节数====================
epsilon = 0.1 * cv2.arcLength(contours[i], True)
approx = cv2.approxPolyDP(contours[i], epsilon, True) # 获取近似轮廓
x1, y1, w1, h1 = cv2.boundingRect(approx)
roi = Net_img[int(y1):int(y1+h1) ,int(x1):int(x1+w1)]
_, roi_contours, hierarchy = cv2.findContours(roi, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
print('len(roi_contours):',len(roi_contours))
if len(roi_contours)<4:continue
src_img1 = cv2.rectangle(src_img, (x1, y1),(x1+w1,y1+h1), (0,255,0), 2)
cut_img = src_img[y1:y1+h1,x1:x1+w1]
cv2.imwrite(cutImg_path+'/'+cutImg_name+'_'+str(i)+'.png',cut_img) # 保存截取的图片
cv2.imshow('src_img_'+str(i),src_img1)
cv2.waitKey(0)
cv2.destroyAllWindows()
# =========================绘出最大轮廓凸包并矫正图像==========================
# epsilon = 0.1 * cv2.arcLength(contours[0], True)
# approx = cv2.approxPolyDP(contours[0], epsilon, True) # 获取近似轮廓
# hull = cv2.convexHull(approx) # 默认返回坐标点
# hull_img = cv2.polylines(src_img, [hull], True, (0, 255, 0), 2)
# cv2.imshow('hull_img', hull_img)
# cv2.waitKey(0)
#
# if len(hull) == 4:
# dst = four_point_transform(src_img, hull.reshape(4,2)) # 矫正变换
# cv2.imwrite(cutImg_path+'/'+cutImg_name+'max.png', dst) # 保存截取的图片
# cv2.imshow("result", dst)
# cv2.waitKey(0)
#
# Img_max = cv2.drawContours(src_img, contours, 0, (0, 255, 0), 2, 1)
# cv2.imshow('ImgMax', Img_max)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
if __name__ == '__main__':
input_Path = 'D:/python_script/ffff/20160727151610172.png'
cutImg_path = 'D:\\python_script\\ffff'
cutImg_name = input_Path.split('/')[-1][:-4]
src_img, Net_img, contours = FindContours(input_Path)
get_Affine_Location(src_img, Net_img, contours,cutImg_name,cutImg_path)
效果截图如下：
参考与鸣谢
