myOCR/dev_image.py

import cv2
import math
from copy import deepcopy
import numpy as np
import os
from TextModel import text_model
import pandas as pd
from image_tools import *
import requests
import json
import time
from tools import get_text



def run_cut(path,draw=False,tabel=True):
    """
    切割图片，生成小图片放在文件夹里
    :param path:
    :param draw:
    :param tabel:
    :return:
    """
    if os.path.exists(r'D:\PaperCut\result\image'):
        os.system(r'rd /s/q D:\PaperCut\result\image')
    if os.path.exists(r'D:\PaperCut\result\text_img'):
        os.system(r'rd /s/q D:\PaperCut\result\text_img')

    if not os.path.exists(r'D:\PaperCut\result\image'):
        os.makedirs(r'D:\PaperCut\result\image')
    if not os.path.exists(r'D:\PaperCut\result\text_img'):
        os.makedirs(r'D:\PaperCut\result\text_img')

    t0 = time.time()
    text_bbx = text_model(path)
    img = cv2.imread(path)

    t1 = time.time()
    print('text时间',t1-t0)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    cv2.imwrite('gray.png', gray)
    ret, binary = cv2.threshold(gray, 150, 255, cv2.THRESH_BINARY)
    if draw:
        cv2.imwrite('ret.png', ret)
        cv2.imwrite('binary.png', binary)
    contours, hierarchy = cv2.findContours(binary, cv2.RETR_LIST, cv2.CHAIN_APPROX_TC89_L1)

    t2= time.time()
    print('轮廓时间',t2-t1)

    text_bbx = get_range(text_bbx) # 坐标转化
    i_box = is_text(text_bbx, contours) # 得到图片坐标和未知坐标
    i_box, text_bbx = processed(np.array(text_bbx),  i_box = np.array(i_box))

    t3 = time.time()

    print('预处理',t3-t2)
    for pixel in i_box:
        x_min, y_min, x_max, y_max = pixel
        if (x_max - x_min) * (y_max - y_min) > 30:
            pts = np.array([[x_min, y_min], [x_max, y_min], [x_max, y_max], [x_min, y_max]], np.int32)
            # 顶点个数：4，矩阵变成4*1*2维
            # OpenCV中需要将多边形的顶点坐标变成顶点数×1×2维的矩阵
            # 这里 reshape 的第一个参数为-1, 表示“任意”，意思是这一维的值是根据后面的维度的计算出来的
            if draw:
                pts = pts.reshape((-1, 1, 2))
                cv2.polylines(img, [pts], True, (0, 0, 255))  # 画文字图

            cv2.imwrite('./result/image/%d-%d-%d-%d.png' % (y_min, y_max, x_min, x_max), img[y_min:y_max, x_min:x_max])


            if tabel:# and (x_max - x_min) * (y_max - y_min) * 8 > img.shape[0] * img.shape[1]:
                files = {'file': open(r'./result/image/%d-%d-%d-%d.png' % (y_min, y_max, x_min, x_max),
                                      'rb')}  # 此处是重点！我们操作文件上传的时候，把目标文件以open打开，然后存储到变量file里面存到一个字典里面
                upload_data = {"parentId": "", "fileCategory": "personal",
                               "fileName": '%d-%d-%d-%d.png' % (y_min, y_max, x_min, x_max),
                               "uoType": 1}

                tabel_img = np.array(
                    json.loads(requests.post('http://192.168.1.192:25255/img2maple', upload_data, files=files).text))
                if np.sum(tabel_img):
                    cv2.imwrite('./result/image/%d-%d-%d-%d.png' % (y_min, y_max, x_min, x_max), tabel_img)

    for pixel in text_bbx:
        x_min, y_min, x_max, y_max = pixel

        pts = np.array([[x_min, y_min], [x_max, y_min], [x_max, y_max], [x_min, y_max]], np.int32)
        # 顶点个数：4，矩阵变成4*1*2维
        # OpenCV中需要将多边形的顶点坐标变成顶点数×1×2维的矩阵
        # 这里 reshape 的第一个参数为-1, 表示“任意”，意思是这一维的值是根据后面的维度的计算出来的
        if draw:
            pts = pts.reshape((-1, 1, 2))
            cv2.polylines(img, [pts], True, (0, 255, 0))  # 画插图



        cv2.imwrite('./result/text_img/%d-%d-%d-%d.png' % (y_min, y_max, x_min, x_max), img[y_min:y_max, x_min:x_max])

    get_text()
    t4=time.time()
    print('画图+ocr',t4-t3)
    print('总时间',t4-t0)

            # contours 轮廓所有的点
    # cv2.drawContours(img, contours, -1, (0, 0, 255), 3)
    if __name__ == '__main__':
        cv2.imshow("img", img)
        cv2.waitKey(0)

        cv2.imwrite('bbb.png', img)


if __name__ == '__main__':

    start = time.time()
    #19 23 29 30
    #
    #36?33
    for i in range(10):
        run_cut('./image_dir/1.jpg',draw=True,tabel=False)