ocr/gendata.py

"""
转换一章数据
"""

import json
import glob
from pathlib import Path
import cv2
import numpy as np


pth_txt = "out/text/text.txt"


PageIds = []

with open(pth_txt, 'r', encoding='utf-8') as f:
    for txtline in f:
        txtline = txtline.strip()
        if not txtline:
            continue
        data = json.loads(txtline)
        lines = data['lines'] 
        for line in lines:
            if "pagePass" in line:
                PageId = line["pagePass"]["PageId"]
                PageIds.append(PageId)


dic = {}
dic_img = {}
num = 0
charTexts = ""
for PageId in PageIds:
    print(PageId)
    paths = glob.glob(f'./out/coords/*.json', recursive=False)
    for pth_json in paths:
        with open(pth_json, "r", encoding="utf-8") as fp:

            stem = Path(pth_json).stem
            image_id = stem.split('_')[-1]  # 坐标关联的图片id

            txt = fp.read()
            data = json.loads(txt)
            pageId2WordBoxContent = data["data"]["pageId2WordBoxContent"]
            if PageId not in pageId2WordBoxContent:
                continue

            wordBoxContents = pageId2WordBoxContent[PageId]  # 这一页对应的坐标



            if PageId not in dic:
                dic[PageId] = True
            else:
                continue



            images = glob.glob(f"./out/images/*{image_id}.webp", recursive=False)
            assert len(images) > 0
            pth_img = images[0]
            pth_img = str( Path(pth_img).resolve() )



            new_img = False
            if PageId not in dic_img:
                new_img = True
                img = cv2.imread(pth_img)
                if len(img.shape) != 3:  # 转彩图
                    img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
                img_color = img.copy()
                dic_img[PageId] = { "img_color": img_color, "img": img }
            else:
                img = dic_img[PageId]["img"]
                img_color = dic_img[PageId]["img_color"]



            

            
            wordBoxList = wordBoxContents["wordBoxList"]
            for wordBox in wordBoxList:
                if "charPoly" not in wordBox:
                    continue
                charPoly = wordBox["charPoly"]
                lu = [ charPoly['x0'], charPoly['y0'] ]
                ru = [ charPoly['x1'], charPoly['y1'] ]
                rb = [ charPoly['x2'], charPoly['y2'] ]
                lb = [ charPoly['x3'], charPoly['y3'] ]
                points = [lu, ru, rb, lb]
                points = np.array(points)   
                cv2.polylines(img_color, [points], isClosed=True, color=(255, 0, 0), thickness=1) 

                charText = wordBox["charText"]
                charTexts += charText
                print(charText, end='')

            charTexts += "\n\n\n"
            print("\n\n\n")


            if new_img:
                cv2.imwrite(f"./out2/{num:04d}_{image_id}.webp", img_color)
                num += 1

        break

with open("out.txt", 'w', encoding='utf-8') as f:
    f.write(charTexts)

pass