comicocr / main.py
fasdfsa's picture
return aliocr format json
e5baede
# see huggingface/BallonsTranslator/main.py
# see huggingface/project/flask_auto_selection.py
from modules.textdetector.ctd.inference import TextDetector as CTDModel
from modules.ocr.mit48px import Model48pxOCR
CTD_ONNX_PATH = 'data/models/comictextdetector.pt.onnx'
device = 'cpu'
detect_size = 1280
ctd_model = CTDModel(CTD_ONNX_PATH, detect_size=detect_size, device=device)
OCR48PXMODEL_PATH = 'data/models/ocr_ar_48px.ckpt'
ocr_model = Model48pxOCR(OCR48PXMODEL_PATH, device)
import json, os, sys, time, io
import os.path as osp
from PIL import Image
import PIL
import cv2
import numpy as np
is_debug = True
dic_cache = {}
from flask import Flask, request, jsonify
app = Flask(__name__)
import base64
import math, re, uuid
def save_json(filename, dics):
with open(filename, 'w', encoding='utf-8') as fp:
json.dump(dics, fp, indent=4, ensure_ascii=False)
fp.close()
def load_json(filename):
with open(filename, encoding='utf-8') as fp:
js = json.load(fp)
fp.close()
return js
def jsonparse(s):
return json.loads(s, strict=False)
def jsonstring(d):
return json.dumps(d, ensure_ascii=False)
def show_img(image, target_width=400):
# 获取原始图片的宽度和高度
original_height, original_width = image.shape[:2]
# 计算缩放比例和目标高度
scale = target_width / original_width
target_height = int(original_height * scale)
# 等比例缩放图片
resized_image = cv2.resize(image, (target_width, target_height), interpolation=cv2.INTER_AREA)
cv2.imshow("green", resized_image)
cv2.waitKey(0)
return resized_image
# see utils\io_utils.py
def imread(imgpath, read_type=cv2.IMREAD_COLOR, max_retry_limit=5, retry_interval=0.1):
if not osp.exists(imgpath):
return None
num_tries = 0
while True:
try:
img = Image.open(imgpath)
if read_type == cv2.IMREAD_GRAYSCALE:
img = img.convert('L')
img = np.array(img)
if read_type != cv2.IMREAD_GRAYSCALE:
if img.ndim == 3 and img.shape[-1] == 1:
img = img[..., :2]
if img.ndim == 2:
img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)
if img.ndim == 3 and img.shape[-1] == 4:
if np.all(img[..., -1] == 255):
img = np.ascontiguousarray(img[..., :3])
break
except PIL.UnidentifiedImageError as e:
# IMG I/O thread might not finished yet
num_tries += 1
if max_retry_limit is not None and num_tries >= max_retry_limit:
return None
time.sleep(retry_interval)
return img
def chunks(lst, n):
"""Yield successive n-sized chunks from lst."""
for i in range(0, len(lst), n):
yield lst[i:i + n]
def ocr(img):
# All text detectors only support 3 channels input
if img.ndim == 3 and img.shape[2] == 4:
img = cv2.cvtColor(img, cv2.COLOR_RGBA2RGB)
_, mask, blk_list = ctd_model(img)
fnt_rsz = 1.0
fnt_max = -1
fnt_min = -1
for blk in blk_list:
sz = blk._detected_font_size * fnt_rsz
if fnt_max > 0:
sz = min(fnt_max, sz)
if fnt_min > 0:
sz = max(fnt_min, sz)
blk.font_size = sz
blk._detected_font_size = sz
ksize = 2
if ksize > 0:
element = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (2 * ksize + 1, 2 * ksize + 1),(ksize, ksize)) # 创建一个椭圆形的结构元素(kernel),用于后续的形态学操作 # 元素的尺寸 # (ksize, ksize) :椭圆的锚点(中心点)
mask = cv2.dilate(mask, element) # 对 mask 图像进行膨胀操作(dilate),使用上面创建的椭圆结构元素。膨胀操作可以让白色区域(通常是前景或目标区域)变大,常用于去除小的黑洞、连接断开的区域等。
for blk in blk_list:
blk.det_model = 'ctd'
need_save_mask = True
detect_counter = 0
detect_counter += 1
for blk in blk_list:
blk.text = []
split_textblk = False
seg_func = None
model_text_height = 48
model_maxwidth = 8100
from utils.textblock import collect_textblock_regions
chunk_size = 16
regions, textblk_lst_indices = collect_textblock_regions(img, blk_list, model_text_height, model_maxwidth, split_textblk, seg_func)
ocr_model(blk_list, regions, textblk_lst_indices, chunk_size=chunk_size)
img_draw = img.copy()
results = []
# ui\mainwindow.py
for blk in blk_list:
texts = blk.text
lines = blk.lines
results.append( { "texts": texts, "lines":lines } )
for line in blk.lines:
img_draw = cv2.rectangle(img_draw, line[0], line[2], (0, 0, 255), 2)
jsn = { "width": img.shape[1], "height": img.shape[0], "prism_wordsInfo": [] }
for result in results:
texts, lines = ( result["texts"], result["lines"])
word = ''.join(texts)
pos = []
charInfo = []
min_x = 999
min_y = 999
max_x = -1
max_y = -1
for text, line in zip(texts, lines):
lu = line[0]
ru = line[1]
rd = line[2]
ld = line[3]
minx = min(lu[0], ld[0])
maxx = max(ru[0], rd[0])
miny = min(lu[1], ru[1])
maxy = max(rd[1], ld[1])
if min_x > minx:
min_x = minx
if max_x < maxx:
max_x = maxx
if min_y > miny:
min_y = miny
if max_y < maxy:
max_y = maxy
for c in text:
charInfo.append( {"word": c, "x":minx , "y":miny, "w":maxx - minx , "h":maxy - miny, "guid": str( uuid.uuid4() ), "isDeleted": 0 } )
pass
pos = [ { "x":min_x, "y":min_y }, { "x":max_x, "y":min_y }, { "x":max_x, "y":max_y }, { "x":min_x, "y":max_y } ]
jsn["prism_wordsInfo"].append( { "word":word, "x":min_x, "y":min_y, "width":max_x - min_x, "height":max_y - min_y, "pos":pos, "charInfo":charInfo} )
# {
# "width": 1200,
# "height": 1801,
# "prism_wordsInfo": [
# {
# "word": "# 简易字",
# "prob": 0.6273085474967957,
# "x": 593,
# "y": 54,
# "width": 127,
# "height": 25,
# "pos": [
# {
# "x": 593,
# "y": 54
# },
# {
# "x": 720,
# "y": 54
# },
# {
# "x": 720,
# "y": 79
# },
# {
# "x": 593,
# "y": 79
# }
# ],
# "charInfo": [
# {
# "h": 25,
# "w": 43,
# "word": " ",
# "x": 595,
# "y": 54,
# "guid": "164e9305-3e8e-4467-bd76-1c13ee9b6a53",
# "isDeleted": 0
# },
# {
# "h": 25,
# "w": 36,
# "word": "易",
# "x": 638,
# "y": 54,
# "guid": "17319ab0-7dca-4492-b5b3-bfe1d3aee0be",
# "isDeleted": 0
# },
# {
# "h": 25,
# "w": 46,
# "word": "字",
# "x": 674,
# "y": 54,
# "guid": "71cdd286-192e-4461-b89f-89b19548e62f",
# "isDeleted": 0
# }
# ]
# },
return jsn, img_draw
@app.route('/comicocr', methods=['post'])
def comicocr():
global dic_cache
# request.json 只能够接受方法为POST、Body为raw,header 内容为 application/json类型的数据
# print(request.json, type(request.json))
img_b64_str = request.json['img']
img_bytes = base64.b64decode(img_b64_str)
imgData = np.frombuffer(img_bytes, dtype=np.uint8)
img = cv2.imdecode(imgData, -1)
# All text detectors only support 3 channels input
if img.ndim == 3 and img.shape[2] == 4:
img = cv2.cvtColor(img, cv2.COLOR_RGBA2RGB)
# cv2.imshow('test', img)
# cv2.waitKey()
jsn, img_draw = ocr(img)
return jsonify(jsn)
def main():
if is_debug:
img = imread('E:/huggingface/BallonsTranslator/assets/kcc-0010.jpg')
jsn, img_draw = ocr(img)
cv2.imwrite("E:/xxxxxxxxxxxxxxxx.jpg", img_draw)
else:
app.run(host="0.0.0.0", port=2393, debug=True)
return
from modules.textdetector.ctd.inference import TextDetector as CTDModel
from modules.ocr.mit48px import Model48pxOCR
CTD_ONNX_PATH = 'data/models/comictextdetector.pt.onnx'
device = 'cpu'
detect_size = 1280
ctd_model = CTDModel(CTD_ONNX_PATH, detect_size=detect_size, device=device)
OCR48PXMODEL_PATH = 'data/models/ocr_ar_48px.ckpt'
ocr_model = Model48pxOCR(OCR48PXMODEL_PATH, device)
img = imread('E:/huggingface/BallonsTranslator/assets/kcc-0010.jpg')
# All text detectors only support 3 channels input
if img.ndim == 3 and img.shape[2] == 4:
img = cv2.cvtColor(img, cv2.COLOR_RGBA2RGB)
_, mask, blk_list = ctd_model(img)
fnt_rsz = 1.0
fnt_max = -1
fnt_min = -1
for blk in blk_list:
sz = blk._detected_font_size * fnt_rsz
if fnt_max > 0:
sz = min(fnt_max, sz)
if fnt_min > 0:
sz = max(fnt_min, sz)
blk.font_size = sz
blk._detected_font_size = sz
ksize = 2
if ksize > 0:
element = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (2 * ksize + 1, 2 * ksize + 1),(ksize, ksize)) # 创建一个椭圆形的结构元素(kernel),用于后续的形态学操作 # 元素的尺寸 # (ksize, ksize) :椭圆的锚点(中心点)
mask = cv2.dilate(mask, element) # 对 mask 图像进行膨胀操作(dilate),使用上面创建的椭圆结构元素。膨胀操作可以让白色区域(通常是前景或目标区域)变大,常用于去除小的黑洞、连接断开的区域等。
for blk in blk_list:
blk.det_model = 'ctd'
need_save_mask = True
detect_counter = 0
detect_counter += 1
# self.ocr.run_ocr(img, blk_list)
for blk in blk_list:
blk.text = []
split_textblk = False
seg_func = None
model_text_height = 48
model_maxwidth = 8100
from utils.textblock import collect_textblock_regions
chunk_size = 16
regions, textblk_lst_indices = collect_textblock_regions(img, blk_list, model_text_height, model_maxwidth, split_textblk, seg_func)
ocr_model(blk_list, regions, textblk_lst_indices, chunk_size=chunk_size)
img_draw = img.copy()
# from qtpy.QtWidgets import QApplication
# from qtpy.QtGui import QIcon, QFontDatabase, QGuiApplication, QFont, QFontMetrics
# ui\mainwindow.py
for blk in blk_list:
text = blk.get_text()
for line in blk.lines:
img_draw = cv2.rectangle(img_draw, line[0], line[3], (0, 0, 255), 2) # 在一行坚排文字的左边画一条红线
# app_font = QFont('Microsoft YaHei UI')
# fontMetrics = QFontMetrics(app_font)
# rect = fontMetrics.boundingRect(text[0])
# textWidth = rect.width()
pass
# blk.text = self.ocrSubWidget.sub_text(text)
cv2.imwrite("E:/xxxxxxxxxxxxxxxx.jpg", img_draw)
pass
if __name__ == '__main__':
main()