5
0
mirror of https://github.com/sml2h3/ddddocr.git synced 2025-05-05 00:22:26 +08:00
ddddocr/test.py
2022-01-25 23:46:53 +08:00

53 lines
1.1 KiB
Python

import ddddocr
# # 识别
# ocr = ddddocr.DdddOcr(old=True)
#
# with open('test.jpg', 'rb') as f:
# img_bytes = f.read()
#
# res = ocr.classification(img_bytes)
# print(res)
# 目标检测
# det = ddddocr.DdddOcr(det=True)
#
# with open('微信图片_20220123122537.jpg', 'rb') as f:
# img_bytes = f.read()
#
# res = det.detection(img_bytes)
# print(res)
# 滑块模板匹配方式
import cv2
im = cv2.imread("b.png")
det = ddddocr.DdddOcr(det=False, ocr=False)
with open('t.png', 'rb') as f:
target_bytes = f.read()
with open('b.png', 'rb') as f:
background_bytes = f.read()
res = det.slide_match(target_bytes, background_bytes)
print(res)
x = res["target"]
print(x)
# im = cv2.line(im, (int(x[0]), 0), (int(x[0]), 200), color=(0, 0, 255), thickness=2)
im = cv2.rectangle(im, (int(x[0]), int(x[1])), (int(x[2]), int(x[3])), color=(0, 0, 255), thickness=2)
cv2.imwrite("res.jpg", im)
# det = ddddocr.DdddOcr(det=False, ocr=False)
#
# with open('bg.jpg', 'rb') as f:
# target_bytes = f.read()
#
# with open('fullpage.jpg', 'rb') as f:
# background_bytes = f.read()
#
#
# res = det.slide_comparison(target_bytes, background_bytes)
#
# print(res)