68 lines
2.4 KiB
Python
68 lines
2.4 KiB
Python
|
|
import cv2
|
|
import os
|
|
import json
|
|
import difflib
|
|
import numpy as np
|
|
from paddleocr import PaddleOCR
|
|
|
|
def similarity(s1, s2):
|
|
return difflib.SequenceMatcher(None, s1, s2).ratio()
|
|
|
|
def test_rotation_ocr():
|
|
target = "威凯检测技术有限公司"
|
|
img_path = "seal_cropped.png"
|
|
if not os.path.exists(img_path):
|
|
print(f"Error: {img_path} not found")
|
|
return
|
|
|
|
img = cv2.imread(img_path)
|
|
h, w = img.shape[:2]
|
|
cx, cy = w // 2, h // 2
|
|
|
|
ocr = PaddleOCR(use_angle_cls=True, lang='ch')
|
|
|
|
results = []
|
|
# Test angles from -90 to 90 with step 15
|
|
for angle in range(-120, 121, 15):
|
|
M = cv2.getRotationMatrix2D((cx, cy), angle, 1.0)
|
|
rotated = cv2.warpAffine(img, M, (w, h), borderValue=(255, 255, 255))
|
|
|
|
# We can also try different scales or enhancements
|
|
res = ocr.ocr(rotated)
|
|
|
|
if res:
|
|
# Handle list of dicts (PaddleX format)
|
|
for page in res:
|
|
if 'rec_texts' in page:
|
|
for text in page['rec_texts']:
|
|
clean_text = text.replace(" ", "")
|
|
sim = similarity(target, clean_text)
|
|
if sim > 0.1:
|
|
print(f"Angle: {angle} | Sim: {sim:.4f} | Text: {clean_text}")
|
|
results.append({
|
|
"angle": angle,
|
|
"text": clean_text,
|
|
"sim": sim
|
|
})
|
|
# Handle standard PaddleOCR format just in case
|
|
elif isinstance(page, list):
|
|
for line in page:
|
|
if isinstance(line, list) and len(line) > 1:
|
|
clean_text = line[1][0].replace(" ", "")
|
|
sim = similarity(target, clean_text)
|
|
if sim > 0.1:
|
|
print(f"Angle: {angle} | Sim: {sim:.4f} | Text: {clean_text}")
|
|
results.append({
|
|
"angle": angle,
|
|
"text": clean_text,
|
|
"sim": sim
|
|
})
|
|
|
|
results.sort(key=lambda x: x['sim'], reverse=True)
|
|
with open("rotation_ocr_results.json", "w", encoding="utf-8") as f:
|
|
json.dump(results, f, ensure_ascii=False, indent=2)
|
|
|
|
if __name__ == "__main__":
|
|
test_rotation_ocr()
|