refactor(underlying): 抽 analyse_pic_word OCR 管道到 ocr_pipeline (1939 → 1560)
underlying_operations.analyse_pic_word L723-1102 是 380 行 OCR 主管道: 路径解析 + 字节缓存 + PIL 加载 + change_color 预处理 + 调试图像保存 + EasyOCR/PaddleOCR 调用 + 置信度提取 + 缓存写回 + 成功识别落盘。 抽到 underlying_operations_pkg/ocr_pipeline.py: - analyse_pic_word(self, picname, change_color, pic_path, return_confidence) - _save_debug_preprocessed: 预处理后的调试图保存(仅 SAVE_OCR_DEBUG_IMAGES=1) - _save_debug_after_ocr: 含 OCR 文本预览 + 置信度的 after_ocr 调试图 - _save_success_pic: 成功识别落盘到 ocr/<category>/<picname>_success.png 主类方法替换为 4 行委托。原行为完全保留:缓存 / 二值化 / EasyOCR vs Paddle predict/ocr 兜底 / [None] 短路 / 异常返回空 / picname 失败诊断日志。 新增 16 个单元测试覆盖: - 早出 4 种(pic_path 空/不存在/带 confidence/OCR 未初始化) - OCR 调用路径(EasyOCR / Paddle predict / Paddle ocr 兜底 / [None] 短路 / 异常) - 缓存命中跳过 OCR / 缓存未命中写回 - change_color 0/1/2 三种模式 - return_confidence True/False 返回类型 - autouse fixture 清 OCR 模块级缓存防止测试污染 1361 passed (+16)。 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
1 parent
b200002a00
commit
8457219241
3 files changed
+640
-369
No files matched your search
@@ -730,376 +730,11 @@ class underlying_operations:
|
||||
change_color: 颜色处理模式,0=保留彩色(推荐),1/2=二值化等
|
||||
pic_path: 可选的图片路径,未提供时通过 `_find_pic_path` 查找
|
||||
return_confidence: 是否返回置信度信息(默认False,保持向后兼容)
|
||||
返回:
|
||||
如果return_confidence=False: 识别到的文本(字符串),识别失败返回空字符串
|
||||
如果return_confidence=True: (text, confidence_info) 元组
|
||||
"""
|
||||
if not pic_path:
|
||||
try:
|
||||
pic_path = self._find_pic_path(picname)
|
||||
except Exception:
|
||||
pic_path = None
|
||||
|
||||
if not pic_path or not os.path.exists(pic_path):
|
||||
return ("", None) if return_confidence else ""
|
||||
|
||||
# ===== 缓存查询:基于图片字节哈希 =====
|
||||
# 同一张图重复 OCR 直接命中(业务里 cut_pic 可能短时间多次重读相同区域)
|
||||
cache_key = None
|
||||
try:
|
||||
from operations.ocr_operations import (
|
||||
_img_fingerprint, cache_get, cache_put,
|
||||
)
|
||||
with open(pic_path, "rb") as _f:
|
||||
_bytes = _f.read()
|
||||
cache_key = _img_fingerprint(_bytes, mode=int(change_color or 0))
|
||||
cached = cache_get(cache_key)
|
||||
if cached is not None:
|
||||
return cached if return_confidence else cached[0]
|
||||
except Exception:
|
||||
cache_key = None
|
||||
|
||||
# 优化:减少图像验证步骤,直接加载(节省50-100ms)
|
||||
img = None
|
||||
try:
|
||||
img = Image.open(pic_path)
|
||||
img.load()
|
||||
except (IOError, OSError, UnidentifiedImageError):
|
||||
return ("", None) if return_confidence else ""
|
||||
|
||||
if img is None:
|
||||
return ("", None) if return_confidence else ""
|
||||
|
||||
# 预处理:实测 PaddleOCR v4-mobile 内部已自带 resize/normalize,
|
||||
# 外层"放大+锐化+对比度增强"反而会引入伪影,让识别变差(tools/test_preprocess_impact.py 实测)。
|
||||
# 因此默认走原图。仅在 change_color=1/2 业务显式要求二值化时才做。
|
||||
if change_color in (1, 2):
|
||||
img = img.convert("L")
|
||||
thresh = 128 if change_color == 1 else 180
|
||||
img = img.point(lambda x, t=thresh: 0 if int(x) < t else 255)
|
||||
img = img.convert("RGB")
|
||||
else:
|
||||
img = img.convert("RGB")
|
||||
|
||||
w, h = img.size
|
||||
img1 = img # 不再做放大/锐化/对比度,PaddleOCR 自己处理
|
||||
|
||||
# 调试图像保存(默认关闭,业务调试时通过 SAVE_OCR_DEBUG_IMAGES=1 启用)
|
||||
try:
|
||||
save_debug = os.environ.get("SAVE_OCR_DEBUG_IMAGES", "0") == "1"
|
||||
if not save_debug:
|
||||
try:
|
||||
save_debug = bool(getattr(self, "save_ocr_debug_images", False))
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
if save_debug:
|
||||
from douyin.handlers import douyin_screenshot_manager
|
||||
from douyin.core import path_config
|
||||
|
||||
ocr_coords = None
|
||||
try:
|
||||
with self._ocr_coords_lock:
|
||||
ocr_coords = self._ocr_coords_cache.get(picname)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
full_screenshot = None
|
||||
try:
|
||||
screenshot_path = path_config.get_screenshot_path()
|
||||
if os.path.exists(screenshot_path):
|
||||
full_screenshot = Image.open(screenshot_path)
|
||||
full_screenshot.load()
|
||||
full_screenshot = full_screenshot.convert("RGB")
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
debug_image = full_screenshot if full_screenshot is not None else img1
|
||||
debug_path = douyin_screenshot_manager.save_debug_image(
|
||||
debug_image,
|
||||
operation_type="ocr",
|
||||
picname=picname,
|
||||
step="preprocessed",
|
||||
coords=ocr_coords,
|
||||
extra_info={
|
||||
"mode": change_color,
|
||||
"size": f"{w}x{h}",
|
||||
"source": pic_path if pic_path else picname,
|
||||
"has_full_screenshot": full_screenshot is not None,
|
||||
"has_coords": ocr_coords is not None,
|
||||
},
|
||||
)
|
||||
if debug_path:
|
||||
try:
|
||||
logging.info("[调试图像] OCR预处理保存: %s", debug_path)
|
||||
except Exception:
|
||||
pass
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
img_np = np.array(img1)
|
||||
|
||||
# 用于收集置信度信息
|
||||
ocr_confidence_info = None
|
||||
|
||||
# 如果 OCR 未初始化,安全降级
|
||||
try:
|
||||
ocr_obj = getattr(self, "ocr", None)
|
||||
ocr_type = getattr(self, "ocr_type", None)
|
||||
except Exception:
|
||||
ocr_obj = None
|
||||
ocr_type = None
|
||||
|
||||
if ocr_obj is None:
|
||||
try:
|
||||
logging.warning("OCR 未初始化,跳过识别:picname=%s", picname)
|
||||
except Exception:
|
||||
pass
|
||||
return ("", None) if return_confidence else ""
|
||||
|
||||
try:
|
||||
if ocr_type == "easyocr":
|
||||
result = ocr_obj.readtext(img_np)
|
||||
extract_result = self.extract_ocr_content(
|
||||
result, return_confidence=True
|
||||
)
|
||||
if isinstance(extract_result, tuple):
|
||||
text, ocr_confidence_info = extract_result
|
||||
else:
|
||||
text = extract_result
|
||||
else:
|
||||
# PaddleOCR 3.x:predict() 是新推荐 API;ocr() 仍兼容
|
||||
_predict = getattr(ocr_obj, "predict", None) or ocr_obj.ocr
|
||||
result = _predict(img_np)
|
||||
if result == [None]:
|
||||
if cache_key:
|
||||
try:
|
||||
cache_put(cache_key, ("", None))
|
||||
except Exception:
|
||||
pass
|
||||
return ("", None) if return_confidence else ""
|
||||
extract_result = self.extract_ocr_content(
|
||||
result, return_confidence=True
|
||||
)
|
||||
if isinstance(extract_result, tuple):
|
||||
text, ocr_confidence_info = extract_result
|
||||
else:
|
||||
text = extract_result
|
||||
except Exception as e:
|
||||
try:
|
||||
logging.warning("OCR 调用失败,返回空结果:%s", e)
|
||||
except Exception:
|
||||
pass
|
||||
return ("", None) if return_confidence else ""
|
||||
|
||||
# 写入缓存
|
||||
if cache_key:
|
||||
try:
|
||||
cache_put(cache_key, (text, ocr_confidence_info))
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# Initialize variables to avoid unbound variable issues
|
||||
img1 = img # img1 is used later, initialize from img
|
||||
img2 = None
|
||||
text2 = None
|
||||
# 禁用二次OCR识别(用户要求:没有GPU,不需要二次OCR)
|
||||
# 如果第一次识别失败,直接返回空结果,避免耗时3-4秒的二次识别
|
||||
# if (not text) and max(w, h) <= 1200:
|
||||
# try:
|
||||
# img2 = img.resize((w * 2, h * 2), _RESAMPLE.LANCZOS)
|
||||
# try:
|
||||
# from PIL import ImageEnhance
|
||||
#
|
||||
# img2_sharpened = img2.filter(
|
||||
# ImageFilter.UnsharpMask(radius=1.5, percent=180, threshold=3)
|
||||
# )
|
||||
# enhancer = ImageEnhance.Contrast(img2_sharpened)
|
||||
# img2 = enhancer.enhance(1.2)
|
||||
# except Exception:
|
||||
# img2 = img2.filter(ImageFilter.SHARPEN)
|
||||
#
|
||||
# if getattr(self, "ocr_type", None) == "easyocr":
|
||||
# result2 = self.ocr.readtext(np.array(img2))
|
||||
# extract_result2 = self.extract_ocr_content(result2, return_confidence=True)
|
||||
# if isinstance(extract_result2, tuple):
|
||||
# text2, ocr_confidence_info = extract_result2
|
||||
# else:
|
||||
# text2 = extract_result2
|
||||
# if text2:
|
||||
# text = text2
|
||||
# result = result2
|
||||
# else:
|
||||
# result2 = self.ocr.ocr(np.array(img2))
|
||||
# if result2 != [None]:
|
||||
# extract_result2 = self.extract_ocr_content(result2, return_confidence=True)
|
||||
# if isinstance(extract_result2, tuple):
|
||||
# text2, ocr_confidence_info = extract_result2
|
||||
# else:
|
||||
# text2 = extract_result2
|
||||
# if text2:
|
||||
# text = text2
|
||||
# result = result2
|
||||
# except Exception:
|
||||
# pass
|
||||
|
||||
if picname in ("福袋内容", "福袋倒计时") and not text:
|
||||
try:
|
||||
if isinstance(result, list) and result and isinstance(result[0], dict):
|
||||
_uo_log(
|
||||
f"OCR调试({picname}): result[0] keys={list(result[0].keys())}"
|
||||
)
|
||||
else:
|
||||
_uo_log(f"OCR调试({picname}): result type={type(result)}")
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
try:
|
||||
from douyin.handlers import douyin_screenshot_manager
|
||||
|
||||
# Ensure img1 is defined before use
|
||||
if "img1" not in locals():
|
||||
img1 = img
|
||||
img_for_attempt = (
|
||||
img2 if (img2 is not None and text2 and text2.strip()) else img1
|
||||
)
|
||||
douyin_screenshot_manager.save_latest_attempt(
|
||||
img_for_attempt, operation_type="ocr"
|
||||
)
|
||||
|
||||
# 保存OCR结果后的调试图像(带OCR文本结果和置信度)- 使用完整截图
|
||||
ocr_result_preview = (
|
||||
(text or text2 or "").strip()[:50] if (text or text2) else ""
|
||||
)
|
||||
extra_info_dict = {
|
||||
"mode": change_color,
|
||||
"text_len": len(text or text2 or ""),
|
||||
"text_preview": ocr_result_preview,
|
||||
"used_img2": bool(img2 is not None and text2 and text2.strip()),
|
||||
}
|
||||
# 添加置信度信息
|
||||
if ocr_confidence_info:
|
||||
extra_info_dict["avg_confidence"] = round(
|
||||
ocr_confidence_info.get("avg_confidence", 0.0), 4
|
||||
)
|
||||
extra_info_dict["min_confidence"] = round(
|
||||
ocr_confidence_info.get("min_confidence", 0.0), 4
|
||||
)
|
||||
extra_info_dict["max_confidence"] = round(
|
||||
ocr_confidence_info.get("max_confidence", 0.0), 4
|
||||
)
|
||||
extra_info_dict["confidence_items"] = ocr_confidence_info.get(
|
||||
"item_count", 0
|
||||
)
|
||||
|
||||
# 尝试获取OCR区域坐标
|
||||
ocr_coords = None
|
||||
try:
|
||||
ocr_coords = self._ocr_coords_cache.get(picname)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# 加载完整截图
|
||||
full_screenshot = None
|
||||
try:
|
||||
from douyin.core import path_config
|
||||
|
||||
screenshot_path = path_config.get_screenshot_path()
|
||||
if os.path.exists(screenshot_path):
|
||||
full_screenshot = Image.open(screenshot_path)
|
||||
full_screenshot.load()
|
||||
full_screenshot = full_screenshot.convert("RGB")
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# 如果成功加载完整截图,使用完整截图;否则使用预处理图像
|
||||
debug_image = (
|
||||
full_screenshot if full_screenshot is not None else img_for_attempt
|
||||
)
|
||||
extra_info_dict["has_full_screenshot"] = full_screenshot is not None
|
||||
extra_info_dict["has_coords"] = ocr_coords is not None
|
||||
|
||||
debug_path = douyin_screenshot_manager.save_debug_image(
|
||||
debug_image,
|
||||
operation_type="ocr",
|
||||
picname=picname,
|
||||
step="after_ocr",
|
||||
coords=ocr_coords,
|
||||
extra_info=extra_info_dict,
|
||||
)
|
||||
|
||||
# 记录OCR置信度到日志
|
||||
if ocr_confidence_info and ocr_confidence_info.get("item_count", 0) > 0:
|
||||
try:
|
||||
logging.info(
|
||||
"[OCR置信度] %s: 平均=%.4f, 最小=%.4f, 最大=%.4f, 项目数=%d",
|
||||
picname,
|
||||
ocr_confidence_info.get("avg_confidence", 0.0),
|
||||
ocr_confidence_info.get("min_confidence", 0.0),
|
||||
ocr_confidence_info.get("max_confidence", 0.0),
|
||||
int(ocr_confidence_info.get("item_count", 0) or 0),
|
||||
)
|
||||
except Exception:
|
||||
pass
|
||||
if debug_path:
|
||||
try:
|
||||
logging.info(
|
||||
"[调试图像] OCR结果保存: %s, 文本: %s",
|
||||
debug_path,
|
||||
ocr_result_preview,
|
||||
)
|
||||
except Exception:
|
||||
pass
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
if text and text.strip() and picname:
|
||||
try:
|
||||
try:
|
||||
from douyin.handlers import douyin_screenshot_manager
|
||||
|
||||
category = douyin_screenshot_manager.get_category_for_picname(
|
||||
picname
|
||||
)
|
||||
except Exception:
|
||||
category = "others"
|
||||
|
||||
from douyin.core import path_config
|
||||
pic_base = path_config.get_pic_dir()
|
||||
category_dir = os.path.join(pic_base, "ocr", category)
|
||||
os.makedirs(category_dir, exist_ok=True)
|
||||
|
||||
success_path = os.path.join(category_dir, f"{picname}_success.png")
|
||||
try:
|
||||
# Ensure img1 is defined before use
|
||||
if "img1" not in locals():
|
||||
img1 = img
|
||||
if img2 is not None and text2 and text2.strip():
|
||||
img_to_save = img2
|
||||
else:
|
||||
img_to_save = img1
|
||||
try:
|
||||
from douyin.handlers import douyin_screenshot_manager
|
||||
|
||||
douyin_screenshot_manager.save_last_success(
|
||||
img_to_save, operation_type="ocr"
|
||||
)
|
||||
except Exception:
|
||||
pass
|
||||
img_to_save.save(success_path)
|
||||
except Exception:
|
||||
try:
|
||||
if img is not None:
|
||||
img.save(success_path)
|
||||
except Exception:
|
||||
pass
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
if return_confidence:
|
||||
return text, ocr_confidence_info
|
||||
else:
|
||||
return text
|
||||
from douyin.core.underlying_operations_pkg.ocr_pipeline import (
|
||||
analyse_pic_word as _impl,
|
||||
)
|
||||
return _impl(self, picname, change_color, pic_path, return_confidence)
|
||||
|
||||
def analyse_pic_digits(self, picname, change_color=1, pic_path=None):
|
||||
return _impl_analyse_pic_digits(self, picname, change_color, pic_path)
|
||||
|
||||
@@ -0,0 +1,367 @@
|
||||
"""douyin/core/underlying_operations_pkg/ocr_pipeline - analyse_pic_word OCR 管道
|
||||
|
||||
把 underlying_operations.analyse_pic_word 整段抽出来。流程:
|
||||
1. 路径解析 + 缓存查询(基于图片字节哈希)
|
||||
2. 加载 + 验证图片完整性
|
||||
3. change_color 预处理(0=保留彩色,1/2=二值化)
|
||||
4. 调试图像保存(preprocessed 步骤,仅当 SAVE_OCR_DEBUG_IMAGES=1)
|
||||
5. 调用 OCR(EasyOCR / PaddleOCR predict/ocr)
|
||||
6. 提取文本 + 置信度
|
||||
7. 写缓存
|
||||
8. 调试图像保存(after_ocr 步骤)
|
||||
9. 成功识别保存到 ocr/<category>/<picname>_success.png
|
||||
|
||||
入口:analyse_pic_word(self, picname, change_color, pic_path, return_confidence)
|
||||
self 必须提供:_find_pic_path / extract_ocr_content / ocr / ocr_type /
|
||||
_ocr_coords_lock / _ocr_coords_cache / save_ocr_debug_images
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import logging
|
||||
import os
|
||||
from typing import Any, Optional
|
||||
|
||||
from PIL import Image, UnidentifiedImageError
|
||||
import numpy as np
|
||||
|
||||
|
||||
def analyse_pic_word(
|
||||
self: Any,
|
||||
picname: str,
|
||||
change_color: int = 0,
|
||||
pic_path: Optional[str] = None,
|
||||
return_confidence: bool = False,
|
||||
):
|
||||
"""对指定图片运行 OCR 并返回拼接后的文本结果。
|
||||
|
||||
Args:
|
||||
self: underlying_operations 实例
|
||||
picname: 图片名(如"福袋内容")
|
||||
change_color: 0=保留彩色,1/2=二值化
|
||||
pic_path: 可选;未提供时通过 self._find_pic_path 查找
|
||||
return_confidence: True 时返回 (text, confidence_info);否则只返回 text
|
||||
|
||||
Returns:
|
||||
return_confidence=False: text(str)
|
||||
return_confidence=True: (text, confidence_info)
|
||||
"""
|
||||
from douyin.core.underlying_operations import _uo_log
|
||||
|
||||
if not pic_path:
|
||||
try:
|
||||
pic_path = self._find_pic_path(picname)
|
||||
except Exception:
|
||||
pic_path = None
|
||||
|
||||
if not pic_path or not os.path.exists(pic_path):
|
||||
return ("", None) if return_confidence else ""
|
||||
|
||||
# ===== 缓存查询:基于图片字节哈希 =====
|
||||
cache_key = None
|
||||
try:
|
||||
from operations.ocr_operations import (
|
||||
_img_fingerprint, cache_get, cache_put,
|
||||
)
|
||||
with open(pic_path, "rb") as _f:
|
||||
_bytes = _f.read()
|
||||
cache_key = _img_fingerprint(_bytes, mode=int(change_color or 0))
|
||||
cached = cache_get(cache_key)
|
||||
if cached is not None:
|
||||
return cached if return_confidence else cached[0]
|
||||
except Exception:
|
||||
cache_key = None
|
||||
|
||||
# 加载图片
|
||||
img = None
|
||||
try:
|
||||
img = Image.open(pic_path)
|
||||
img.load()
|
||||
except (IOError, OSError, UnidentifiedImageError):
|
||||
return ("", None) if return_confidence else ""
|
||||
|
||||
if img is None:
|
||||
return ("", None) if return_confidence else ""
|
||||
|
||||
# 预处理:实测 PaddleOCR v4-mobile 内部已自带 resize/normalize,
|
||||
# 外层"放大+锐化+对比度增强"反而会引入伪影,让识别变差。
|
||||
# 因此默认走原图。仅在 change_color=1/2 业务显式要求二值化时才做。
|
||||
if change_color in (1, 2):
|
||||
img = img.convert("L")
|
||||
thresh = 128 if change_color == 1 else 180
|
||||
img = img.point(lambda x, t=thresh: 0 if int(x) < t else 255)
|
||||
img = img.convert("RGB")
|
||||
else:
|
||||
img = img.convert("RGB")
|
||||
|
||||
w, h = img.size
|
||||
img1 = img # 不再做放大/锐化/对比度
|
||||
|
||||
# 调试图像保存(preprocessed 步骤)
|
||||
_save_debug_preprocessed(self, img1, picname, pic_path, change_color, w, h)
|
||||
|
||||
img_np = np.array(img1)
|
||||
ocr_confidence_info = None
|
||||
|
||||
try:
|
||||
ocr_obj = getattr(self, "ocr", None)
|
||||
ocr_type = getattr(self, "ocr_type", None)
|
||||
except Exception:
|
||||
ocr_obj = None
|
||||
ocr_type = None
|
||||
|
||||
if ocr_obj is None:
|
||||
try:
|
||||
logging.warning("OCR 未初始化,跳过识别:picname=%s", picname)
|
||||
except Exception:
|
||||
pass
|
||||
return ("", None) if return_confidence else ""
|
||||
|
||||
# 执行 OCR
|
||||
try:
|
||||
if ocr_type == "easyocr":
|
||||
result = ocr_obj.readtext(img_np)
|
||||
extract_result = self.extract_ocr_content(result, return_confidence=True)
|
||||
if isinstance(extract_result, tuple):
|
||||
text, ocr_confidence_info = extract_result
|
||||
else:
|
||||
text = extract_result
|
||||
else:
|
||||
# PaddleOCR 3.x:predict() 是新推荐 API;ocr() 仍兼容
|
||||
_predict = getattr(ocr_obj, "predict", None) or ocr_obj.ocr
|
||||
result = _predict(img_np)
|
||||
if result == [None]:
|
||||
if cache_key:
|
||||
try:
|
||||
cache_put(cache_key, ("", None))
|
||||
except Exception:
|
||||
pass
|
||||
return ("", None) if return_confidence else ""
|
||||
extract_result = self.extract_ocr_content(result, return_confidence=True)
|
||||
if isinstance(extract_result, tuple):
|
||||
text, ocr_confidence_info = extract_result
|
||||
else:
|
||||
text = extract_result
|
||||
except Exception as e:
|
||||
try:
|
||||
logging.warning("OCR 调用失败,返回空结果:%s", e)
|
||||
except Exception:
|
||||
pass
|
||||
return ("", None) if return_confidence else ""
|
||||
|
||||
# 写缓存
|
||||
if cache_key:
|
||||
try:
|
||||
cache_put(cache_key, (text, ocr_confidence_info))
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# 调试日志:福袋内容 / 福袋倒计时 OCR 失败时打印 result 结构
|
||||
if picname in ("福袋内容", "福袋倒计时") and not text:
|
||||
try:
|
||||
if isinstance(result, list) and result and isinstance(result[0], dict):
|
||||
_uo_log(f"OCR调试({picname}): result[0] keys={list(result[0].keys())}")
|
||||
else:
|
||||
_uo_log(f"OCR调试({picname}): result type={type(result)}")
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# 调试图像保存(after_ocr 步骤)+ 置信度日志
|
||||
_save_debug_after_ocr(
|
||||
self, img1, picname, change_color, text,
|
||||
ocr_confidence_info,
|
||||
)
|
||||
|
||||
# 成功识别 → 保存到 ocr/<category>/<picname>_success.png
|
||||
if text and text.strip() and picname:
|
||||
_save_success_pic(img, img1, picname, text)
|
||||
|
||||
if return_confidence:
|
||||
return text, ocr_confidence_info
|
||||
else:
|
||||
return text
|
||||
|
||||
|
||||
def _save_debug_preprocessed(
|
||||
self: Any, img1, picname: str, pic_path: str,
|
||||
change_color: int, w: int, h: int,
|
||||
) -> None:
|
||||
"""保存预处理后的调试图像(OCR 输入),仅当 SAVE_OCR_DEBUG_IMAGES=1 时执行。"""
|
||||
try:
|
||||
save_debug = os.environ.get("SAVE_OCR_DEBUG_IMAGES", "0") == "1"
|
||||
if not save_debug:
|
||||
try:
|
||||
save_debug = bool(getattr(self, "save_ocr_debug_images", False))
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
if not save_debug:
|
||||
return
|
||||
|
||||
from douyin.handlers import douyin_screenshot_manager
|
||||
from douyin.core import path_config
|
||||
|
||||
ocr_coords = None
|
||||
try:
|
||||
with self._ocr_coords_lock:
|
||||
ocr_coords = self._ocr_coords_cache.get(picname)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
full_screenshot = None
|
||||
try:
|
||||
screenshot_path = path_config.get_screenshot_path()
|
||||
if os.path.exists(screenshot_path):
|
||||
full_screenshot = Image.open(screenshot_path)
|
||||
full_screenshot.load()
|
||||
full_screenshot = full_screenshot.convert("RGB")
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
debug_image = full_screenshot if full_screenshot is not None else img1
|
||||
debug_path = douyin_screenshot_manager.save_debug_image(
|
||||
debug_image,
|
||||
operation_type="ocr",
|
||||
picname=picname,
|
||||
step="preprocessed",
|
||||
coords=ocr_coords,
|
||||
extra_info={
|
||||
"mode": change_color,
|
||||
"size": f"{w}x{h}",
|
||||
"source": pic_path if pic_path else picname,
|
||||
"has_full_screenshot": full_screenshot is not None,
|
||||
"has_coords": ocr_coords is not None,
|
||||
},
|
||||
)
|
||||
if debug_path:
|
||||
try:
|
||||
logging.info("[调试图像] OCR预处理保存: %s", debug_path)
|
||||
except Exception:
|
||||
pass
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
|
||||
def _save_debug_after_ocr(
|
||||
self: Any, img1, picname: str, change_color: int,
|
||||
text: Optional[str], ocr_confidence_info: Optional[dict],
|
||||
) -> None:
|
||||
"""保存 OCR 完成后的调试图像(含识别文本预览 + 置信度)。"""
|
||||
try:
|
||||
from douyin.handlers import douyin_screenshot_manager
|
||||
|
||||
douyin_screenshot_manager.save_latest_attempt(img1, operation_type="ocr")
|
||||
|
||||
ocr_result_preview = (text or "").strip()[:50] if text else ""
|
||||
extra_info_dict = {
|
||||
"mode": change_color,
|
||||
"text_len": len(text or ""),
|
||||
"text_preview": ocr_result_preview,
|
||||
"used_img2": False, # 二次 OCR 已禁用
|
||||
}
|
||||
if ocr_confidence_info:
|
||||
extra_info_dict["avg_confidence"] = round(
|
||||
ocr_confidence_info.get("avg_confidence", 0.0), 4
|
||||
)
|
||||
extra_info_dict["min_confidence"] = round(
|
||||
ocr_confidence_info.get("min_confidence", 0.0), 4
|
||||
)
|
||||
extra_info_dict["max_confidence"] = round(
|
||||
ocr_confidence_info.get("max_confidence", 0.0), 4
|
||||
)
|
||||
extra_info_dict["confidence_items"] = ocr_confidence_info.get(
|
||||
"item_count", 0
|
||||
)
|
||||
|
||||
ocr_coords = None
|
||||
try:
|
||||
ocr_coords = self._ocr_coords_cache.get(picname)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
full_screenshot = None
|
||||
try:
|
||||
from douyin.core import path_config
|
||||
|
||||
screenshot_path = path_config.get_screenshot_path()
|
||||
if os.path.exists(screenshot_path):
|
||||
full_screenshot = Image.open(screenshot_path)
|
||||
full_screenshot.load()
|
||||
full_screenshot = full_screenshot.convert("RGB")
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
debug_image = full_screenshot if full_screenshot is not None else img1
|
||||
extra_info_dict["has_full_screenshot"] = full_screenshot is not None
|
||||
extra_info_dict["has_coords"] = ocr_coords is not None
|
||||
|
||||
debug_path = douyin_screenshot_manager.save_debug_image(
|
||||
debug_image,
|
||||
operation_type="ocr",
|
||||
picname=picname,
|
||||
step="after_ocr",
|
||||
coords=ocr_coords,
|
||||
extra_info=extra_info_dict,
|
||||
)
|
||||
|
||||
# 记录 OCR 置信度到日志
|
||||
if ocr_confidence_info and ocr_confidence_info.get("item_count", 0) > 0:
|
||||
try:
|
||||
logging.info(
|
||||
"[OCR置信度] %s: 平均=%.4f, 最小=%.4f, 最大=%.4f, 项目数=%d",
|
||||
picname,
|
||||
ocr_confidence_info.get("avg_confidence", 0.0),
|
||||
ocr_confidence_info.get("min_confidence", 0.0),
|
||||
ocr_confidence_info.get("max_confidence", 0.0),
|
||||
int(ocr_confidence_info.get("item_count", 0) or 0),
|
||||
)
|
||||
except Exception:
|
||||
pass
|
||||
if debug_path:
|
||||
try:
|
||||
logging.info(
|
||||
"[调试图像] OCR结果保存: %s, 文本: %s",
|
||||
debug_path,
|
||||
ocr_result_preview,
|
||||
)
|
||||
except Exception:
|
||||
pass
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
|
||||
def _save_success_pic(img, img1, picname: str, text: str) -> None:
|
||||
"""成功识别后保存到 ocr/<category>/<picname>_success.png。"""
|
||||
try:
|
||||
try:
|
||||
from douyin.handlers import douyin_screenshot_manager
|
||||
|
||||
category = douyin_screenshot_manager.get_category_for_picname(picname)
|
||||
except Exception:
|
||||
category = "others"
|
||||
|
||||
from douyin.core import path_config
|
||||
|
||||
pic_base = path_config.get_pic_dir()
|
||||
category_dir = os.path.join(pic_base, "ocr", category)
|
||||
os.makedirs(category_dir, exist_ok=True)
|
||||
|
||||
success_path = os.path.join(category_dir, f"{picname}_success.png")
|
||||
try:
|
||||
img_to_save = img1
|
||||
try:
|
||||
from douyin.handlers import douyin_screenshot_manager
|
||||
|
||||
douyin_screenshot_manager.save_last_success(
|
||||
img_to_save, operation_type="ocr"
|
||||
)
|
||||
except Exception:
|
||||
pass
|
||||
img_to_save.save(success_path)
|
||||
except Exception:
|
||||
try:
|
||||
if img is not None:
|
||||
img.save(success_path)
|
||||
except Exception:
|
||||
pass
|
||||
except Exception:
|
||||
pass
|
||||
@@ -0,0 +1,269 @@
|
||||
"""douyin/core/underlying_operations_pkg/ocr_pipeline 单元测试。
|
||||
|
||||
analyse_pic_word 主流程的早出和分支测试。OCR 实际调用通过 mock 完成。
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import os
|
||||
from unittest.mock import MagicMock, patch
|
||||
|
||||
import pytest
|
||||
|
||||
try:
|
||||
from PIL import Image
|
||||
except ImportError: # pragma: no cover
|
||||
pytest.skip("PIL unavailable", allow_module_level=True)
|
||||
|
||||
from douyin.core.underlying_operations_pkg.ocr_pipeline import analyse_pic_word
|
||||
|
||||
|
||||
@pytest.fixture(autouse=True)
|
||||
def _clear_ocr_cache():
|
||||
"""每个测试前清 operations.ocr_operations 模块级 OCR 缓存,避免互相污染。"""
|
||||
try:
|
||||
from operations import ocr_operations
|
||||
with ocr_operations._ocr_cache_lock:
|
||||
ocr_operations._ocr_cache.clear()
|
||||
except Exception:
|
||||
pass
|
||||
yield
|
||||
try:
|
||||
from operations import ocr_operations
|
||||
with ocr_operations._ocr_cache_lock:
|
||||
ocr_operations._ocr_cache.clear()
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
|
||||
def _mk_self(*, ocr=None, ocr_type=None, save_debug=False):
|
||||
s = MagicMock()
|
||||
s.ocr = ocr
|
||||
s.ocr_type = ocr_type
|
||||
s.save_ocr_debug_images = save_debug
|
||||
s._find_pic_path = MagicMock(return_value="")
|
||||
s._ocr_coords_lock = MagicMock()
|
||||
s._ocr_coords_lock.__enter__ = MagicMock(return_value=None)
|
||||
s._ocr_coords_lock.__exit__ = MagicMock(return_value=False)
|
||||
s._ocr_coords_cache = {}
|
||||
# extract_ocr_content 默认返回 ("", None)
|
||||
s.extract_ocr_content = MagicMock(return_value=("", None))
|
||||
return s
|
||||
|
||||
|
||||
def _make_real_pic(tmp_path):
|
||||
"""生成一张真实 PNG 文件作为 OCR 输入,避免 mock PIL.Image 引入复杂度。"""
|
||||
p = tmp_path / "tmp.png"
|
||||
Image.new("RGB", (50, 50), (200, 200, 200)).save(str(p))
|
||||
return str(p)
|
||||
|
||||
|
||||
# ============================================================
|
||||
# 早出条件
|
||||
# ============================================================
|
||||
|
||||
class TestEarlyReturns:
|
||||
def test_pic_path_empty_returns_empty(self):
|
||||
s = _mk_self()
|
||||
s._find_pic_path = MagicMock(return_value="")
|
||||
result = analyse_pic_word(s, "X", change_color=0)
|
||||
assert result == ""
|
||||
|
||||
def test_pic_path_nonexistent_returns_empty(self):
|
||||
s = _mk_self()
|
||||
s._find_pic_path = MagicMock(return_value="/nonexistent/path.png")
|
||||
result = analyse_pic_word(s, "X", change_color=0)
|
||||
assert result == ""
|
||||
|
||||
def test_pic_path_nonexistent_with_confidence(self):
|
||||
s = _mk_self()
|
||||
s._find_pic_path = MagicMock(return_value="/nonexistent/path.png")
|
||||
text, conf = analyse_pic_word(
|
||||
s, "X", change_color=0, return_confidence=True
|
||||
)
|
||||
assert text == ""
|
||||
assert conf is None
|
||||
|
||||
def test_ocr_not_initialized_returns_empty(self, tmp_path):
|
||||
s = _mk_self(ocr=None)
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
result = analyse_pic_word(s, "X", change_color=0, pic_path=pic_path)
|
||||
assert result == ""
|
||||
|
||||
|
||||
# ============================================================
|
||||
# OCR 调用路径
|
||||
# ============================================================
|
||||
|
||||
class TestOcrCallPaths:
|
||||
def test_easyocr_path(self, tmp_path):
|
||||
ocr = MagicMock()
|
||||
ocr.readtext = MagicMock(return_value=[("bbox", "识别出的文本", 0.95)])
|
||||
s = _mk_self(ocr=ocr, ocr_type="easyocr")
|
||||
s.extract_ocr_content = MagicMock(
|
||||
return_value=("识别出的文本", {"avg_confidence": 0.95, "item_count": 1})
|
||||
)
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
result = analyse_pic_word(s, "X", change_color=0, pic_path=pic_path)
|
||||
assert result == "识别出的文本"
|
||||
ocr.readtext.assert_called_once()
|
||||
|
||||
def test_paddleocr_predict_path(self, tmp_path):
|
||||
ocr = MagicMock()
|
||||
ocr.predict = MagicMock(return_value=[{"some": "result"}])
|
||||
s = _mk_self(ocr=ocr, ocr_type="paddleocr")
|
||||
s.extract_ocr_content = MagicMock(return_value=("paddle text", None))
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
result = analyse_pic_word(s, "X", change_color=0, pic_path=pic_path)
|
||||
assert result == "paddle text"
|
||||
ocr.predict.assert_called_once()
|
||||
|
||||
def test_paddleocr_ocr_fallback(self, tmp_path):
|
||||
"""没有 predict 方法时回退到 ocr() 调用。"""
|
||||
ocr = MagicMock(spec=["ocr"])
|
||||
ocr.ocr = MagicMock(return_value=[{"some": "result"}])
|
||||
s = _mk_self(ocr=ocr, ocr_type="paddleocr")
|
||||
s.extract_ocr_content = MagicMock(return_value=("ocr text", None))
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
result = analyse_pic_word(s, "X", change_color=0, pic_path=pic_path)
|
||||
assert result == "ocr text"
|
||||
ocr.ocr.assert_called_once()
|
||||
|
||||
def test_paddleocr_returns_none_short_circuits(self, tmp_path):
|
||||
"""PaddleOCR 返回 [None] 时直接返回空(不调用 extract_ocr_content)。"""
|
||||
ocr = MagicMock()
|
||||
ocr.predict = MagicMock(return_value=[None])
|
||||
s = _mk_self(ocr=ocr, ocr_type="paddleocr")
|
||||
s.extract_ocr_content = MagicMock(return_value=("不应被调用", None))
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
result = analyse_pic_word(s, "X", change_color=0, pic_path=pic_path)
|
||||
assert result == ""
|
||||
s.extract_ocr_content.assert_not_called()
|
||||
|
||||
def test_ocr_throws_returns_empty(self, tmp_path):
|
||||
"""OCR 调用抛异常时安全返回空。"""
|
||||
ocr = MagicMock()
|
||||
ocr.predict = MagicMock(side_effect=RuntimeError("OCR fail"))
|
||||
s = _mk_self(ocr=ocr, ocr_type="paddleocr")
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
result = analyse_pic_word(s, "X", change_color=0, pic_path=pic_path)
|
||||
assert result == ""
|
||||
|
||||
|
||||
# ============================================================
|
||||
# 缓存路径
|
||||
# ============================================================
|
||||
|
||||
class TestCachePath:
|
||||
def test_cache_hit_skips_ocr(self, tmp_path):
|
||||
"""缓存命中时不应调用 OCR。"""
|
||||
ocr = MagicMock()
|
||||
ocr.predict = MagicMock(return_value=[{"X": 1}])
|
||||
s = _mk_self(ocr=ocr, ocr_type="paddleocr")
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
with patch(
|
||||
"operations.ocr_operations._img_fingerprint",
|
||||
return_value="cache_key_X",
|
||||
), patch(
|
||||
"operations.ocr_operations.cache_get",
|
||||
return_value=("缓存的文本", {"avg_confidence": 0.99}),
|
||||
):
|
||||
result = analyse_pic_word(s, "X", change_color=0, pic_path=pic_path)
|
||||
assert result == "缓存的文本"
|
||||
# OCR 不应被调用
|
||||
ocr.predict.assert_not_called()
|
||||
|
||||
def test_cache_miss_runs_ocr_and_writes_back(self, tmp_path):
|
||||
ocr = MagicMock()
|
||||
ocr.predict = MagicMock(return_value=[{"X": 1}])
|
||||
s = _mk_self(ocr=ocr, ocr_type="paddleocr")
|
||||
s.extract_ocr_content = MagicMock(return_value=("新识别的文本", None))
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
with patch(
|
||||
"operations.ocr_operations._img_fingerprint", return_value="key1"
|
||||
), patch(
|
||||
"operations.ocr_operations.cache_get", return_value=None
|
||||
), patch(
|
||||
"operations.ocr_operations.cache_put"
|
||||
) as mock_put:
|
||||
result = analyse_pic_word(s, "X", change_color=0, pic_path=pic_path)
|
||||
assert result == "新识别的文本"
|
||||
# OCR 被调用 + 结果写回缓存
|
||||
ocr.predict.assert_called_once()
|
||||
mock_put.assert_called_once_with("key1", ("新识别的文本", None))
|
||||
|
||||
|
||||
# ============================================================
|
||||
# 颜色预处理分支
|
||||
# ============================================================
|
||||
|
||||
class TestChangeColor:
|
||||
def test_mode_0_keeps_color(self, tmp_path):
|
||||
ocr = MagicMock()
|
||||
ocr.predict = MagicMock(return_value=[{"X": 1}])
|
||||
s = _mk_self(ocr=ocr, ocr_type="paddleocr")
|
||||
s.extract_ocr_content = MagicMock(return_value=("text", None))
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
# 不抛即可
|
||||
result = analyse_pic_word(s, "X", change_color=0, pic_path=pic_path)
|
||||
assert isinstance(result, str)
|
||||
|
||||
def test_mode_1_binarizes(self, tmp_path):
|
||||
ocr = MagicMock()
|
||||
ocr.predict = MagicMock(return_value=[{"X": 1}])
|
||||
s = _mk_self(ocr=ocr, ocr_type="paddleocr")
|
||||
s.extract_ocr_content = MagicMock(return_value=("binarized text", None))
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
# change_color=1 走二值化分支
|
||||
result = analyse_pic_word(s, "X", change_color=1, pic_path=pic_path)
|
||||
assert isinstance(result, str)
|
||||
|
||||
def test_mode_2_binarizes_high_threshold(self, tmp_path):
|
||||
ocr = MagicMock()
|
||||
ocr.predict = MagicMock(return_value=[{"X": 1}])
|
||||
s = _mk_self(ocr=ocr, ocr_type="paddleocr")
|
||||
s.extract_ocr_content = MagicMock(return_value=("text", None))
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
result = analyse_pic_word(s, "X", change_color=2, pic_path=pic_path)
|
||||
assert isinstance(result, str)
|
||||
|
||||
|
||||
# ============================================================
|
||||
# return_confidence
|
||||
# ============================================================
|
||||
|
||||
class TestReturnConfidence:
|
||||
def test_returns_text_only_by_default(self, tmp_path):
|
||||
ocr = MagicMock()
|
||||
ocr.predict = MagicMock(return_value=[{"X": 1}])
|
||||
s = _mk_self(ocr=ocr, ocr_type="paddleocr")
|
||||
s.extract_ocr_content = MagicMock(return_value=("text", {"item_count": 1}))
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
result = analyse_pic_word(s, "X", change_color=0, pic_path=pic_path)
|
||||
# 默认只返回字符串
|
||||
assert isinstance(result, str)
|
||||
assert result == "text"
|
||||
|
||||
def test_returns_tuple_with_flag(self, tmp_path):
|
||||
ocr = MagicMock()
|
||||
ocr.predict = MagicMock(return_value=[{"X": 1}])
|
||||
conf = {"avg_confidence": 0.9, "item_count": 1}
|
||||
s = _mk_self(ocr=ocr, ocr_type="paddleocr")
|
||||
s.extract_ocr_content = MagicMock(return_value=("text", conf))
|
||||
pic_path = _make_real_pic(tmp_path)
|
||||
|
||||
text, ret_conf = analyse_pic_word(
|
||||
s, "X", change_color=0, pic_path=pic_path, return_confidence=True
|
||||
)
|
||||
assert text == "text"
|
||||
assert ret_conf == conf
|
||||
Reference in new issue
Block a user