Improve OCR speed and preprocessing

VRWT_Tool/VRC_OSC/src/ocr/ocr_actions.py

@@ -1,12 +1,6 @@
 import os
 from datetime import datetime
 
-import pyperclip
-from paddleocr import PaddleOCR
-
-from common.runtime_log import appendRuntimeLog
-from screen.screen_actions import captureOcrRegion
-
 os.environ["FLAGS_enable_pir_api"] = "0"
 os.environ["FLAGS_use_onednn"] = "0"
 os.environ["FLAGS_use_mkldnn"] = "0"
@@ -14,6 +8,17 @@ os.environ["FLAGS_use_onednn_bfloat16"] = "0"
 os.environ["MKLDNN_DISABLE_WORKSPACE"] = "1"
 os.environ["ONEDNN_VERBOSE"] = "0"
 
+import pyperclip
+from paddleocr import PaddleOCR
+from PIL import Image
+from PIL import ImageFilter
+from PIL import ImageOps
+
+from common.env_config import getEnvBool
+from common.env_config import getEnvFloat
+from common.runtime_log import appendRuntimeLog
+from screen.screen_actions import captureOcrRegion
+
 _ocr_engine = None
 
 
@@ -27,10 +32,17 @@ def getOcrEngine():
 
     if _ocr_engine is None:
         log("INFO", "PaddleOCR initializing")
-        _ocr_engine = PaddleOCR(
-            lang="japan",
-            use_angle_cls=True,
-        )
+        options = {
+            "lang": "japan",
+            "use_angle_cls": getEnvBool("OCR_USE_ANGLE_CLS", False),
+            "show_log": False,
+        }
+
+        try:
+            _ocr_engine = PaddleOCR(**options)
+        except (TypeError, ValueError):
+            options.pop("show_log", None)
+            _ocr_engine = PaddleOCR(**options)
 
     return _ocr_engine
 
@@ -99,6 +111,37 @@ def printOcrResult(text):
         print(line, flush=True)
 
 
+def preprocessOcrImage(image_path):
+    enabled = getEnvBool("OCR_PREPROCESS", True)
+    if not enabled:
+        return image_path
+
+    try:
+        image = Image.open(image_path)
+        image = image.convert("L")
+        image = ImageOps.autocontrast(image)
+        image = image.filter(ImageFilter.SHARPEN)
+
+        scale = max(1.0, min(2.0, getEnvFloat("OCR_SCALE", 1.5)))
+        if scale > 1.0:
+            width, height = image.size
+            resampling = getattr(
+                getattr(Image, "Resampling", Image),
+                "LANCZOS",
+            )
+            image = image.resize(
+                (int(width * scale), int(height * scale)),
+                resampling,
+            )
+
+        image.save(image_path)
+        log("INFO", f"OCR image preprocessed={image_path}")
+    except Exception as e:
+        log("ERROR", f"OCR preprocess failed detail={e}")
+
+    return image_path
+
+
 def runPaddleOcr(engine, image_path):
     if hasattr(engine, "predict"):
         return engine.predict(str(image_path))
@@ -109,6 +152,7 @@ def runOcrFromImage(image_path, show_result=True):
     log("ACTION", f"PaddleOCR start image={image_path}")
 
     try:
+        image_path = preprocessOcrImage(image_path)
         engine = getOcrEngine()
         result = runPaddleOcr(engine, image_path)
     except Exception as e: