OCR用のapp.pyを作成

app.py

@@ -0,0 +1,56 @@
+import gradio as gr
+import pytesseract
+from PIL import Image
+import fitz  # PyMuPDF
+import io
+import cv2
+import numpy as np
+import os
+
+pytesseract.pytesseract.tesseract_cmd = "tesseract"  # Hugging Face内はこれでOK
+
+def preprocess_image_for_ocr(image):
+    gray = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2GRAY)
+    clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
+    gray = clahe.apply(gray)
+    blurred = cv2.bilateralFilter(gray, 9, 75, 75)
+    sharp = cv2.addWeighted(blurred, 1.5, cv2.GaussianBlur(blurred, (0, 0), 5), -0.5, 0)
+    denoised = cv2.fastNlMeansDenoising(sharp, h=20)
+    adaptive = cv2.adaptiveThreshold(denoised, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
+                                     cv2.THRESH_BINARY, 35, 11)
+    _, otsu = cv2.threshold(denoised, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
+    binary = cv2.bitwise_and(adaptive, otsu)
+    opened = cv2.morphologyEx(binary, cv2.MORPH_OPEN, np.ones((1, 1), np.uint8))
+    h, w = opened.shape
+    opened = cv2.resize(opened, (w * 2, h * 2), interpolation=cv2.INTER_CUBIC)
+    return Image.fromarray(opened)
+
+def ocr_from_file(file):
+    ext = os.path.splitext(file.name)[1].lower()
+    result_text = ""
+    try:
+        if ext in ['.png', '.jpg', '.jpeg', '.bmp', '.tiff']:
+            image = Image.open(file)
+            processed = preprocess_image_for_ocr(image)
+            result_text = pytesseract.image_to_string(processed, lang="jpn+eng")
+        elif ext == '.pdf':
+            doc = fitz.open(file.name)
+            for page_num in range(min(len(doc), 3)):
+                page = doc.load_page(page_num)
+                pix = page.get_pixmap(matrix=fitz.Matrix(4, 4))
+                img = Image.open(io.BytesIO(pix.tobytes()))
+                processed = preprocess_image_for_ocr(img)
+                result_text += pytesseract.image_to_string(processed, lang="jpn+eng") + "\n"
+        else:
+            return "非対応形式です。"
+    except Exception as e:
+        return f"OCRエラー: {e}"
+    return result_text.strip()
+
+gr.Interface(
+    fn=ocr_from_file,
+    inputs=gr.File(label="画像またはPDFをアップロード"),
+    outputs=gr.Textbox(label="OCR結果"),
+    title="🚀 高精度OCR Lite",
+    description="日本語・英語対応の高精度OCRツール（Lite体験版）"
+).launch()