import os os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE" import sys import cv2 import numpy as np import re import shutil import subprocess from datetime import datetime from PIL import Image # ========================================== # 0. 【合体版の肝】YomiTokuを安全に裏方で動かすためのセルフ判定 # ========================================== if "--cli-mode" in sys.argv: # GUIを一切起動せず、上書きされた環境下でYomiTokuの解析処理だけをピンポイントで実行します。 # メインアプリのインポートと隔離されているため、不具合を100%回避できます。 try: from yomitoku import DocumentAnalyzer # 引数から入力画像と出力先テキストを取得 img_idx = sys.argv.index("--input") + 1 out_idx = sys.argv.index("--output") + 1 input_path = sys.argv[img_idx] output_path = sys.argv[out_idx] cropped = cv2.imread(input_path) if cropped is not None: analyzer = DocumentAnalyzer(visualize=False, device="cpu") results, _, _ = analyzer(cropped) # 🚀 元プログラムにあったYomiToku専用のスマート重複排除ロジック def deep_extract_strings(obj): res = [] if isinstance(obj, str): if obj.strip(): res.append(obj.strip()) elif isinstance(obj, dict): for v in obj.values(): res.extend(deep_extract_strings(v)) elif isinstance(obj, (list, tuple, set)): for item in obj: res.extend(deep_extract_strings(item)) else: if hasattr(obj, "to_dict"): try: res.extend(deep_extract_strings(obj.to_dict())) except: pass elif hasattr(obj, "dict"): try: res.extend(deep_extract_strings(obj.dict())) except: pass elif hasattr(obj, "__dict__"): try: res.extend(deep_extract_strings(obj.__dict__)) except: pass return res raw_strings = deep_extract_strings(results) unique_strings = [] seen = set() for s in raw_strings: if s not in seen: unique_strings.append(s) seen.add(s) sorted_strings = sorted(unique_strings, key=len, reverse=True) exclude_set = set() joined_check_text = "" for s in sorted_strings: if s in joined_check_text: exclude_set.add(s) else: joined_check_text += " " + s clean_texts = [] jp_pattern = re.compile(r'[\u3040-\u309F\u30A0-\u30FF\u4E00-\u9FFF\u3000-\u303F]') for s in unique_strings: if s in exclude_set: continue if s.replace('.', '', 1).isdigit() or s in ['[', ']', '(', ')', ',', '.']: continue if jp_pattern.search(s): clean_texts.append(s) text_out = "\n".join(clean_texts) if not text_out.strip(): text_out = "(文字は検出されませんでした)" with open(output_path, "w", encoding="utf-8") as f: f.write(text_out) except Exception as e: # エラーが起きた場合はテキストファイルに書き出してメイン側に報告 if "-o" in sys.argv or "--output" in sys.argv: try: out_idx = sys.argv.index("--output") + 1 with open(sys.argv[out_idx], "w", encoding="utf-8") as f: f.write(f"YomiToku裏方実行エラー: {e}") except: pass sys.exit(0) # --- ここから下は通常のGUIアプリ用のインポート --- from PySide6.QtCore import QThread, Signal, Slot, QTimer, QRect, QPoint, Qt from PySide6.QtWidgets import (QApplication, QMainWindow, QPushButton, QTextEdit, QVBoxLayout, QHBoxLayout, QWidget, QFileDialog, QLabel, QScrollArea, QSplitter, QRadioButton, QButtonGroup) from PySide6.QtGui import QPixmap, QImage, QPainter, QPen, QColor, QDragEnterEvent, QDropEvent, QTextCursor from pdf2image import convert_from_path, pdfinfo_from_path # 🚀 各種外部エンジンの基本パス指定 POPPLER_BASE_PATH = r"C:\ut\poppler-26.02.0" NDL_OCR_COMMAND = r"C:\ut\ndlocr-lite-master\src\ocr.py" # ========================================== # 1. 範囲指定可能な画像表示用カスタムラベル # ========================================== class ImageLabel(QLabel): rect_selected = Signal(int, int, int, int) def __init__(self): super().__init__() self.start_pos = QPoint() self.end_pos = QPoint() self.is_drawing = False self.orig_pixmap = None self.setCursor(Qt.CrossCursor) self.setSizePolicy(self.sizePolicy().Policy.Fixed, self.sizePolicy().Policy.Fixed) self.setScaledContents(False) def set_pixmap(self, pixmap): self.orig_pixmap = pixmap self.setPixmap(pixmap) if pixmap: self.resize(pixmap.size()) self.setMinimumSize(pixmap.size()) def mousePressEvent(self, event): if event.button() == Qt.LeftButton and self.orig_pixmap: self.start_pos = event.position().toPoint() self.end_pos = self.start_pos self.is_drawing = True self.update() def mouseMoveEvent(self, event): if self.is_drawing: self.end_pos = event.position().toPoint() self.update() def mouseReleaseEvent(self, event): if event.button() == Qt.LeftButton and self.is_drawing: self.is_drawing = False self.update() x1 = min(self.start_pos.x(), self.end_pos.x()) y1 = min(self.start_pos.y(), self.end_pos.y()) x2 = max(self.start_pos.x(), self.end_pos.x()) y2 = max(self.start_pos.y(), self.end_pos.y()) if (x2 - x1) > 5 and (y2 - y1) > 5: self.rect_selected.emit(x1, y1, x2, y2) def paintEvent(self, event): super().paintEvent(event) if self.is_drawing: painter = QPainter(self) pen = QPen(QColor(255, 0, 0), 2, Qt.DashLine) painter.setPen(pen) rect = QRect(self.start_pos, self.end_pos) painter.drawRect(rect) # ========================================== # 2. バックグラウンドで選択されたOCRを実行するスレッド # ========================================== class CropOCRWorker(QThread): result_ready = Signal(str) status_changed = Signal(str) def __init__(self, cv_image, x1, y1, x2, y2, scale_factor, ocr_mode): super().__init__() self.cv_image = cv_image.copy() if cv_image is not None else None self.x1 = x1 self.y1 = y1 self.x2 = x2 self.y2 = y2 self.scale_factor = scale_factor self.ocr_mode = ocr_mode # "NDL" または "YomiToku" def run(self): if self.cv_image is None: self.result_ready.emit("画像データが存在しません。") return # タイムスタンプを付与した一意のテンポラリフォルダ名を作成 timestamp = datetime.now().strftime("%Y%m%d_%H%M%S_%f")[:-3] temp_dir_name = f"ndl_ocr_temp_{timestamp}" temp_dir = os.path.abspath(temp_dir_name) os.makedirs(temp_dir, exist_ok=True) # 切り抜いた画像を一時保存 img_input_path = os.path.join(temp_dir, "crop_input.png") h, w = self.cv_image.shape[:2] rx1 = max(0, int(self.x1 / self.scale_factor)) ry1 = max(0, int(self.y1 / self.scale_factor)) rx2 = min(w, int(self.x2 / self.scale_factor)) ry2 = min(h, int(self.y2 / self.scale_factor)) cropped = self.cv_image[ry1:ry2, rx1:rx2] if cropped.size == 0: self.result_ready.emit("(有効な範囲が選択されませんでした)") try: shutil.rmtree(temp_dir) except: pass return cv2.imwrite(img_input_path, cropped) output_dir = os.path.join(temp_dir, "ocr_out") os.makedirs(output_dir, exist_ok=True) txt_out_path = os.path.join(output_dir, "crop_input.txt") text_out = "" # 🚀 選択されたモードによって処理を分岐 if self.ocr_mode == "NDL": self.status_changed.emit("NDL用の部品に切り替え中（数秒〜十数秒かかります）...") # 【対策1】裏でのpip自動更新がフリーズの原因になる場合、この1行を頭に「#」をつけて無効化してください subprocess.run([sys.executable, "-m", "pip", "install", "-r", r"C:\ut\ndlocr-lite-master\requirements.txt", "--user", "--quiet"]) self.status_changed.emit("NDL OCRでスキャン中...") # 【対策2】一時フォルダの作成場所を、安全な C:\ut の中に強制隔離 import shutil safe_output_dir = os.path.join(r"C:\ut", os.path.basename(output_dir)) os.makedirs(safe_output_dir, exist_ok=True) safe_img_input_path = os.path.join(safe_output_dir, "crop_input.png") # 元の場所に画像があれば、安全な C:\ut 側にコピーして引き渡す if os.path.exists(img_input_path): shutil.copy2(img_input_path, safe_img_input_path) # コマンドの引数をすべて C:\ut 側の安全なパスに書き換える cmd = [sys.executable, NDL_OCR_COMMAND, "--sourceimg", safe_img_input_path, "--output", safe_output_dir] txt_out_path = os.path.join(safe_output_dir, "crop_input.txt") # 【対策3】text=False にしてWindowsの日本語パス破壊を防ぐ result_raw = subprocess.run(cmd, capture_output=True, text=False) # 【対策4】PyMuPDF等の警告文（文字コードエラー）を安全に受け流す class DecodedResult: stdout = result_raw.stdout.decode('cp932', errors='replace') stderr = result_raw.stderr.decode('cp932', errors='replace') returncode = result_raw.returncode result = DecodedResult if os.path.exists(txt_out_path): try: with open(txt_out_path, "r", encoding="utf-8") as f: text_out = f.read() except Exception as e: text_out = f"テキスト読み込みエラー: {e}" else: if result.returncode != 0: err_msg = result.stderr.strip() if result.stderr else result.stdout.strip() text_out = f"NDL OCRエラー詳細:\n{err_msg}" if err_msg else "NDLの解析スクリプトの起動に失敗しました。" else: text_out = "(文字は検出されませんでした)" else: self.status_changed.emit("YomiToku用の部品に切り替え中（数秒〜十数秒かかります）...") # YomiTokuの依存関係を復旧 ## subprocess.run([sys.executable, "-m", "pip", "install", "networkx>=3.4.2", "reportlab>=4.4.1", "--user", "--quiet"]) self.status_changed.emit("YomiTokuでスキャン中...") # 自分自身（sys.argv[0]）を --cli-mode で裏方として呼び出す高等テクニック cmd = [sys.executable, sys.argv[0], "--cli-mode", "--input", img_input_path, "--output", txt_out_path] result = subprocess.run(cmd, capture_output=True, text=True) if os.path.exists(txt_out_path): try: with open(txt_out_path, "r", encoding="utf-8") as f: text_out = f.read() except Exception as e: text_out = f"テキスト読み込みエラー: {e}" else: err_msg = result.stderr.strip() if result.stderr else result.stdout.strip() text_out = f"YomiTokuエラー詳細:\n{err_msg}" if err_msg else "YomiToku裏方処理の起動に失敗しました。" # 安全クリーンアップ try: shutil.rmtree(temp_dir) except: pass self.result_ready.emit(text_out) # ========================================== # 3. メインの画面クラス（合体・モード切り替え対応） # ========================================== class MainWindow(QMainWindow): def __init__(self): super().__init__() self.setWindowTitle("高精度クロップスキャン OCRツール (NDL / YomiToku ハイブリッド)") self.resize(1200, 750) self.setAcceptDrops(True) self.current_file_path = None self.current_page = 1 self.total_pages = 1 self.current_cv_image = None self.scale_factor = 1.0 self.elapsed_seconds = 0.0 self.timer = QTimer() self.timer.timeout.connect(self.update_timer_label) # UI部品の作成 self.btn_select = QPushButton("画像またはPDFを開く") self.btn_prev = QPushButton("◀ 前のページ") self.btn_next = QPushButton("次のページ ▶") self.lbl_page_num = QLabel("ページ: 0 / 0") self.lbl_page_num.setAlignment(Qt.AlignmentFlag.AlignCenter) self.btn_prev.setEnabled(False) self.btn_next.setEnabled(False) # 🚀 【合体版の肝】エンジン切り替えラジオボタンの作成 self.radio_ndl = QRadioButton("🤖 NDLエンジン") self.radio_yomi = QRadioButton("🦆 YomiToku") self.radio_ndl.setChecked(True) # デフォルトはNDL self.mode_group = QButtonGroup(self) self.mode_group.addButton(self.radio_ndl) self.mode_group.addButton(self.radio_yomi) self.btn_reverse = QPushButton("🔄 行順を反転") self.btn_remove_nl = QPushButton("✂️ 選択範囲の改行を削除") self.btn_save = QPushButton("テキスト保存") self.btn_save.setEnabled(False) self.lbl_status = QLabel("画像・PDFを開くか、ファイルをドラッグ＆ドロップしてください。") self.splitter = QSplitter(Qt.Orientation.Horizontal) self.splitter.setChildrenCollapsible(False) self.scroll_area = QScrollArea() self.img_label = ImageLabel() self.scroll_area.setWidget(self.img_label) self.scroll_area.setWidgetResizable(False) self.text_edit = QTextEdit() self.text_edit.setPlaceholderText("マウスで囲んだ部分のテキストがここに表示されます...") # 🚀 【追加】老眼でもクッキリ見えるよう、右側の文字サイズを大きくします # 16 の数字を「18」や「20」に変えることで、好みの大きさにさらに調整できます self.text_edit.setStyleSheet("font-size: 12pt; font-family: 'Noto Serif JP', 'MS Gothic', sans-serif;") self.splitter.addWidget(self.scroll_area) self.splitter.addWidget(self.text_edit) # レイアウトへの組み込み page_layout = QHBoxLayout() page_layout.addWidget(self.btn_prev) page_layout.addWidget(self.lbl_page_num, stretch=1) page_layout.addWidget(self.btn_next) # モード選択ボタンを並べるレイアウト mode_layout = QHBoxLayout() mode_layout.addWidget(self.radio_ndl) mode_layout.addWidget(self.radio_yomi) top_layout = QHBoxLayout() top_layout.addWidget(self.btn_select) top_layout.addLayout(page_layout, stretch=1) top_layout.addLayout(mode_layout) # 🚀 切り替えスイッチをナビに配置 top_layout.addWidget(self.btn_reverse) top_layout.addWidget(self.btn_remove_nl) top_layout.addWidget(self.btn_save) root_layout = QVBoxLayout() root_layout.addLayout(top_layout) root_layout.addWidget(self.lbl_status) root_layout.addWidget(self.splitter, stretch=1) container = QWidget() container.setLayout(root_layout) self.setCentralWidget(container) # イベント接続 self.btn_select.clicked.connect(self.select_file) self.btn_prev.clicked.connect(self.prev_page) self.btn_next.clicked.connect(self.next_page) self.btn_reverse.clicked.connect(self.reverse_text) self.btn_remove_nl.clicked.connect(self.remove_newlines) self.btn_save.clicked.connect(self.save_file) self.img_label.rect_selected.connect(self.on_rect_selected) def showEvent(self, event): super().showEvent(event) self.splitter.setSizes([700, 500]) def reverse_text(self): cursor = self.text_edit.textCursor() if cursor.hasSelection(): selected_text = cursor.selectedText() lines = selected_text.replace('\u2029', '\n').split('\n') reversed_text = '\n'.join(lines[::-1]) cursor.insertText(reversed_text) self.lbl_status.setText("選択された部分の行順を反転しました。") else: current_text = self.text_edit.toPlainText() if current_text: lines = current_text.split('\n') self.text_edit.setText('\n'.join(lines[::-1])) self.lbl_status.setText("すべての行順を反転しました。") def remove_newlines(self): cursor = self.text_edit.textCursor() if cursor.hasSelection(): selected_text = cursor.selectedText() cleaned_text = selected_text.replace('\u2029', '').replace('\n', '').replace('\r', '') cursor.insertText(cleaned_text) self.lbl_status.setText("選択範囲の改行コードを削除して文章を繋げました。") else: self.lbl_status.setText("エラー: 改行を消したい文章の範囲を、右側のテキストエリアで選択してください。") def dragEnterEvent(self, event: QDragEnterEvent): if event.mimeData().hasUrls(): event.acceptProposedAction() def dropEvent(self, event: QDropEvent): urls = event.mimeData().urls() if urls: raw_path = urls[0].toLocalFile() if os.path.exists(raw_path): self.current_file_path = raw_path self.current_page = 1 self.load_document(raw_path) def select_file(self): file_path, _ = QFileDialog.getOpenFileName( self, "ファイルを開く", "", "対応ファイル (*.pdf *.png *.jpg *.jpeg *.bmp)" ) if file_path: self.current_file_path = file_path self.current_page = 1 self.load_document(file_path) def prev_page(self): if self.current_page > 1: self.current_page -= 1 self.load_document(self.current_file_path) def next_page(self): if self.current_page < self.total_pages: self.current_page += 1 self.load_document(self.current_file_path) def update_page_buttons(self): self.lbl_page_num.setText(f"ページ: {self.current_page} / {self.total_pages}") self.btn_prev.setEnabled(self.current_page > 1) self.btn_next.setEnabled(self.current_page < self.total_pages) def load_document(self, file_path): if isinstance(file_path, (tuple, list)): file_path = file_path[0] ext = os.path.splitext(file_path)[1].lower() pil_img = None self.lbl_status.setText(f"ドキュメントの {self.current_page} ページ目を読み込み中...") QApplication.processEvents() possible_bins = [ POPPLER_BASE_PATH, os.path.join(POPPLER_BASE_PATH, "bin"), os.path.join(POPPLER_BASE_PATH, "Library", "bin") ] actual_poppler_bin = None for p in possible_bins: if os.path.exists(os.path.join(p, "pdftoppm.exe")): actual_poppler_bin = p break if ext == '.pdf': if not actual_poppler_bin: self.lbl_status.setText(f"エラー: {POPPLER_BASE_PATH} 内に pdftoppm.exe が見つかりません。") return try: info = pdfinfo_from_path(file_path, poppler_path=actual_poppler_bin) self.total_pages = int(info.get("Pages", 1)) images = convert_from_path( file_path, first_page=self.current_page, last_page=self.current_page, poppler_path=actual_poppler_bin ) if images: pil_img = images[0] except Exception as e: self.lbl_status.setText(f"PDF読み込みエラー: {e}") return elif ext in ['.png', '.jpg', '.jpeg', '.bmp']: self.total_pages = 1 self.current_page = 1 try: pil_img = Image.open(file_path) except Exception as e: self.lbl_status.setText(f"画像読み込みエラー: {e}") return self.update_page_buttons() if pil_img: pil_img = pil_img.convert('RGB') self.current_cv_image = cv2.cvtColor(np.array(pil_img), cv2.COLOR_RGB2BGR) display_height = 650 w, h = pil_img.size self.scale_factor = display_height / h nw, nh = int(w * self.scale_factor), int(h * self.scale_factor) rgb_image = cv2.cvtColor(self.current_cv_image, cv2.COLOR_BGR2RGB) bytes_per_line = 3 * w qimg = QImage(rgb_image.data, w, h, bytes_per_line, QImage.Format_RGB888) safe_qimg = qimg.copy() pixmap = QPixmap.fromImage(safe_qimg).scaled(nw, nh, Qt.KeepAspectRatio, Qt.SmoothTransformation) self.img_label.set_pixmap(pixmap) self.lbl_status.setText(f"{self.current_page} ページ目の読み込み成功！ 範囲をドラッグしてスキャンしてください。") self.btn_save.setEnabled(True) def on_rect_selected(self, x1, y1, x2, y2): if self.current_cv_image is None: return self.elapsed_seconds = 0.0 self.lbl_status.setText("準備中... (0.0秒経過)") self.timer.start(100) # 🚀 画面上のラジオボタンがどちらを選択しているかを判定 ocr_mode = "NDL" if self.radio_ndl.isChecked() else "YomiToku" # モード情報をワーカーに引き渡して起動 self.worker = CropOCRWorker(self.current_cv_image, x1, y1, x2, y2, self.scale_factor, ocr_mode) self.worker.status_changed.connect(self.lbl_status.setText) self.worker.result_ready.connect(self.on_ocr_finished) self.worker.start() @Slot(str) def on_ocr_finished(self, text): self.timer.stop() current_text = self.text_edit.toPlainText() if current_text: current_text += "\n\n" self.text_edit.setText(current_text + text) self.text_edit.moveCursor(QTextCursor.MoveOperation.End) ocr_mode = "NDL" if self.radio_ndl.isChecked() else "YomiToku" self.lbl_status.setText(f"[{ocr_mode}モード] スキャン完了！ ({self.elapsed_seconds:.1f}秒) 必要に応じて文章を補正してください。") def update_timer_label(self): self.elapsed_seconds += 0.1 # ステータスラベルの末尾の時間表示だけをリアルタイム更新 current_status = self.lbl_status.text() if "(" in current_status and "秒経過" in current_status: base_status = current_status.split("(")[0].strip() self.lbl_status.setText(f"{base_status} ({self.elapsed_seconds:.1f}秒経過)") elif "中..." in current_status: self.lbl_status.setText(f"{current_status} ({self.elapsed_seconds:.1f}秒経過)") def save_file(self): file_path, _ = QFileDialog.getSaveFileName( self, "テキストを保存", "ocr_result.txt", "Text Files (*.txt)" ) if file_path: with open(file_path, "w", encoding="utf-8") as f: f.write(self.text_edit.toPlainText()) self.lbl_status.setText(f"保存しました: {file_path}") if __name__ == "__main__": app = QApplication(sys.argv) window = MainWindow() window.show() sys.exit(app.exec())