import sys import json import cv2 import numpy as np import pytesseract import re def preprocess_image(image_path, scale=2): image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE) height, width = image.shape[:2] image = cv2.resize(image, (int(width * scale), int(height * scale)), interpolation=cv2.INTER_CUBIC) _, binary = cv2.threshold(image, 128, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU) image = cv2.equalizeHist(image) image = cv2.GaussianBlur(image, (3, 3), 0) # ノイズ除去 return binary def enhance_lines(binary): horizontal_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (60, 1)) horizontal_lines = cv2.morphologyEx(binary, cv2.MORPH_OPEN, horizontal_kernel, iterations=3) vertical_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, 60)) vertical_lines = cv2.morphologyEx(binary, cv2.MORPH_OPEN, vertical_kernel, iterations=3) return horizontal_lines, vertical_lines def remove_lines(binary, horizontal_lines, vertical_lines): mask = cv2.add(horizontal_lines, vertical_lines) binary_no_lines = cv2.bitwise_and(binary, cv2.bitwise_not(mask)) return binary_no_lines def extract_text_from_image(image): return pytesseract.image_to_string(image, config="--oem 3 --psm 6", lang="jpn") def correct_misrecognized_minus(text): return text.replace('--', '-') # def correct_misrecognized_spaces(text): # """数値の間の不要なスペースを削除""" # # return re.sub(r'(-?\d+)\s+(\.\d+)', r'\1\2', text) # # return re.sub(r'(?<=\d)\s+(?=\.\d)', '', text) # # return re.sub(r'(?<=-?\d+)\s+(?=\.\d+)', '', text) # # return re.sub(r'(?<=\d)\s+(?=\.\d)', '', text) # 数値と小数点の間のスペースのみ削除 # # return re.sub(r'(?<=-?\d{1,6})\s+(?=\.\d+)', '', text) # 数値と小数点の間のスペースのみ削除 # # text = re.sub(r'(?<=-?\d{1,6})\s+(?=\.\d+)', '', text) # 数値と小数点の間のスペースのみ削除 # # text = re.sub(r'(?<=\d)\s+(?=\d)', '', text) # 数値間の不要なスペースを削除 # # return text # # text = re.sub(r'(?<=-?\d)\s+(?=\.\d+)', '', text) # 小数点の前の不要なスペース削除 # # text = re.sub(r'(?<=\d)\s+(?=\d)', '', text) # 数値間の不要なスペースを削除 # # return text # text = re.sub(r'(?<=\d{1,10})\s+(?=\.\d+)', '', text) # 小数点の前の不要なスペース削除（固定長の look-behind） # text = re.sub(r'(?<=\d)\s+(?=\d)', '', text) # 数値間の不要なスペースを削除 # return text def correct_misrecognized_characters(text): """OCR誤認識の修正（7→/ の誤認識を修正）""" # "7/" → "7" text = re.sub(r'7/', '7', text) # "/7" → "7" text = re.sub(r'/7', '7', text) # 残りのスラッシュを "7" に text = text.replace('/', '7') text = text.replace('ー', '-') text = text.replace('一', '-') # 連続したマイナス(--)以上を単一のマイナス(-)に修正 text = re.sub(r'-{2,}', '-', text) # 指定された文字を削除 text = re.sub(r'[|ー、。「]', '', text) text = text.replace('ー', '-') text = text.replace('一', '-') text = re.sub(r'[|ー、。「]', '', text) # 指定された文字を削除 return text def split_text_into_2d_array(text): lines = text.strip().split('\n') result = [] for line in lines: line = line.strip() line = re.sub(r'(-?\d+\.\d+)(-\d+\.\d+)', r'\1 \2', line) # 数値の連続を分割 split_line = re.split(r'\s+', line) result.append(split_line) return result def merge_split_decimal_numbers(table): """ 数値のあとに . があり、次のセルが数値の場合、それらを結合する """ merged_table = [] for row in table: merged_row = [] skip_next = False for i in range(len(row)): if skip_next: skip_next = False continue if i < len(row) - 1 and re.match(r'^-?\d+\.$', row[i]) and re.match(r'^\d+$', row[i + 1]): merged_row.append(row[i] + row[i + 1]) # 結合 skip_next = True # 次の要素をスキップ else: merged_row.append(row[i]) merged_table.append(merged_row) return merged_table def merge_single_characters(table): # '-' 一文字だけの要素を削除 table = [[cell for cell in row if cell != '-'] for row in table] """ 一文字または二文字の要素の後に一文字の要素が続いた場合、それらを結合する """ merged_table = [] for row in table: merged_row = [] skip_next = False for i in range(len(row)): if skip_next: skip_next = False continue if i < len(row) - 1 and len(row[i]) in [1, 2] and len(row[i + 1]) == 1: merged_row.append(row[i] + row[i + 1]) # 結合 skip_next = True # 次の要素をスキップ else: merged_row.append(row[i]) merged_table.append(merged_row) return merged_table def propagate_minus_signs(table): """ 前の行の同じ列にマイナスがついていたら、自己にもマイナスをつける """ num_cols = max(len(row) for row in table if row) # 最大列数を取得（不均一な行も処理） for col in range(num_cols): # 列単位で処理 for row in range(1, len(table)): # 1行目から処理開始 if col < len(table[row - 1]) and col < len(table[row]): # 列の範囲をチェック prev_value = str(table[row - 1][col]).strip() current_value = str(table[row][col]).strip() # 可能な限り条件を緩める（マイナスで始まる、またはマイナス記号を含む） if ('-' in prev_value) and ('-' not in current_value): table[row][col] = '-' + current_value # マイナスを適用 return table def is_number(value): """ 文字列が小数（小数点を含む数値）かどうかを判定 """ try: cleaned_value = value.replace(' ', '') # 空白を削除 return '.' in cleaned_value and float(cleaned_value) except ValueError: return False def filter_numeric_rows(data): """ 2列目と3列目が数値の行のみを抽出 """ filtered_data = [] for row in data: if len(row) >= 3 and is_number(row[1]) and is_number(row[2]): filtered_data.append(row) return filtered_data def extract_table_from_image(image_path): binary = preprocess_image(image_path) horizontal_lines, vertical_lines = enhance_lines(binary) binary_no_lines = remove_lines(binary, horizontal_lines, vertical_lines) raw_output = extract_text_from_image(binary_no_lines) corrected_text = correct_misrecognized_minus(raw_output) corrected_text = correct_misrecognized_characters(corrected_text) # 7→/ の誤認識を修正 text_2d_array = split_text_into_2d_array(corrected_text) merged_text_2d_array = merge_split_decimal_numbers(text_2d_array) fully_merged_text_2d_array = merge_single_characters(merged_text_2d_array) finalized_table = propagate_minus_signs(fully_merged_text_2d_array) # 2列目と3列目が数値の行のみを抽出 filtered_data = filter_numeric_rows(fully_merged_text_2d_array ) finalized_table = propagate_minus_signs(filtered_data) return { "success": True, "raw_output": raw_output, "test_data": fully_merged_text_2d_array , "structured_data": filtered_data, } if __name__ == "__main__": if len(sys.argv) < 2: print(json.dumps({"success": False, "error": "Missing image path"})) sys.exit(1) image_path = sys.argv[1] result = extract_table_from_image(image_path) print(json.dumps(result, ensure_ascii=False, indent=2)) # import sys # import json # import cv2 # import numpy as np # import pytesseract # import re # def preprocess_image(image_path, scale=2): # image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE) # height, width = image.shape[:2] # image = cv2.resize(image, (int(width * scale), int(height * scale)), interpolation=cv2.INTER_CUBIC) # _, binary = cv2.threshold(image, 128, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU) # image = cv2.equalizeHist(image) # image = cv2.GaussianBlur(image, (3, 3), 0) # ノイズ除去 # return binary # def enhance_lines(binary): # horizontal_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (60, 1)) # horizontal_lines = cv2.morphologyEx(binary, cv2.MORPH_OPEN, horizontal_kernel, iterations=3) # vertical_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, 60)) # vertical_lines = cv2.morphologyEx(binary, cv2.MORPH_OPEN, vertical_kernel, iterations=3) # return horizontal_lines, vertical_lines # def remove_lines(binary, horizontal_lines, vertical_lines): # mask = cv2.add(horizontal_lines, vertical_lines) # binary_no_lines = cv2.bitwise_and(binary, cv2.bitwise_not(mask)) # return binary_no_lines # def extract_cells(binary, horizontal_lines, vertical_lines): # table_structure = cv2.add(horizontal_lines, vertical_lines) # contours, _ = cv2.findContours(table_structure, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) # cells = [cv2.boundingRect(cnt) for cnt in contours if cv2.boundingRect(cnt)[2] > 30 and cv2.boundingRect(cnt)[3] > 20] # return sorted(cells, key=lambda c: (c[1] // 10, c[0])) # def extract_cells(binary, horizontal_lines, vertical_lines): # """ 罫線からセルを抽出する """ # table_structure = cv2.add(horizontal_lines, vertical_lines) # contours, _ = cv2.findContours(table_structure, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) # cells = [] # for cnt in contours: # x, y, w, h = cv2.boundingRect(cnt) # if w > 30 and h > 20: # 小さすぎる領域は無視 # cells.append((x, y, w, h)) # # 上から順にソート（行単位でグループ化） # cells = sorted(cells, key=lambda c: (c[1] // 10, c[0])) # `// 10` で近い値をまとめる # return cells # def is_number(value): # """ 文字列が数値（整数または小数）かどうかを判定 """ # try: # cleaned_value = value.replace(' ', '') # 空白を削除 # float(cleaned_value) # return True # except ValueError: # return False # def filter_numeric_rows(data): # """ 2列目と3列目が数値の行のみを抽出 """ # filtered_data = [] # for row in data: # if len(row) >= 3 and is_number(row[1]) and is_number(row[2]): # filtered_data.append(row) # return filtered_data # def correct_misrecognized_numbers(text): # """ OCR誤認識を補正（3→5、5→3の誤認識修正） """ # text = text.replace('５', '3').replace('３', '5') # text = re.sub(r'(? 0 and len(test_datas[0]) > 1 else "" # return { # "success": True, # "structured_data": filtered_data, # "raw_output": raw_output, # "chiban_data": chiban_data, # "test_data": test_datas # } # if __name__ == "__main__": # if len(sys.argv) < 2: # print(json.dumps({"success": False, "error": "Missing image path"})) # sys.exit(1) # image_path = sys.argv[1] # result = extract_table_from_image(image_path) # print(json.dumps(result, ensure_ascii=False, indent=2)) # import sys # import json # import cv2 # import numpy as np # import pytesseract # import re # def preprocess_image(image_path, scale=2): # """ 画像の前処理（グレースケール変換、二値化、ノイズ除去） """ # image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE) # # 拡大処理 # height, width = image.shape[:2] # image = cv2.resize(image, (int(width * scale), int(height * scale)), interpolation=cv2.INTER_CUBIC) # # 二値化（適応的閾値処理） # binary = cv2.adaptiveThreshold(image, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY_INV, 15, 10) # return binary # def enhance_lines(binary): # """ 罫線を強調する処理 """ # horizontal_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (40, 1)) # horizontal_lines = cv2.morphologyEx(binary, cv2.MORPH_OPEN, horizontal_kernel, iterations=2) # horizontal_lines = cv2.dilate(horizontal_lines, horizontal_kernel, iterations=1) # vertical_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, 40)) # vertical_lines = cv2.morphologyEx(binary, cv2.MORPH_OPEN, vertical_kernel, iterations=2) # vertical_lines = cv2.dilate(vertical_lines, vertical_kernel, iterations=1) # return horizontal_lines, vertical_lines # def remove_lines(binary, horizontal_lines, vertical_lines): # """ 罫線を削除し、OCR用の画像を作成 """ # mask = cv2.add(horizontal_lines, vertical_lines) # 水平線と垂直線を合成 # binary_no_lines = cv2.bitwise_and(binary, cv2.bitwise_not(mask)) # return binary_no_lines # def extract_cells(binary, horizontal_lines, vertical_lines): # """ 罫線からセルを抽出する """ # table_structure = cv2.add(horizontal_lines, vertical_lines) # contours, _ = cv2.findContours(table_structure, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) # cells = [] # for cnt in contours: # x, y, w, h = cv2.boundingRect(cnt) # if w > 30 and h > 20: # 小さすぎる領域は無視 # cells.append((x, y, w, h)) # # 上から順にソート（行単位でグループ化） # cells = sorted(cells, key=lambda c: (c[1] // 10, c[0])) # `// 10` で近い値をまとめる # return cells # def is_number(value): # """ 文字列が数値（整数または小数）かどうかを判定 """ # try: # float(value) # return True # except ValueError: # return False # def filter_numeric_rows(data): # """ 2列目と3列目が数値の行のみを抽出 """ # filtered_data = [] # for row in data: # if len(row) >= 3 and is_number(row[1]) and is_number(row[2]): # filtered_data.append(row) # return filtered_data # def correct_misrecognized_numbers(text): # """ OCR誤認識を補正 """ # corrected_text = re.sub(r'(?<=\d)e(?=\d)', '6', text) # corrected_text = corrected_text.replace('ー', '-') # corrected_text = corrected_text.replace('一', '-') # return corrected_text # def extract_text_from_cells(image, cells): # """ 各セルのOCR結果を取得 """ # structured_data = [] # row = [] # last_y = None # raw_output_list = [] # 生データ用リスト # for x, y, w, h in cells: # if last_y is not None and abs(y - last_y) > 20: # structured_data.append(row) # row = [] # cell_image = image[y:y+h, x:x+w] # cell_image = cv2.bitwise_not(cell_image) # 反転してOCRしやすく # ocr_result = pytesseract.image_to_string(cell_image, config="--oem 1 --psm 6", lang="jpn").strip() # corrected_text = correct_misrecognized_numbers(ocr_result) # row.append(corrected_text) # raw_output_list.append(ocr_result) # 生データに追加 # last_y = y # if row: # structured_data.append(row) # raw_output = "\n".join(raw_output_list) # すべてのOCR結果を結合 # return structured_data, raw_output # def extract_table_from_image(image_path, scale=2): # """ 画像から表データを抽出 """ # binary = preprocess_image(image_path, scale) # horizontal_lines, vertical_lines = enhance_lines(binary) # binary_no_lines = remove_lines(binary, horizontal_lines, vertical_lines) # cells = extract_cells(binary, horizontal_lines, vertical_lines) # extracted_text, raw_output = extract_text_from_cells(binary_no_lines, cells) # # 2列目と3列目が数値の行のみを抽出 # filtered_data = filter_numeric_rows(extracted_text) # return { # "success": True, # "structured_data": filtered_data, # "raw_output": raw_output # } # if __name__ == "__main__": # if len(sys.argv) < 2: # print(json.dumps({"success": False, "error": "Missing image path"})) # sys.exit(1) # image_path = sys.argv[1] # result = extract_table_from_image(image_path) # print(json.dumps(result, ensure_ascii=False, indent=2)) # import sys # import json # import cv2 # import numpy as np # import pytesseract # import re # def preprocess_image(image_path,scale): # image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE) # # 画像を2倍に拡大 # height, width = image.shape[:2] # image = cv2.resize(image, (int(width * scale), int(height * scale)), interpolation=cv2.INTER_CUBIC) # _, binary = cv2.threshold(image, 128, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU) # # コントラスト強調 # image = cv2.equalizeHist(image) # # ガウシアンブラーでノイズ除去 # image = cv2.GaussianBlur(image, (3, 3), 0) # horizontal_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (40, 1)) # detected_lines = cv2.morphologyEx(binary, cv2.MORPH_OPEN, horizontal_kernel, iterations=2) # binary = cv2.subtract(binary, detected_lines) # vertical_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, 40)) # detected_lines = cv2.morphologyEx(binary, cv2.MORPH_OPEN, vertical_kernel, iterations=2) # binary = cv2.subtract(binary, detected_lines) # temp_path = "/tmp/processed_image.png" # cv2.imwrite(temp_path, binary) # return temp_path # def correct_six_misrecognition(text): # corrected_text = re.sub(r'(?<=\d)e(?=\d)', '6', text) # corrected_text = re.sub(r'-e', '-6', corrected_text) # corrected_text = corrected_text.replace('ー', '-') # OCRの'ー'を'-'に変換 # corrected_text = re.sub(r'(?<=\d)z(?=\d)', '2', corrected_text) # 数値に挟まれた'z'を'2'に変換 # return corrected_text # def clean_numeric(cell): # return re.sub(r'[^0-9\.-]', '', cell) # def merge_split_numbers(cells): # """ # -69022. 285 のように、数値が分割されている場合に結合する。 # """ # merged_cells = [] # i = 0 # while i < len(cells): # if i < len(cells) - 1 and re.match(r'-?\d+\.$', cells[i]) and re.match(r'\d+', cells[i + 1]): # merged_cells.append(cells[i] + cells[i + 1]) # i += 2 # else: # merged_cells.append(cells[i]) # i += 1 # return merged_cells # def extract_text_from_image(image_path, scale): # processed_image = preprocess_image(image_path, scale) # if processed_image is None: # return {"success": False, "error": "Failed to process image"} # ocr_result = pytesseract.image_to_string(processed_image, config="--oem 1 --psm 6 --dpi 1000", lang="jpn") # ocr_text = correct_six_misrecognition(ocr_result.strip().replace('|', '')) # lines = [line for line in ocr_text.split('\n') if line] # structured_data = [] # chiban_data = "" # if lines: # first_line_words = lines[0].split() # if first_line_words: # chiban_data = first_line_words[-1] # 最後の単語を抽出 # for line in lines: # cells = line.split() # cleaned_cells = [clean_numeric(cell) for cell in cells if clean_numeric(cell)] # merged_cells = merge_split_numbers(cleaned_cells) # col1 = next((cell for cell in cells if not re.match(r'-?\d+\.\d+', cell)), '9') # col2, col3, col4 = None, None, None # for cell in merged_cells: # if len(cell) >= 5 and re.match(r'-?\d+\.\d+', cell): # 5桁以上の数値、小数点を考慮 # if col2 is None: # col2 = cell # elif col3 is None: # col3 = cell # elif col4 is None: # col4 = cell # break # if col2 and col3 and col4: # 3つの数値がそろった行のみを出力 # structured_data.append([ # col1, # col2, # col3, # col4 # ]) # return {"success": True, "raw_output": ocr_text, "structured_data": structured_data, "chiban_data": chiban_data} # if __name__ == "__main__": # if len(sys.argv) < 3: # print(json.dumps({"success": False, "error": "Missing arguments. Usage: script.py "})) # sys.exit(1) # image_path = sys.argv[1] # scale = float(sys.argv[2]) # result = extract_text_from_image(image_path, scale) # print(json.dumps(result, ensure_ascii=False, indent=2))