import re with open('youtube_tab_to_pdf.py', 'r', encoding='utf-8') as f: code = f.read() new_func = """def extract_unique_scroll(frames: List[np.ndarray], threshold: float = SIMILARITY_THRESHOLD) -> List[np.ndarray]: print(f"[4/5] 순차 Number Sprite Template 앵커 기반 마디 추출 중...") strip_tops, strip_bottoms = [], [] for frame in frames[:50]: strip = _find_white_tab_strip(frame) if strip: strip_tops.append(strip[0]) strip_bottoms.append(strip[1]) if not strip_tops: return [] median_top = int(np.median(strip_tops)) median_bottom = int(np.median(strip_bottoms)) unique_measures = [] chunk_width = 1280 def get_number_sprite(m_img): # We explicitly use inverse thresholding to capture the tiny white number on black background gray = np.max(m_img, axis=2) _, thresh = cv2.threshold(gray, 200, 255, cv2.THRESH_BINARY) row_sums = np.sum(thresh, axis=1) / 255 staff_lines = np.where(row_sums > m_img.shape[1] * 0.5)[0] y_staff = staff_lines[0] if len(staff_lines) > 0 else 50 crop_y1 = max(0, y_staff - 35) crop_y2 = max(0, y_staff - 2) crop_x1 = 0 crop_x2 = min(60, m_img.shape[1]) if crop_y2 <= crop_y1 or crop_x2 <= crop_x1: return None sprite = thresh[crop_y1:crop_y2, crop_x1:crop_x2] # MUST BE STRICT: If there are fewer than 8 white pixels, it's a BLANK SPRITE. # Blank sprites caused the catastrophic 1->36 time-travel deletion! if np.count_nonzero(sprite > 127) < 8: return None return sprite for frame_idx, frame in enumerate(frames): h = frame.shape[0] tab_crop = frame[max(0, median_top):min(h, median_bottom), :] if not _has_tab_content(tab_crop): continue gray_page = _extract_print_channel(tab_crop) bar_coords = _detect_measure_bars(gray_page) if not bar_coords: continue coords = [0] + bar_coords + [tab_crop.shape[1]] coords = sorted(list(set(coords))) page_measures = [] for i in range(len(coords) - 1): x_start = coords[i] x_end = coords[i+1] if x_end - x_start < 40: continue page_measures.append(tab_crop[:, x_start:x_end]) if not page_measures: continue if not unique_measures: unique_measures.extend(page_measures) first_sprite = get_number_sprite(page_measures[0]) has_pixels = np.count_nonzero(first_sprite > 127) if first_sprite is not None else 0 print(f" -> [초기화] 첫 프레임 배열 등록: {len(page_measures)}개 마디 (Sprite Pixels: {has_pixels})") continue first_m = page_measures[0] first_sprite = get_number_sprite(first_m) anchored = False new_start_offset = 0 best_val = 0.0 # Only attempt anchor if the first measure explicitly displays a sequence number. # If it's blank, we DO NOT blindly match it to other blank measures! if first_sprite is not None: # We can scan backwards up to 15 measures because clear Number Sprites are completely unique IDs. for scan_dist in range(1, min(15, len(unique_measures) + 1)): past_idx = len(unique_measures) - scan_dist past_m = unique_measures[past_idx] past_sprite = get_number_sprite(past_m) if past_sprite is not None: hs = min(first_sprite.shape[0], past_sprite.shape[0]) ws = min(first_sprite.shape[1], past_sprite.shape[1]) s1 = first_sprite[:hs, :ws] s2 = past_sprite[:hs, :ws] template = s1[2:-2, 2:-2] if template.shape[0] >= 5 and template.shape[1] >= 5: res = cv2.matchTemplate(s2, template, cv2.TM_CCOEFF_NORMED) max_val = res[0][0] if max_val > best_val: best_val = max_val new_start_offset = len(unique_measures) - past_idx if best_val > 0.85: anchored = True # If we failed to anchor via Sprite (maybe this page has no numbers at all), # we fallback to strict whole-measure Template Matching (TM_CCOEFF_NORMED on greyscale prints to survive subpixel scroll drift) if not anchored: bin_first = _extract_print_channel(first_m) # greyscale thresholded for scan_dist in range(1, min(4, len(unique_measures) + 1)): # strictly limit to 4 to prevent musical loops past_idx = len(unique_measures) - scan_dist past_m = unique_measures[past_idx] bin_past = _extract_print_channel(past_m) if abs(bin_first.shape[1] - bin_past.shape[1]) <= 30: hs = min(bin_first.shape[0], bin_past.shape[0]) ws = min(bin_first.shape[1], bin_past.shape[1]) s1 = bin_first[:hs, :ws] s2 = bin_past[:hs, :ws] template = s1[10:-10, 10:-10] if template.shape[0] >= 10 and template.shape[1] >= 10: res = cv2.matchTemplate(s2, template, cv2.TM_CCOEFF_NORMED) max_val = res[0][0] if max_val > 0.85: new_start_offset = len(unique_measures) - past_idx anchored = True break if anchored and new_start_offset < len(page_measures): unique_measures.extend(page_measures[new_start_offset:]) elif not anchored: unique_measures.extend(page_measures) print(f" -> 동기화 중복 제거 완료: Number Sprite 시계열 기반 {len(unique_measures)}개 마디 보존") final_chunks = [] current_row_measures = [] current_row_width = 0 for measure_img in unique_measures: measure_w = measure_img.shape[1] if current_row_width + measure_w > chunk_width and len(current_row_measures) > 0: row_img = np.hstack(current_row_measures) pad_w = chunk_width - row_img.shape[1] if pad_w > 0: pad_img = np.full((row_img.shape[0], pad_w, 3), 255, dtype=np.uint8) row_img = np.hstack([row_img, pad_img]) final_chunks.append(row_img) current_row_measures = [measure_img] current_row_width = measure_w else: current_row_measures.append(measure_img) current_row_width += measure_w if current_row_measures: row_img = np.hstack(current_row_measures) if row_img.shape[1] > chunk_width: row_img = row_img[:, :chunk_width] else: pad_w = chunk_width - row_img.shape[1] if pad_w > 0: pad_img = np.full((row_img.shape[0], pad_w, 3), 255, dtype=np.uint8) row_img = np.hstack([row_img, pad_img]) final_chunks.append(row_img) print(f" -> A4 분할 컷: {len(final_chunks)}개 줄(Row)") return final_chunks """ pattern = r'def extract_unique_scroll\(frames: List\[np\.ndarray\], threshold: float = SIMILARITY_THRESHOLD\) -> List\[np\.ndarray\]:.*?return final_chunks' new_code = re.sub(pattern, new_func, code, flags=re.DOTALL) with open('youtube_tab_to_pdf.py', 'w', encoding='utf-8') as f: f.write(new_code) print("Supreme Logic Embedded.")