feat(math-ct): ЦТ 2020 V1 — 32 задания (5 с PNG-изображениями) + инфраструктура PDF→PNG

backend/scripts/detect_table_rows.py

@@ -0,0 +1,77 @@
+"""
+Detect horizontal table borders in a scanned PDF page PNG
+and extract row bounding boxes.
+
+Usage:
+  python detect_table_rows.py <page_png> [--min-width 0.7] [--debug]
+
+Prints detected row y-ranges as normalized (0-1) coordinates.
+"""
+import sys
+import argparse
+import numpy as np
+from PIL import Image
+
+
+def detect_rows(page_png, min_width_frac=0.7, debug=False):
+    img = Image.open(page_png).convert('L')  # grayscale
+    arr = np.array(img)
+    h, w = arr.shape
+
+    # Binarize: dark pixels (potential lines) = True
+    dark = arr < 128
+    # Count dark pixels per row
+    row_dark_count = dark.sum(axis=1)
+    min_dark = int(min_width_frac * w)
+
+    # Find rows that are mostly dark (horizontal lines)
+    is_line = row_dark_count > min_dark
+
+    # Group consecutive line pixels into bands
+    line_bands = []
+    in_band = False
+    band_start = 0
+    for y in range(h):
+        if is_line[y] and not in_band:
+            in_band = True
+            band_start = y
+        elif not is_line[y] and in_band:
+            in_band = False
+            band_end = y
+            line_bands.append((band_start, band_end))
+    if in_band:
+        line_bands.append((band_start, h))
+
+    if not line_bands:
+        print("No table lines detected. Try reducing --min-width.", file=sys.stderr)
+        return []
+
+    # Extract row y-ranges between consecutive line bands
+    rows = []
+    for i in range(len(line_bands) - 1):
+        y_top = line_bands[i][1]      # bottom of upper border
+        y_bot = line_bands[i + 1][0]  # top of lower border
+        if y_bot - y_top > 5:  # skip tiny gaps
+            rows.append((y_top / h, y_bot / h))
+
+    if debug:
+        print(f"Detected {len(line_bands)} line bands:")
+        for b in line_bands:
+            print(f"  pixels {b[0]}-{b[1]}  (y={b[0]/h:.3f}-{b[1]/h:.3f})")
+        print(f"\nDetected {len(rows)} content rows:")
+        for i, (y0, y1) in enumerate(rows):
+            print(f"  row {i}: y={y0:.3f}-{y1:.3f}  (pixels {int(y0*h)}-{int(y1*h)}, height={int((y1-y0)*h)}px)")
+
+    return rows
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('page_png')
+    parser.add_argument('--min-width', type=float, default=0.7)
+    parser.add_argument('--debug', action='store_true')
+    args = parser.parse_args()
+    rows = detect_rows(args.page_png, min_width_frac=args.min_width, debug=args.debug)
+    if not args.debug:
+        for i, (y0, y1) in enumerate(rows):
+            print(f"row {i}: {y0:.4f} - {y1:.4f}")