| import numpy as np
|
|
|
| def merge_line_boxes(det_res, x_margin_ratio=2.5, y_overlap_thresh=0.45):
|
| """
|
| Умное объединение разрезанных боксов одной строки (Horizontal Line Merging).
|
|
|
| :param det_res: список кортежей (box, text, score) от PaddleOCR/RapidOCR
|
| :param x_margin_ratio: максимальное расстояние между словами одной строки (в долях высоты)
|
| :param y_overlap_thresh: минимальный коэффициент вертикального перекрытия для принадлежности к одной строке
|
| """
|
| if det_res is None or len(det_res) == 0:
|
| return []
|
|
|
|
|
| rects = []
|
| for item in det_res:
|
| if isinstance(item, (list, tuple)) and len(item) > 0 and isinstance(item[0], (list, tuple, np.ndarray)):
|
| box = np.array(item[0]).astype(int)
|
| else:
|
| box = np.array(item).astype(int)
|
|
|
| if box.ndim == 2 and box.shape[0] >= 4:
|
| xmin = int(np.min(box[:, 0]))
|
| xmax = int(np.max(box[:, 0]))
|
| ymin = int(np.min(box[:, 1]))
|
| ymax = int(np.max(box[:, 1]))
|
| elif len(box) == 4:
|
| xmin, ymin, xmax, ymax = int(box[0]), int(box[1]), int(box[2]), int(box[3])
|
| else:
|
| continue
|
|
|
| h = max(1, ymax - ymin)
|
| w = max(1, xmax - xmin)
|
| rects.append({
|
| "box": [xmin, ymin, xmax, ymax],
|
| "xmin": xmin, "xmax": xmax,
|
| "ymin": ymin, "ymax": ymax,
|
| "h": h, "w": w,
|
| "cy": (ymin + ymax) / 2.0
|
| })
|
|
|
| if not rects:
|
| return []
|
|
|
|
|
| rects.sort(key=lambda r: (r["ymin"], r["xmin"]))
|
|
|
|
|
| lines = []
|
| for r in rects:
|
| placed = False
|
| for line in lines:
|
| line_ymin = min(x["ymin"] for x in line)
|
| line_ymax = max(x["ymax"] for x in line)
|
| line_cy = (line_ymin + line_ymax) / 2.0
|
| line_h = max(1, line_ymax - line_ymin)
|
|
|
|
|
| center_diff = abs(r["cy"] - line_cy)
|
| if center_diff <= min(r["h"], line_h) * 0.45:
|
| line.append(r)
|
| placed = True
|
| break
|
| if not placed:
|
| lines.append([r])
|
|
|
|
|
| merged_boxes = []
|
| for line in lines:
|
| line.sort(key=lambda r: r["xmin"])
|
|
|
| cur_xmin = line[0]["xmin"]
|
| cur_xmax = line[0]["xmax"]
|
| cur_ymin = line[0]["ymin"]
|
| cur_ymax = line[0]["ymax"]
|
|
|
| for next_r in line[1:]:
|
| avg_h = ((cur_ymax - cur_ymin) + next_r["h"]) / 2.0
|
| dist_x = next_r["xmin"] - cur_xmax
|
|
|
|
|
| if dist_x <= avg_h * x_margin_ratio:
|
| cur_xmax = max(cur_xmax, next_r["xmax"])
|
| cur_ymin = min(cur_ymin, next_r["ymin"])
|
| cur_ymax = max(cur_ymax, next_r["ymax"])
|
| else:
|
|
|
| merged_boxes.append([cur_xmin, cur_ymin, cur_xmax, cur_ymax])
|
| cur_xmin = next_r["xmin"]
|
| cur_xmax = next_r["xmax"]
|
| cur_ymin = next_r["ymin"]
|
| cur_ymax = next_r["ymax"]
|
|
|
| merged_boxes.append([cur_xmin, cur_ymin, cur_xmax, cur_ymax])
|
|
|
|
|
| merged_boxes.sort(key=lambda b: (b[1], b[0]))
|
| return merged_boxes
|
|
|
