refactor: 后处理阈值完全自动化，零人工参数

- keep_largest_object: 每格仅留最大块，不设最低门槛
- remove_small_objects: 统计全局面积中位数，<25%自动判定为噪声

src/cDNA_segmentation.py

@@ -203,27 +203,36 @@ def gridding(gray: np.ndarray) -> tuple:
 # 第四部分：后处理（参考choice.m, choosemaxobj.m）
 # ============================================================
 
-def remove_small_objects(binary: np.ndarray, cell_area: int = 1225, pct: float = 0.02) -> np.ndarray:
+def remove_small_objects(binary: np.ndarray) -> np.ndarray:
-    """去除面积小于 cell_area*pct 的连通域（pct默认2%）"""
+    """
-    min_size = int(cell_area * pct)
+    自动去除小连通域。
+
+    统计所有连通域的面积中位数，小于中位数 25% 的视为噪声，
+    自动剔除，不需要人设定阈值。
+    """
     labeled, num = ndimage.label(binary)
+    if num == 0:
+        return binary
+
+    # 统计每个连通域的面积
+    areas = [int(np.sum(labeled == i)) for i in range(1, num + 1)]
+    median_area = np.median(areas)
+    min_size = max(1, int(median_area * 0.25))  # 中位数的25%，最少1像素
+
     result = binary.copy()
     for i in range(1, num + 1):
-        if np.sum(labeled == i) < min_size:
+        if areas[i - 1] < min_size:
             result[labeled == i] = 0
     return result
 
 
-def keep_largest_object(binary: np.ndarray, cell_area: int = 1225, pct: float = 0.01) -> np.ndarray:
+def keep_largest_object(binary: np.ndarray) -> np.ndarray:
-    """只保留最大连通域，若最大块面积 < cell_area*pct 则整块抹黑（pct默认1%）"""
+    """每个格子里只保留面积最大的连通域（无需设定阈值）"""
-    min_size = int(cell_area * pct)
     labeled, num = ndimage.label(binary)
     if num == 0:
-        return binary
-    areas = [int(np.sum(labeled == i)) for i in range(1, num + 1)]
-    max_area = max(areas)
-    if max_area < min_size:
         return np.zeros_like(binary)
+
+    areas = [int(np.sum(labeled == i)) for i in range(1, num + 1)]
     max_idx = int(np.argmax(areas)) + 1
     return (labeled == max_idx).astype(np.uint8)
 
@@ -375,11 +384,6 @@ def main() -> None:
 
     # ---- 步骤3: 全图逐块分割（Otsu + TV去噪） ----
     print("\n[步骤3] 全图逐块分割...")
-    # 计算每个格子的面积，用于自适应后处理阈值
-    cell_h = int(np.median(np.diff(y_grid))) if len(y_grid) > 1 else 35
-    cell_w = int(np.median(np.diff(x_grid))) if len(x_grid) > 1 else 35
-    cell_area = cell_h * cell_w
-
     bw_full = np.zeros_like(gray)
     if len(x_grid) >= 2 and len(y_grid) >= 2:
         for i in range(len(y_grid) - 1):
@@ -403,12 +407,12 @@ def main() -> None:
                     bw_blk = (blk_denoised > T).astype(np.uint8)
                 except ValueError:
                     bw_blk = np.zeros(blk.shape, dtype=np.uint8)
-                # 后处理：保留最大连通域（最小面积 = 格子面积的1%）
+                # 后处理：保留最大连通域（无需阈值）
-                bw_blk = keep_largest_object(bw_blk, cell_area=cell_area)
+                bw_blk = keep_largest_object(bw_blk)
                 bw_full[r1:r2, c1:c2] = bw_blk
 
-    # 全局去小连通域（最小面积 = 格子面积的2%）
+    # 全局去除小连通域（中位数的25%以下自动判为噪声）
-    bw_full = remove_small_objects(bw_full, cell_area=cell_area)
+    bw_full = remove_small_objects(bw_full)
 
     fig_full = plot_full_segmentation(gray, bw_full, "全图逐块Otsu分割结果")
     fig_full.savefig(os.path.join(OUTPUT_DIR, 'result_full_segmentation.png'), dpi=150, bbox_inches='tight')