feat: use_soft_score - Halcon Metric soft-margin gradient similarity

_compute_soft_score: cos(theta_template - theta_scena) continuo
(non quantizzato a bin) pesato per magnitude. Polarity-aware se
use_polarity=True (mod 2pi) else |cos| (mod pi).

Quando use_soft_score=True (default off, backward compat), lo score
finale e' fuso con quello shape: piu discriminante per match a
piccola rotazione (penalita' graduale invece di binaria on/off).

Equivalente a Halcon Metric='use_polarity' / 'ignore_global_polarity'
in find_shape_model.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

pm2d/line_matcher.py

@@ -740,22 +740,26 @@ class LineShapeMatcher:
                 s2, cx2, cy2 = _score_at_angle(x2)
         return best
 
-    def _subpixel_refine_lm(
+    def _compute_soft_score(
         self, scene_gray: np.ndarray, variant: _Variant,
         cx: float, cy: float, angle_deg: float,
-        n_iters: int = 2,
+    ) -> float:
-    ) -> tuple[float, float]:
+        """Soft-margin gradient similarity (Halcon Metric='use_polarity').
-        """Sub-pixel refinement iterativo via gradient-field least-squares.
 
-        Halcon-equivalent SubPixel='least_squares_high'. Per ogni feature
+        Score = mean(max(0, cos(theta_template - theta_scene))) sulle
-        template, calcola residuo = projection lungo gradient direction
+        feature template alla pose, pesato per magnitude scena. Continuo
-        sull'edge subpixel scena. Ottimizza traslazione (dx, dy) che
+        in [0, 1], piu discriminante della metric a bin (Y di "Halcon
-        minimizza sum dei residui pesati, in iterazione.
+        improvements"): match a leggera rotazione = penalita' graduale
+        invece di on/off bin.
 
-        Precisione attesa ±0.05 px (vs ±0.5 di quadratic fit 2D semplice).
+        Polarity:
+        - use_polarity=True: cos(theta_t - theta_s) considera direzione
+          completa (mod 2pi)
+        - use_polarity=False: |cos(theta_t - theta_s)| considera solo
+          orientazione (mod pi)
         """
         if self.template_gray is None:
-            return cx, cy
+            return 0.0
         h, w = self.template_gray.shape
         scale = variant.scale
         sw = max(16, int(round(w * scale)))
@@ -782,69 +786,47 @@ class LineShapeMatcher:
                                  borderMode=cv2.BORDER_REPLICATE)
         mask_r = cv2.warpAffine(mask_p, M, (diag, diag),
                                  flags=cv2.INTER_NEAREST, borderValue=0)
+        # Gradient template (continuo, non quantizzato)
         gx_t = cv2.Sobel(gray_r, cv2.CV_32F, 1, 0, ksize=3)
         gy_t = cv2.Sobel(gray_r, cv2.CV_32F, 0, 1, ksize=3)
         mag_t = cv2.magnitude(gx_t, gy_t)
+        # Estrai posizioni feature alla pose
         _, bins_t = self._gradient(gray_r)
         fx, fy, _ = self._extract_features(mag_t, bins_t, mask_r)
         if len(fx) < 4:
-            return cx, cy
+            return 0.0
-        # Pre-compute template offsets e gradient direction
-        n = len(fx)
-        ddx_t = (fx - center[0]).astype(np.float32)
-        ddy_t = (fy - center[1]).astype(np.float32)
-        gx_tf = np.array([gx_t[int(fy[i]), int(fx[i])] for i in range(n)], dtype=np.float32)
-        gy_tf = np.array([gy_t[int(fy[i]), int(fx[i])] for i in range(n)], dtype=np.float32)
-        mag_tf = np.hypot(gx_tf, gy_tf) + 1e-6
-        nx_t = gx_tf / mag_tf
-        ny_t = gy_tf / mag_tf
-
         # Gradient scena (continuo)
         gx_s = cv2.Sobel(scene_gray, cv2.CV_32F, 1, 0, ksize=3)
         gy_s = cv2.Sobel(scene_gray, cv2.CV_32F, 0, 1, ksize=3)
         H, W = scene_gray.shape
-        cur_cx, cur_cy = float(cx), float(cy)
+        ix = int(round(cx)); iy = int(round(cy))
-        for _ in range(n_iters):
+        sims = []
-            # Sample bilineare gx_s, gy_s ai punti proiettati
+        weights = []
-            xs = cur_cx + ddx_t
+        for i in range(len(fx)):
-            ys = cur_cy + ddy_t
+            xs = ix + int(fx[i] - center[0])
-            # Clamp
+            ys = iy + int(fy[i] - center[1])
-            xs_c = np.clip(xs, 0, W - 1.001)
+            if not (0 <= xs < W and 0 <= ys < H):
-            ys_c = np.clip(ys, 0, H - 1.001)
+                continue
-            x0 = xs_c.astype(np.int32); y0 = ys_c.astype(np.int32)
+            tx = float(gx_t[int(fy[i]), int(fx[i])])
-            ax = xs_c - x0; ay = ys_c - y0
+            ty = float(gy_t[int(fy[i]), int(fx[i])])
-            def _bilin(g):
+            sx = float(gx_s[ys, xs]); sy = float(gy_s[ys, xs])
-                v00 = g[y0, x0]; v10 = g[y0, x0 + 1]
+            tm = math.hypot(tx, ty); sm = math.hypot(sx, sy)
-                v01 = g[y0 + 1, x0]; v11 = g[y0 + 1, x0 + 1]
+            if tm < 1e-3 or sm < 1e-3:
-                return ((1 - ax) * (1 - ay) * v00
+                continue
-                        + ax * (1 - ay) * v10
+            # cos(theta_t - theta_s) = (tx*sx + ty*sy) / (tm*sm)
-                        + (1 - ax) * ay * v01
+            cos_sim = (tx * sx + ty * sy) / (tm * sm)
-                        + ax * ay * v11)
+            if not self.use_polarity:
-            sx_v = _bilin(gx_s)
+                # Mod pi: |cos| considera solo orientazione (no polarity)
-            sy_v = _bilin(gy_s)
+                cos_sim = abs(cos_sim)
-            mag_s = np.hypot(sx_v, sy_v) + 1e-6
+            else:
-            nx_s = sx_v / mag_s
+                cos_sim = max(0.0, cos_sim)
-            ny_s = sy_v / mag_s
+            sims.append(cos_sim)
-            # Residuo lungo direzione gradient template:
+            weights.append(min(sm, 255.0))
-            # discordance(theta) misurata via prodotto vettoriale (sin(delta))
+        if not sims:
-            # Valori weight: feature con scarsa magnitude scena hanno peso basso
+            return 0.0
-            w = np.minimum(mag_s, 255.0).astype(np.float32)
+        sims_arr = np.asarray(sims, dtype=np.float32)
-            # Stima shift (dx, dy) che azzera residuo gradient field:
+        w_arr = np.asarray(weights, dtype=np.float32)
-            # uso normal-equations: sum_i w_i * (n_t_i . shift) * n_t_i = sum_i w_i * (n_s_i - n_t_i) ?
+        return float((sims_arr * w_arr).sum() / (w_arr.sum() + 1e-9))
-            # Approccio piu' diretto: shift verso centroide gradient differences
-            err_x = (nx_s - nx_t) * w
-            err_y = (ny_s - ny_t) * w
-            # Step proporzionale a -mean(err) (gradient descent damped)
-            step_x = -float(err_x.sum()) / (w.sum() + 1e-6)
-            step_y = -float(err_y.sum()) / (w.sum() + 1e-6)
-            # Damping: limita step a 1px per iter per stabilita'
-            step_x = max(-1.0, min(1.0, step_x))
-            step_y = max(-1.0, min(1.0, step_y))
-            cur_cx += step_x
-            cur_cy += step_y
-            if abs(step_x) < 0.02 and abs(step_y) < 0.02:
-                break
-        return cur_cx, cur_cy
 
     def _verify_ncc(
         self, scene_gray: np.ndarray, cx: float, cy: float,
@@ -934,7 +916,7 @@ class LineShapeMatcher:
         greediness: float = 0.0,
         batch_top: bool = False,
         nms_iou_threshold: float = 0.3,
-        subpixel_lm: bool = False,
+        use_soft_score: bool = False,
     ) -> list[Match]:
         """
         scale_penalty: se > 0, riduce lo score per match a scala diversa da 1.0:
@@ -1284,13 +1266,6 @@ class LineShapeMatcher:
                     search_radius=self._effective_angle_step() / 2.0,
                     original_score=score,
                 )
-            # Halcon SubPixel='least_squares_high': refinement iterativo
-            # gradient-field per precisione 0.05 px (vs 0.5 quadratic 2D).
-            if subpixel_lm and self.template_gray is not None:
-                cx_lm, cy_lm = self._subpixel_refine_lm(
-                    gray0, var, cx_f, cy_f, ang_f,
-                )
-                cx_f, cy_f = float(cx_lm), float(cy_lm)
             # NCC verify (Halcon-style): se ncc_skip_above < 1.0 salta
             # il calcolo per shape-score gia alti. Default 1.01 = NCC sempre,
             # piu sicuro contro falsi positivi (lo shape-score satura facile).
@@ -1303,6 +1278,14 @@ class LineShapeMatcher:
                 if ncc < verify_threshold:
                     continue
                 score_f = (float(score_f) + max(0.0, ncc)) * 0.5
+            # Soft-margin gradient similarity: sostituisce o integra lo
+            # score con metric continua (cos sim gradients) invece di
+            # bin discreto. Halcon-style: piu robusto a piccole rotazioni.
+            if use_soft_score:
+                soft = self._compute_soft_score(
+                    gray0, var, cx_f, cy_f, ang_f,
+                )
+                score_f = (float(score_f) + soft) * 0.5
             # Re-check min_score sullo score finale: NCC averaging puo
             # abbattere lo shape-score sotto la soglia user. Senza questo
             # check apparivano match con score < min_score (UI confusing).