feat: PM2D standalone shape-based matcher

Programma standalone Pattern Matching 2D con GUI cv2/tk + algoritmo
puro riusabile. Due backend:

- LineShapeMatcher (default): porting Python di line2Dup (linemod-style)
  - Gradient orientation quantized 8-bin modulo π + spreading
  - Feature sparse top-magnitude con spacing minimo
  - Score via shift-add vettorizzato numpy (O(N_features·H·W))
  - Piramide multi-risoluzione con pruning varianti al top-level
  - Supporto mask binaria per modello non-rettangolare

- EdgeShapeMatcher (fallback): Canny + matchTemplate multi-rotazione

GUI separata da algoritmo. Benchmark clip.png (13 istanze):
  - Edge backend:  84s, 6/13 score ~0.3
  - Line backend:  4.1s, 13/13 score 0.98-1.00

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-04-24 00:46:59 +02:00
commit b9a4d51fac
14 changed files with 2499 additions and 0 deletions
+320
View File
@@ -0,0 +1,320 @@
"""Pattern Matching 2D shape-based via edge template matching multi-rotazione/scala.
Algoritmo equivalente a Fase Alpha del documento tecnico Vision Suite:
- Estrazione edge Canny dal template (invarianza illuminazione)
- Generazione varianti del template edge per ogni (angolo, scala)
- matchTemplate NCC sulla scena edge per ogni variante
- Picchi locali con NMS spaziale per multi-istanza
Uso: vedi `EdgeShapeMatcher.train` e `EdgeShapeMatcher.find`.
"""
from __future__ import annotations
from dataclasses import dataclass
import cv2
import numpy as np
@dataclass
class Match:
"""Singola istanza trovata nella scena."""
cx: float # baricentro x [px] nella scena
cy: float # baricentro y [px] nella scena
angle_deg: float # rotazione [0, 360)
scale: float # fattore scala (1.0 = template originale)
score: float # similarità NCC [0, 1]
bbox: tuple[int, int, int, int] # x, y, w, h del template ruotato/scalato
@dataclass
class Template:
"""Variante precomputata del template a un dato (angolo, scala)."""
angle_deg: float
scale: float
edge: np.ndarray # immagine edge ruotata+scalata (uint8 0/255)
mask: np.ndarray # maschera supporto (uint8 0/255)
cx_local: float # baricentro nel sistema locale variante
cy_local: float
class EdgeShapeMatcher:
"""Matcher shape-based su edge Canny con rotazione + scala precomputate."""
def __init__(
self,
canny_low: int = 50,
canny_high: int = 150,
angle_step_deg: float = 5.0,
angle_range_deg: tuple[float, float] = (0.0, 360.0),
scale_range: tuple[float, float] = (1.0, 1.0),
scale_step: float = 0.1,
match_method: int = cv2.TM_CCOEFF_NORMED,
pyramid_levels: int = 3,
top_score_factor: float = 0.6,
) -> None:
self.canny_low = canny_low
self.canny_high = canny_high
self.angle_step_deg = angle_step_deg
self.angle_range_deg = angle_range_deg
self.scale_range = scale_range
self.scale_step = scale_step
self.match_method = match_method
self.pyramid_levels = max(1, pyramid_levels)
self.top_score_factor = top_score_factor
self.templates: list[Template] = []
self.template_size: tuple[int, int] = (0, 0) # w, h originale
@staticmethod
def _to_gray(img: np.ndarray) -> np.ndarray:
if img.ndim == 3:
return cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
return img
def _edges(self, gray: np.ndarray) -> np.ndarray:
return cv2.Canny(gray, self.canny_low, self.canny_high)
def _scale_list(self) -> list[float]:
s0, s1 = self.scale_range
if s0 >= s1 or self.scale_step <= 0:
return [float(s0)]
n = int(np.floor((s1 - s0) / self.scale_step)) + 1
return [float(s0 + i * self.scale_step) for i in range(n)]
def _angle_list(self) -> list[float]:
a0, a1 = self.angle_range_deg
if self.angle_step_deg <= 0 or a0 >= a1:
return [float(a0)]
n = int(np.floor((a1 - a0) / self.angle_step_deg))
return [float(a0 + i * self.angle_step_deg) for i in range(n)]
def train(self, template_bgr: np.ndarray) -> int:
"""Genera varianti per tutte le combinazioni (angolo, scala)."""
gray = self._to_gray(template_bgr)
h, w = gray.shape
self.template_size = (w, h)
edge_orig = self._edges(gray)
mask_orig = np.full((h, w), 255, dtype=np.uint8)
self.templates.clear()
scales = self._scale_list()
angles = self._angle_list()
for s in scales:
sw = max(8, int(round(w * s)))
sh = max(8, int(round(h * s)))
edge_s = cv2.resize(edge_orig, (sw, sh), interpolation=cv2.INTER_LINEAR)
mask_s = cv2.resize(mask_orig, (sw, sh), interpolation=cv2.INTER_NEAREST)
# Re-thresh dopo resize
_, edge_s = cv2.threshold(edge_s, 64, 255, cv2.THRESH_BINARY)
# Padding diagonale per rotazione senza cropping
diag = int(np.ceil(np.hypot(sh, sw))) + 4
pad_y = (diag - sh) // 2
pad_x = (diag - sw) // 2
edge_p = cv2.copyMakeBorder(
edge_s, pad_y, diag - sh - pad_y, pad_x, diag - sw - pad_x,
cv2.BORDER_CONSTANT, value=0,
)
mask_p = cv2.copyMakeBorder(
mask_s, pad_y, diag - sh - pad_y, pad_x, diag - sw - pad_x,
cv2.BORDER_CONSTANT, value=0,
)
center = (diag / 2.0, diag / 2.0)
for ang in angles:
M = cv2.getRotationMatrix2D(center, ang, 1.0)
edge_r = cv2.warpAffine(
edge_p, M, (diag, diag),
flags=cv2.INTER_LINEAR, borderValue=0,
)
mask_r = cv2.warpAffine(
mask_p, M, (diag, diag),
flags=cv2.INTER_NEAREST, borderValue=0,
)
# Crop minimo bounding mask
ys, xs = np.where(mask_r > 0)
if len(xs) == 0:
continue
x0, x1 = xs.min(), xs.max() + 1
y0, y1 = ys.min(), ys.max() + 1
edge_c = edge_r[y0:y1, x0:x1]
mask_c = mask_r[y0:y1, x0:x1]
cx_local = (mask_c.shape[1] - 1) / 2.0
cy_local = (mask_c.shape[0] - 1) / 2.0
self.templates.append(
Template(
angle_deg=float(ang),
scale=float(s),
edge=edge_c,
mask=mask_c,
cx_local=cx_local,
cy_local=cy_local,
)
)
return len(self.templates)
def _pyrdown_binary(self, img: np.ndarray) -> np.ndarray:
"""pyrDown + re-thresh per mantenere binario 0/255."""
d = cv2.pyrDown(img)
_, d = cv2.threshold(d, 32, 255, cv2.THRESH_BINARY)
return d
def find(
self,
scene_bgr: np.ndarray,
min_score: float = 0.5,
max_matches: int = 10,
nms_radius: int | None = None,
) -> list[Match]:
"""Cerca istanze del template nella scena con strategia piramidale.
- Top-level: matching brute-force a bassa risoluzione (veloce, soglia ridotta)
- Refinement: re-match locale a risoluzione piena per ogni candidato
"""
if not self.templates:
raise RuntimeError("Matcher non addestrato: chiamare train() prima.")
gray = self._to_gray(scene_bgr)
scene_edge0 = self._edges(gray)
# Piramide scena edge
scene_pyr = [scene_edge0]
for _ in range(self.pyramid_levels - 1):
scene_pyr.append(self._pyrdown_binary(scene_pyr[-1]))
top = len(scene_pyr) - 1
sf = 2 ** top # scale factor top→0
scene_top = scene_pyr[top]
if nms_radius is None:
nms_radius = max(8, min(self.template_size) // 2)
top_thresh = min_score * self.top_score_factor
# Top-level brute-force
candidates: list[tuple[float, int, int, int]] = []
for ti, tpl in enumerate(self.templates):
edge_top = tpl.edge.copy()
mask_top = tpl.mask.copy()
for _ in range(top):
edge_top = self._pyrdown_binary(edge_top)
mask_top = self._pyrdown_binary(mask_top)
th, tw = edge_top.shape
if th < 6 or tw < 6:
continue
if scene_top.shape[0] < th or scene_top.shape[1] < tw:
continue
res = cv2.matchTemplate(
scene_top, edge_top, self.match_method, mask=mask_top,
)
res = np.nan_to_num(res, nan=-1.0, posinf=-1.0, neginf=-1.0)
ys, xs = np.where(res >= top_thresh)
for y, x in zip(ys, xs):
candidates.append((float(res[y, x]), int(x), int(y), ti))
# Refinement a risoluzione piena: per ogni candidato top, finestra locale
refined: list[tuple[float, int, int, int]] = []
margin = sf + 4
for _, xt, yt, ti in candidates:
tpl = self.templates[ti]
th, tw = tpl.edge.shape
x0 = xt * sf
y0 = yt * sf
sx0 = max(0, x0 - margin)
sy0 = max(0, y0 - margin)
sx1 = min(scene_edge0.shape[1], x0 + tw + margin)
sy1 = min(scene_edge0.shape[0], y0 + th + margin)
roi = scene_edge0[sy0:sy1, sx0:sx1]
if roi.shape[0] < th or roi.shape[1] < tw:
continue
res = cv2.matchTemplate(
roi, tpl.edge, self.match_method, mask=tpl.mask,
)
res = np.nan_to_num(res, nan=-1.0, posinf=-1.0, neginf=-1.0)
_, max_val, _, max_loc = cv2.minMaxLoc(res)
if max_val < min_score:
continue
bx = sx0 + max_loc[0]
by = sy0 + max_loc[1]
refined.append((float(max_val), bx, by, ti))
refined.sort(key=lambda c: -c[0])
# NMS spaziale baricentri
kept: list[Match] = []
r2 = nms_radius * nms_radius
for score, x, y, ti in refined:
tpl = self.templates[ti]
cx = x + tpl.cx_local
cy = y + tpl.cy_local
if any((k.cx - cx) ** 2 + (k.cy - cy) ** 2 < r2 for k in kept):
continue
th, tw = tpl.edge.shape
kept.append(
Match(
cx=cx, cy=cy,
angle_deg=tpl.angle_deg,
scale=tpl.scale,
score=score,
bbox=(x, y, tw, th),
)
)
if len(kept) >= max_matches:
break
return kept
# --- Persistenza modello ---
def save(self, path: str) -> None:
"""Salva matcher su disco (.npz)."""
meta = np.array(
[(t.angle_deg, t.scale, t.cx_local, t.cy_local) for t in self.templates],
dtype=np.float32,
)
params = np.array(
[self.canny_low, self.canny_high, self.angle_step_deg,
self.angle_range_deg[0], self.angle_range_deg[1],
self.scale_range[0], self.scale_range[1], self.scale_step,
self.template_size[0], self.template_size[1], self.match_method,
self.pyramid_levels, self.top_score_factor],
dtype=np.float32,
)
arrays = {f"edge_{i}": t.edge for i, t in enumerate(self.templates)}
arrays.update({f"mask_{i}": t.mask for i, t in enumerate(self.templates)})
np.savez_compressed(path, params=params, meta=meta, **arrays)
@classmethod
def load(cls, path: str) -> "EdgeShapeMatcher":
z = np.load(path)
p = z["params"]
m = cls(
canny_low=int(p[0]),
canny_high=int(p[1]),
angle_step_deg=float(p[2]),
angle_range_deg=(float(p[3]), float(p[4])),
scale_range=(float(p[5]), float(p[6])),
scale_step=float(p[7]),
match_method=int(p[10]),
pyramid_levels=int(p[11]) if len(p) > 11 else 3,
top_score_factor=float(p[12]) if len(p) > 12 else 0.6,
)
m.template_size = (int(p[8]), int(p[9]))
meta = z["meta"]
for i in range(len(meta)):
m.templates.append(
Template(
angle_deg=float(meta[i, 0]),
scale=float(meta[i, 1]),
edge=z[f"edge_{i}"],
mask=z[f"mask_{i}"],
cx_local=float(meta[i, 2]),
cy_local=float(meta[i, 3]),
)
)
return m