setiastrosuitepro 1.6.7__py3-none-any.whl

setiastro/saspro/rgbalign.py

@@ -0,0 +1,727 @@
+# pro/rgbalign.py
+from __future__ import annotations
+
+import os
+import numpy as np
+
+from PyQt6.QtCore import Qt, QThread, pyqtSignal
+from PyQt6.QtWidgets import (
+    QDialog, QVBoxLayout, QHBoxLayout, QLabel, QPushButton, QApplication,
+    QComboBox, QCheckBox, QMessageBox, QProgressBar, QPlainTextEdit, QSpinBox
+)
+
+
+import astroalign
+
+import sep
+
+import cv2
+
+# try to reuse poly from star_alignment if present
+try:
+    from setiastro.saspro.star_alignment import PolynomialTransform
+except Exception:
+    PolynomialTransform = None
+
+
+# ─────────────────────────────────────────────────────────────────────
+# Worker
+# ─────────────────────────────────────────────────────────────────────
+class RGBAlignWorker(QThread):
+    progress = pyqtSignal(int, str)     # (percent, message)
+    done = pyqtSignal(np.ndarray)       # aligned RGB image
+    failed = pyqtSignal(str)
+    EDGE_FRAC = 0.55      # 55% of max radius
+    MIN_EDGE_PTS = 6      # want at least 6 out there
+    EDGE_INNER_FRAC = 0.38   # toss center 38% radius
+    MATCH_MAX_DIST = 10.0    # px, generous for CA
+    MIN_MATCHES = 6
+
+
+    def __init__(self, img: np.ndarray, model: str, sep_sigma: float = 3.0):
+        super().__init__()
+        self.img = img
+        self.model = model
+        self.sep_sigma = float(sep_sigma)
+
+        self.r_xform = None
+        self.b_xform = None
+        self.r_pairs = None
+        self.b_pairs = None
+
+    def _pts_too_central(self, pts: np.ndarray | None, shape) -> bool:
+        """
+        Return True if the matched points are all bunched near the center.
+        pts: (N, 2) in x,y
+        shape: (H, W)
+        """
+        if pts is None or len(pts) == 0:
+            return True
+        h, w = shape[:2]
+        cx, cy = w * 0.5, h * 0.5
+        # distance of each point from center
+        r = np.hypot(pts[:, 0] - cx, pts[:, 1] - cy)
+        rmax = np.hypot(cx, cy)  # radius to corner
+        edge_mask = r > (self.EDGE_FRAC * rmax)
+        return edge_mask.sum() < self.MIN_EDGE_PTS
+
+    def _sep_detect_points(self, img: np.ndarray):
+        """Return (N,2) points from SEP, brightest first, using user sigma."""
+        if sep is None:
+            return None
+        data = img.astype(np.float32, copy=False)
+        bkg = sep.Background(data)
+        data_sub = data - bkg
+        # use the promoted sigma here 👇
+        objs = sep.extract(data_sub, self.sep_sigma, err=bkg.globalrms)
+        if objs is None or len(objs) == 0:
+            return None
+        idx = np.argsort(objs["peak"])[::-1]
+        pts = np.stack([objs["x"][idx], objs["y"][idx]], axis=1)
+        return pts
+
+    def _filter_edge_ring(self, pts: np.ndarray, shape, inner_frac=EDGE_INNER_FRAC):
+        """Keep only points outside inner_frac * Rmax."""
+        if pts is None or pts.size == 0:
+            return None
+        h, w = shape[:2]
+        cx, cy = w * 0.5, h * 0.5
+        r = np.hypot(pts[:,0] - cx, pts[:,1] - cy)
+        rmax = np.hypot(cx, cy)
+        mask = r >= (inner_frac * rmax)
+        pts_edge = pts[mask]
+        return pts_edge if pts_edge.size else None
+
+    def _pair_edge_points(self, src_img, ref_img, shape):
+        """Detect in BOTH images, keep only edge ring in REF, then NN-match in SRC."""
+        ref_pts = self._sep_detect_points(ref_img)
+        src_pts = self._sep_detect_points(src_img)
+        if ref_pts is None or src_pts is None:
+            return None, None
+
+        ref_edge = self._filter_edge_ring(ref_pts, shape)
+        if ref_edge is None:
+            return None, None
+
+        # brute-force NN, small N, so ok
+        src_arr = np.asarray(src_pts, dtype=np.float32)
+        pairs_src = []
+        pairs_dst = []
+        for (x_ref, y_ref) in ref_edge:
+            dxy = src_arr - np.array([x_ref, y_ref], dtype=np.float32)
+            dist = np.hypot(dxy[:,0], dxy[:,1])
+            j = np.argmin(dist)
+            if dist[j] <= self.MATCH_MAX_DIST:
+                # src point is in the channel we want to warp → source
+                pairs_src.append(src_arr[j])
+                # ref point is the green channel → destination
+                pairs_dst.append([x_ref, y_ref])
+
+        if len(pairs_src) < self.MIN_MATCHES:
+            return None, None
+
+        return (np.array(pairs_src, dtype=np.float32),
+                np.array(pairs_dst, dtype=np.float32))
+
+
+    def run(self):
+        if self.img is None or self.img.ndim != 3 or self.img.shape[2] < 3:
+            self.failed.emit("Image must be RGB (3 channels).")
+            return
+        if astroalign is None:
+            self.failed.emit("astroalign is not available.")
+            return
+
+        try:
+            self.progress.emit(5, "Preparing channels…")
+            R = np.ascontiguousarray(self.img[..., 0].astype(np.float32, copy=False))
+            G = np.ascontiguousarray(self.img[..., 1].astype(np.float32, copy=False))
+            B = np.ascontiguousarray(self.img[..., 2].astype(np.float32, copy=False))
+
+            # R → G
+            self.progress.emit(15, "Aligning Red → Green…")
+            kind_R, X_R, (r_src, r_dst) = self._estimate_transform(R, G, self.model)
+            self.r_xform = (kind_R, X_R)
+            self.r_pairs = (r_src, r_dst)
+            self.progress.emit(35, f"Red transform = {kind_R}")
+            R_aligned = self._warp_channel(R, kind_R, X_R, G.shape)
+
+            # B → G
+            self.progress.emit(55, "Aligning Blue → Green…")
+            kind_B, X_B, (b_src, b_dst) = self._estimate_transform(B, G, self.model)
+            self.b_xform = (kind_B, X_B)
+            self.b_pairs = (b_src, b_dst)
+            self.progress.emit(75, f"Blue transform = {kind_B}")
+            B_aligned = self._warp_channel(B, kind_B, X_B, G.shape)
+
+            out = np.stack([R_aligned, G, B_aligned], axis=2).astype(self.img.dtype, copy=False)
+            self.progress.emit(100, "Done.")
+            self.done.emit(out)
+        except Exception as e:
+            self.failed.emit(str(e))
+
+
+    # ───── helpers (basically mini versions of your big star alignment logic) ─────
+    def _estimate_transform(self, src: np.ndarray, ref: np.ndarray, model: str):
+        H, W = ref.shape[:2]
+
+        # ── 0) edge-only, SEP-based path ─────────────────────────────
+        if model == "edge-sep":
+            src_xy, dst_xy = self._pair_edge_points(src, ref, (H, W))
+            if src_xy is not None and dst_xy is not None and cv2 is not None:
+                # 0a) try homography first (better for corner warp)
+                Hh, inliers = cv2.findHomography(
+                    src_xy, dst_xy,
+                    method=cv2.RANSAC,
+                    ransacReprojThreshold=2.5,
+                    maxIters=2000,
+                    confidence=0.999,
+                )
+                if Hh is not None:
+                    return ("homography", Hh, (src_xy, dst_xy))
+
+                # 0b) fallback → affine
+                A, inliers = cv2.estimateAffine2D(
+                    src_xy, dst_xy,
+                    method=cv2.RANSAC,
+                    ransacReprojThreshold=2.5,
+                    maxIters=2000,
+                    confidence=0.999,
+                )
+                if A is not None:
+                    return ("affine", A, (src_xy, dst_xy))
+            # if SEP failed or cv2 missing → fall through to astroalign normal path
+
+        # ─────────────────────────────────────────────────────────────
+        # 1) astroalign normal pass
+        # ─────────────────────────────────────────────────────────────
+        tform, (src_pts, dst_pts) = astroalign.find_transform(
+            np.ascontiguousarray(src),
+            np.ascontiguousarray(ref),
+            max_control_points=50,
+            detection_sigma=5.0,
+            min_area=5,
+        )
+
+        # 2) 'hungry' pass if too central
+        if self._pts_too_central(dst_pts, ref.shape):
+            tform2, (src_pts2, dst_pts2) = astroalign.find_transform(
+                np.ascontiguousarray(src),
+                np.ascontiguousarray(ref),
+                max_control_points=120,
+                detection_sigma=3.0,
+                min_area=3,
+            )
+            if not self._pts_too_central(dst_pts2, ref.shape):
+                tform, src_pts, dst_pts = tform2, src_pts2, dst_pts2
+
+        # 3) original branching
+        P = np.asarray(tform.params, dtype=np.float64)
+        src_xy = np.asarray(src_pts, dtype=np.float32)
+        dst_xy = np.asarray(dst_pts, dtype=np.float32)
+
+        # affine
+        if model == "affine":
+            if cv2 is None:
+                return ("affine", P[0:2, :], (src_xy, dst_xy))
+            A, _ = cv2.estimateAffine2D(
+                src_xy, dst_xy, method=cv2.RANSAC, ransacReprojThreshold=3.0
+            )
+            if A is None:
+                return ("affine", P[0:2, :], (src_xy, dst_xy))
+            return ("affine", A, (src_xy, dst_xy))
+
+        # homography
+        if model == "homography":
+            if cv2 is None:
+                if P.shape == (3, 3):
+                    return ("homography", P, (src_xy, dst_xy))
+                A3 = np.vstack([P[0:2, :], [0, 0, 1]])
+                return ("homography", A3, (src_xy, dst_xy))
+            Hh, _ = cv2.findHomography(
+                src_xy, dst_xy, method=cv2.RANSAC, ransacReprojThreshold=3.0
+            )
+            if Hh is None:
+                if P.shape == (3, 3):
+                    return ("homography", P, (src_xy, dst_xy))
+                A3 = np.vstack([P[0:2, :], [0, 0, 1]])
+                return ("homography", A3, (src_xy, dst_xy))
+            return ("homography", Hh, (src_xy, dst_xy))
+
+        # poly3 / poly4
+        if model in ("poly3", "poly4") and PolynomialTransform is not None and cv2 is not None:
+            order = 3 if model == "poly3" else 4
+            scale_vec = np.array([W, H], dtype=np.float32)
+            src_n = src_xy / scale_vec
+            dst_n = dst_xy / scale_vec
+
+            t_poly = PolynomialTransform()
+            ok = t_poly.estimate(dst_n, src_n, order=order)  # dst → src
+            if not ok:
+                Hh, _ = cv2.findHomography(
+                    src_xy, dst_xy, method=cv2.RANSAC, ransacReprojThreshold=3.0
+                )
+                return ("homography", Hh, (src_xy, dst_xy))
+
+            def _warp_poly(img: np.ndarray, out_shape: tuple[int, int]):
+                Hh_, Ww_ = out_shape
+                yy, xx = np.mgrid[0:Hh_, 0:Ww_].astype(np.float32)
+                coords = np.stack([xx, yy], axis=-1).reshape(-1, 2)
+                coords_n = coords / scale_vec
+                mapped_n = t_poly(coords_n)
+                mapped = mapped_n * scale_vec
+                map_x = mapped[:, 0].reshape(Hh_, Ww_).astype(np.float32)
+                map_y = mapped[:, 1].reshape(Hh_, Ww_).astype(np.float32)
+                return cv2.remap(
+                    img, map_x, map_y,
+                    interpolation=cv2.INTER_LANCZOS4,
+                    borderMode=cv2.BORDER_CONSTANT, borderValue=0
+                )
+
+            return (model, _warp_poly, (src_xy, dst_xy))
+
+        # fallback → homography
+        if cv2 is None:
+            if P.shape == (3, 3):
+                return ("homography", P, (src_xy, dst_xy))
+            A3 = np.vstack([P[0:2, :], [0, 0, 1]])
+            return ("homography", A3, (src_xy, dst_xy))
+
+        Hh, _ = cv2.findHomography(
+            src_xy, dst_xy, method=cv2.RANSAC, ransacReprojThreshold=3.0
+        )
+        return ("homography", Hh, (src_xy, dst_xy))
+
+
+
+    def _pick_edge_stars_with_sep(self, img, tiles=(3,3), per_tile=2):
+        if sep is None:
+            return []
+        data = img.astype(np.float32, copy=False)
+        bkg = sep.Background(data)
+        data_sub = data - bkg
+        objs = sep.extract(data_sub, 1.5, err=bkg.globalrms)
+        H, W = data.shape[:2]
+        th, tw = H // tiles[0], W // tiles[1]
+        picked = []
+        for ty in range(tiles[0]):
+            for tx in range(tiles[1]):
+                y0, y1 = ty*th, min((ty+1)*th, H)
+                x0, x1 = tx*tw, min((tx+1)*tw, W)
+                box = objs[
+                    (objs['y'] >= y0) & (objs['y'] < y1) &
+                    (objs['x'] >= x0) & (objs['x'] < x1)
+                ]
+                if len(box) == 0:
+                    continue
+                # brightest first
+                box = box[np.argsort(box['peak'])][::-1][:per_tile]
+                for o in box:
+                    picked.append((float(o['x']), float(o['y'])))
+        return picked
+
+
+    def _warp_channel(self, ch: np.ndarray, kind: str, X, ref_shape):
+        H, W = ref_shape[:2]
+        if kind == "affine":
+            # Just assume cv2 is available (standard dependency) for perf
+            A = np.asarray(X, dtype=np.float32).reshape(2, 3)
+            return cv2.warpAffine(ch, A, (W, H), flags=cv2.INTER_LANCZOS4,
+                                  borderMode=cv2.BORDER_CONSTANT, borderValue=0)
+
+        if kind == "homography":
+            Hm = np.asarray(X, dtype=np.float32).reshape(3, 3)
+            return cv2.warpPerspective(ch, Hm, (W, H), flags=cv2.INTER_LANCZOS4,
+                                       borderMode=cv2.BORDER_CONSTANT, borderValue=0)
+
+        if kind.startswith("poly"):
+            return X(ch, (H, W))
+
+        return ch
+
+
+# ─────────────────────────────────────────────────────────────────────
+# Dialog
+# ─────────────────────────────────────────────────────────────────────
+class RGBAlignDialog(QDialog):
+    def __init__(self, parent=None, document=None):
+        super().__init__(parent)
+        self.setWindowTitle(self.tr("RGB Align"))
+        self.setWindowFlag(Qt.WindowType.Window, True)
+        self.setWindowModality(Qt.WindowModality.NonModal)
+        self.setModal(False)
+        try:
+            self.setAttribute(Qt.WidgetAttribute.WA_DeleteOnClose, True)
+        except Exception:
+            pass  # older PyQt6 versions        
+        self.parent = parent
+        # document could be a view; try to unwrap
+        self.doc_view = document
+        self.doc = getattr(document, "document", document)
+        self.image = getattr(self.doc, "image", None) if self.doc is not None else None
+
+        lay = QVBoxLayout(self)
+        lay.addWidget(QLabel("Align R and B channels to G.\n"
+                             "Select model and run."))
+
+        hl = QHBoxLayout()
+        hl.addWidget(QLabel(self.tr("Alignment model:")))
+        self.model_combo = QComboBox()
+        self.model_combo.addItems([
+            "EDGE",           # ← first, new default
+            "Homography",
+            "Affine",
+            "Poly 3",
+            "Poly 4",
+        ])
+        self.model_combo.setCurrentIndex(0)
+
+        # tooltips for each mode
+        self.model_combo.setItemData(
+            0,
+            (
+                "EDGE (Edge-Detected Guided Estimator)\n"
+                "• Detect stars in both channels with SEP\n"
+                "• Keep only outer-ring stars (ignore center)\n"
+                "• Try homography first for corner CA\n"
+                "• If homography fails → try affine\n"
+                "• If that fails → fall back to astroalign"
+            ),
+            Qt.ItemDataRole.ToolTipRole,
+        )
+        self.model_combo.setItemData(
+            1,
+            "Standard homography using astroalign matches (good general-purpose choice).",
+            Qt.ItemDataRole.ToolTipRole,
+        )
+        self.model_combo.setItemData(
+            2,
+            "Affine (shift + scale + rotate + shear). Good when channels are mostly parallel.",
+            Qt.ItemDataRole.ToolTipRole,
+        )
+        self.model_combo.setItemData(
+            3,
+            "Polynomial (order 3). Use when you have mild field distortion.",
+            Qt.ItemDataRole.ToolTipRole,
+        )
+        self.model_combo.setItemData(
+            4,
+            "Polynomial (order 4). Use for stronger distortion, but needs more/better matches.",
+            Qt.ItemDataRole.ToolTipRole,
+        )
+        hl.addWidget(self.model_combo)
+        lay.addLayout(hl)
+
+        # ── SEP controls ─────────────────────────
+        sep_row = QHBoxLayout()
+        sep_row.addWidget(QLabel(self.tr("SEP sigma:")))
+
+        self.sep_spin = QSpinBox()
+        self.sep_spin.setRange(1, 100)
+        self.sep_spin.setValue(5)   # default; 1.5 was too hungry
+        self.sep_spin.setToolTip("Detection threshold (σ) for SEP star finding in EDGE mode.\n"
+                                 "Higher = fewer stars, lower = more stars.")
+        sep_row.addWidget(self.sep_spin)
+
+        self.btn_trial_sep = QPushButton(self.tr("Trial detect stars"))
+        self.btn_trial_sep.setToolTip("Run SEP on the green channel with this sigma and report how many "
+                                      "stars it finds and how many are in the EDGE ring.")
+        self.btn_trial_sep.clicked.connect(self._trial_sep_detect)
+        sep_row.addWidget(self.btn_trial_sep)
+
+        lay.addLayout(sep_row)
+
+
+        self.chk_new_doc = QCheckBox(self.tr("Create new document (keep original)"))
+        self.chk_new_doc.setChecked(True)
+        lay.addWidget(self.chk_new_doc)
+
+        # progress
+        self.progress_label = QLabel("Idle.")
+        self.progress_bar = QProgressBar()
+        self.progress_bar.setRange(0, 100)
+        self.progress_bar.setValue(0)
+        lay.addWidget(self.progress_label)
+        lay.addWidget(self.progress_bar)
+
+        self.summary_box = QPlainTextEdit()
+        self.summary_box.setReadOnly(True)
+        self.summary_box.setPlaceholderText("Transform summary will appear here…")
+        self.summary_box.setMinimumHeight(140)
+        # optional: monospace
+        self.summary_box.setStyleSheet("font-family: Consolas, 'Courier New', monospace; font-size: 11px;")
+        lay.addWidget(self.summary_box)
+
+        btns = QHBoxLayout()
+        self.btn_run = QPushButton(self.tr("Align"))
+        self.btn_close = QPushButton(self.tr("Close"))
+        btns.addWidget(self.btn_run)
+        btns.addWidget(self.btn_close)
+        lay.addLayout(btns)
+
+        self.btn_run.clicked.connect(self._start_align)
+        self.btn_close.clicked.connect(self.close)
+
+        self.worker: RGBAlignWorker | None = None
+
+    def _trial_sep_detect(self):
+        if self.image is None:
+            QMessageBox.warning(self, "RGB Align", "No image loaded.")
+            return
+        if sep is None:
+            QMessageBox.warning(self, "RGB Align", "python-sep is not available.")
+            return
+        self.progress_label.setText(f"Trial Detection In Progress…")
+        QApplication.processEvents()
+        # use green channel as reference, same as align
+        G = np.ascontiguousarray(self.image[..., 1].astype(np.float32, copy=False))
+        sigma = float(self.sep_spin.value())
+
+        # run a mini version of what the worker does
+        bkg = sep.Background(G)
+        data_sub = G - bkg
+        objs = sep.extract(data_sub, sigma, err=bkg.globalrms)
+        total = 0 if objs is None else len(objs)
+
+        # compute how many are in the EDGE ring, using same logic/constants
+        h, w = G.shape[:2]
+        cx, cy = w * 0.5, h * 0.5
+        rmax = np.hypot(cx, cy)
+        edge_inner = RGBAlignWorker.EDGE_INNER_FRAC * rmax
+
+        if objs is not None and total > 0:
+            r = np.hypot(objs["x"] - cx, objs["y"] - cy)
+            edge_mask = r >= edge_inner
+            edge_count = int(edge_mask.sum())
+        else:
+            edge_count = 0
+
+        msg = (f"[Trial SEP]\n"
+               f"sigma = {sigma}\n"
+               f"total stars (green): {total}\n"
+               f"outer-ring stars (used by EDGE): {edge_count}")
+        self.summary_box.setPlainText(msg)
+        self.progress_label.setText(f"Trial SEP: {total} stars, {edge_count} edge")
+
+
+    def _start_align(self):
+        if self.image is None:
+            QMessageBox.warning(self, "RGB Align", "No image found in active view.")
+            return
+        if self.image.ndim != 3 or self.image.shape[2] < 3:
+            QMessageBox.warning(self, "RGB Align", "Image must be RGB (3 channels).")
+            return
+        if astroalign is None:
+            QMessageBox.warning(self, "RGB Align", "astroalign is not available.")
+            return
+
+        model = self._selected_model()
+        sep_sigma = float(self.sep_spin.value())
+        self.progress_label.setText("Starting…")
+        self.progress_bar.setValue(0)
+
+        self.worker = RGBAlignWorker(self.image, model, sep_sigma=sep_sigma)
+        self.worker.progress.connect(self._on_worker_progress)
+        self.worker.done.connect(self._on_worker_done)
+        self.worker.failed.connect(self._on_worker_failed)
+        self.worker.start()
+        self.btn_run.setEnabled(False)
+
+    def _selected_model(self) -> str:
+        txt = self.model_combo.currentText().lower()
+        if "edge" in txt:
+            return "edge-sep"
+        if "affine" in txt:
+            return "affine"
+        if "poly 3" in txt:
+            return "poly3"
+        if "poly 4" in txt:
+            return "poly4"
+        if "homography" in txt:
+            return "homography"
+        return "edge-sep"  # super-safe fallback
+
+    # slots
+    def _on_worker_progress(self, pct: int, msg: str):
+        self.progress_bar.setValue(pct)
+        self.progress_label.setText(msg)
+
+    def _on_worker_failed(self, err: str):
+        self.btn_run.setEnabled(True)
+        self.progress_bar.setValue(0)
+        self.progress_label.setText("Failed.")
+        QMessageBox.critical(self, "RGB Align", err)
+
+    def _on_worker_done(self, out: np.ndarray):
+        self.btn_run.setEnabled(True)
+        self.progress_bar.setValue(100)
+        self.progress_label.setText("Applying…")
+
+        summary_lines = []
+        w = self.worker  # type: ignore
+
+        if w is not None:
+            def _fmt_mat(M):
+                return "\n".join(
+                    ["    " + "  ".join(f"{v: .6f}" for v in row) for row in M]
+                )
+
+            def _spread_stats(pts, shape):
+                if pts is None:
+                    return "  points: 0"
+                pts = np.asarray(pts, dtype=float)
+                if pts.size == 0:
+                    return "  points: 0"
+                h, w_ = shape[:2]
+                cx, cy = w_ * 0.5, h * 0.5
+                if pts.ndim != 2 or pts.shape[1] != 2:
+                    return f"  points: {len(pts)} (unusual shape {pts.shape})"
+                r = np.hypot(pts[:, 0] - cx, pts[:, 1] - cy)
+                rmax = np.hypot(cx, cy)
+                edge = r > (RGBAlignWorker.EDGE_FRAC * rmax)
+                return (
+                    f"  points: {len(pts)} "
+                    f"(edge: {edge.sum()} ≥{RGBAlignWorker.EDGE_FRAC*100:.0f}%Rmax)"
+                )
+
+            h_img, w_img = self.image.shape[:2]
+
+            # ── R → G ──
+            if w.r_xform is not None:
+                kind, X = w.r_xform
+                summary_lines.append("Red → Green:")
+                if w.r_pairs is not None and len(w.r_pairs) == 2:
+                    summary_lines.append(_spread_stats(w.r_pairs[1], (h_img, w_img)))
+                summary_lines.append(f"  model: {kind}")
+                if kind == "affine":
+                    A = np.asarray(X, dtype=float).reshape(2, 3)
+                    M = np.vstack([A, [0, 0, 1]])
+                    summary_lines.append(_fmt_mat(M))
+                elif kind == "homography":
+                    Hm = np.asarray(X, dtype=float).reshape(3, 3)
+                    summary_lines.append(_fmt_mat(Hm))
+                else:
+                    summary_lines.append("  (non-matrix; warp callable)")
+
+            # ── B → G ──
+            if w.b_xform is not None:
+                kind, X = w.b_xform
+                summary_lines.append("")
+                summary_lines.append("Blue → Green:")
+                if w.b_pairs is not None and len(w.b_pairs) == 2:
+                    summary_lines.append(_spread_stats(w.b_pairs[1], (h_img, w_img)))
+                summary_lines.append(f"  model: {kind}")
+                if kind == "affine":
+                    A = np.asarray(X, dtype=float).reshape(2, 3)
+                    M = np.vstack([A, [0, 0, 1]])
+                    summary_lines.append(_fmt_mat(M))
+                elif kind == "homography":
+                    Hm = np.asarray(X, dtype=float).reshape(3, 3)
+                    summary_lines.append(_fmt_mat(Hm))
+                else:
+                    summary_lines.append("  (non-matrix; warp callable)")
+
+        summary_text = "\n".join(summary_lines) if summary_lines else "No transform info."
+        self.summary_box.setPlainText(summary_text)
+
+        if self.parent is not None and hasattr(self.parent, "_log") and callable(self.parent._log):
+            self.parent._log("[RGB Align]\n" + summary_text)
+
+        try:
+            if self.chk_new_doc.isChecked():
+                dm = getattr(self.parent, "docman", None)
+                if dm is not None:
+                    dm.open_array(out, {"display_name": "RGB Aligned"}, title="RGB Aligned")
+                else:
+                    if hasattr(self.doc, "apply_edit"):
+                        self.doc.apply_edit(out, {"step_name": "RGB Align"}, step_name="RGB Align")
+                    else:
+                        self.doc.image = out
+            else:
+                if hasattr(self.doc, "apply_edit"):
+                    self.doc.apply_edit(out, {"step_name": "RGB Align"}, step_name="RGB Align")
+                else:
+                    self.doc.image = out
+
+            self.progress_label.setText("Done.")
+        except Exception as e:
+            self.progress_label.setText("Apply failed.")
+            QMessageBox.warning(self, "RGB Align", f"Aligned image created, but applying failed:\n{e}")
+
+
+
+
+
+def align_rgb_array(img: np.ndarray, model: str = "edge-sep", sep_sigma: float = 3.0) -> np.ndarray:
+    """
+    Headless core: returns a new RGB image with R,B aligned to G.
+    Raises RuntimeError on problems.
+    """
+    if img is None or img.ndim != 3 or img.shape[2] < 3:
+        raise RuntimeError("Image must be RGB (3 channels).")
+    if astroalign is None:
+        raise RuntimeError("astroalign is not available.")
+
+    worker = RGBAlignWorker(img, model, sep_sigma=sep_sigma)
+
+    try:
+        R = np.ascontiguousarray(img[..., 0].astype(np.float32, copy=False))
+        G = np.ascontiguousarray(img[..., 1].astype(np.float32, copy=False))
+        B = np.ascontiguousarray(img[..., 2].astype(np.float32, copy=False))
+
+        def _estimate_and_warp(src, ref):
+            # NOTE: _estimate_transform now returns 3 values
+            kind, X, _pairs = worker._estimate_transform(src, ref, model)
+            return worker._warp_channel(src, kind, X, ref.shape)
+
+        R_aligned = _estimate_and_warp(R, G)
+        B_aligned = _estimate_and_warp(B, G)
+
+        out = np.stack([R_aligned, G, B_aligned], axis=2)
+        if img.dtype != out.dtype:
+            out = out.astype(img.dtype, copy=False)
+        return out
+    except Exception as e:
+        raise RuntimeError(str(e))
+
+def run_rgb_align_headless(main_window, document, preset: dict | None = None):
+    if document is None:
+        QMessageBox.warning(main_window, "RGB Align", "No active document.")
+        return
+
+    img = np.asarray(document.image)
+    p = dict(preset or {})
+    model = p.get("model", "edge").lower()
+    sep_sigma = float(p.get("sep_sigma", 3.0))
+    create_new = bool(p.get("new_doc", False))
+
+    sb = getattr(main_window, "statusBar", None)
+    if callable(sb):
+        sb().showMessage(f"RGB Align ({model})…", 3000)
+
+    try:
+        out = align_rgb_array(img, model=model if model != "edge" else "edge-sep",
+                              sep_sigma=sep_sigma)
+    except Exception as e:
+        QMessageBox.critical(main_window, "RGB Align (headless)", str(e))
+        return
+
+    if create_new:
+        dm = getattr(main_window, "docman", None)
+        if dm is not None:
+            dm.open_array(out, {"display_name": "RGB Aligned"}, title="RGB Aligned")
+        else:
+            # fallback to replace if we can't create new
+            try:
+                document.apply_edit(out, {"step_name": "RGB Align"})
+            except Exception:
+                document.image = out
+    else:
+        # in-place
+        if hasattr(document, "apply_edit"):
+            document.apply_edit(out, {"step_name": "RGB Align"}, step_name="RGB Align")
+        else:
+            document.image = out
+
+    if callable(sb):
+        sb().showMessage("RGB Align done.", 3000)