diff --git a/CHANGELOG.md b/CHANGELOG.md
index adeec4a..88fddcc 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,6 +1,10 @@
 Changelog
 =========
 
+v0.3.10 (2026-08-03)
+--------------------
+* Make `SkewCorrector` faster
+
 v0.3.9 (2026-06-19)
 -------------------
 * Add `OrientationClassifier` class
diff --git a/VERSION b/VERSION
index ed63cdf..81de5c5 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-0.3.9
\ No newline at end of file
+0.3.10
\ No newline at end of file
diff --git a/dedocutils/preprocessing/skew_corrector.py b/dedocutils/preprocessing/skew_corrector.py
index ddd7dbd..ef54174 100644
--- a/dedocutils/preprocessing/skew_corrector.py
+++ b/dedocutils/preprocessing/skew_corrector.py
@@ -14,37 +14,38 @@ class SkewCorrector(AbstractPreprocessor):
     The projection method is used to determine the rotation angle.
     """
     def __init__(self) -> None:
-        self.step = 1  # step
-        self.max_angle = 45  # max angle
+        self._step = 1  # step
+        self._max_angle = 45  # max angle
+        self._min_side = 1000  # the fine sweep runs on an image downscaled to this long side (never upscales a small page)
+        self._coarse_side = 512  # the coarse guess runs on this smaller thumbnail
 
     def preprocess(self, image: np.ndarray, parameters: Optional[dict] = None) -> Tuple[np.ndarray, dict]:
         parameters = {} if parameters is None else parameters
         orientation_angle = parameters.get("orientation_angle", 0)
 
         if orientation_angle:
-            rotation_nums = orientation_angle // 90
-            image = np.rot90(image, rotation_nums)
+            image = np.rot90(image, orientation_angle // 90)
 
         gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
         thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]
+        scale = min(1.0, self._min_side / max(thresh.shape[:2]))
+        if scale < 1.0:
+            thresh = cv2.resize(thresh, None, fx=scale, fy=scale, interpolation=cv2.INTER_AREA)
 
-        angles = np.arange(-self.max_angle, self.max_angle + self.step, self.step)
-        scores = [self.__determine_score(thresh, angle) for angle in angles]
+        coarse_scale = min(1.0, self._coarse_side / max(thresh.shape[:2]))
+        thumb = cv2.resize(thresh, None, fx=coarse_scale, fy=coarse_scale, interpolation=cv2.INTER_AREA) if coarse_scale < 1.0 else thresh
 
-        max_idx = scores.index(max(scores))
-        if max_idx >= 2 and scores[max_idx - 2] > scores[max_idx] * 0.98:
-            # if there are 2 approximately equal scores +- 1 step by max_score it will utilize angle between them
-            best_angle = angles[max_idx - 1]
-        elif max_idx < len(scores) - 2 and scores[max_idx + 2] > scores[max_idx] * 0.98:
-            best_angle = angles[max_idx + 1]
-        else:
-            best_angle = angles[scores.index(max(scores))]
+        coarse_angles = np.arange(-self._max_angle, self._max_angle + 1, 3)
+        coarse = float(coarse_angles[int(np.argmax([self._score(thumb, angle) for angle in coarse_angles]))])
+        lo, hi = max(coarse - 4, -self._max_angle), min(coarse + 4, self._max_angle)
+        fine_angles = np.arange(lo, hi + 0.001, self._step)
+        best_angle = float(fine_angles[int(np.argmax([self._score(thresh, angle) for angle in fine_angles]))])
 
-        rotated = rotate_image(image, best_angle)
+        rotated = image if best_angle == 0 else rotate_image(image, best_angle)
         return rotated, {"rotated_angle": float(orientation_angle + best_angle)}
 
-    def __determine_score(self, arr: np.ndarray, angle: int) -> Tuple[np.ndarray, float]:
+    @staticmethod
+    def _score(arr: np.ndarray, angle: float) -> float:
         data = rotate_image(arr, angle)
         histogram = np.sum(data, axis=1, dtype=float)
-        score = np.sum((histogram[1:] - histogram[:-1]) ** 2, dtype=float)
-        return score
+        return float(np.sum((histogram[1:] - histogram[:-1]) ** 2, dtype=float))