Downsize large images

Author: danielgarthur

scripts/deskew.py

@@ -4,12 +4,14 @@
 from show_baselines_and_textlines import show_baselines_and_textlines
 
 sys.path.append("../src")
-from util import to_binary, deskew
+from util import downsize, deskew
 
 
 def correct_skew(img_path, output_path):
     img = cv2.imread(img_path, cv2.IMREAD_GRAYSCALE)
 
+    img = downsize(img)
+
     import time
 
     print("starting deskew")

src/ocr.py

@@ -212,7 +212,9 @@ def process_image(image, model, metadata, split_lr=False):
 def prepare_image(image):
     page = PageAnalysis()
 
-    angle, corrected = util.deskew(image, limit=5, delta=1)
+    resized = util.downsize(image)
+
+    angle, corrected = util.deskew(resized, limit=5, delta=1)
 
     binary = util.to_binary(corrected)
 

src/util.py

@@ -116,6 +116,7 @@ def gss_max(f, a, b, tolerance=1e-5):
     while b - a > tolerance:
         c = b - (b - a) * invphi
         d = a + (b - a) * invphi
+
         if f(c) > f(d):
             b = d
         else:
@@ -140,11 +141,20 @@ def determine_score(angle):
 
     best_angle = gss_max(determine_score, lower_limit, upper_limit, 0.1)
 
-    (h, w) = binary_image.shape[:2]
+    (h, w) = img.shape[:2]
     center = (w // 2, h // 2)
     M = cv2.getRotationMatrix2D(center, best_angle, 1.0)
     corrected = cv2.warpAffine(
         img, M, (w, h), flags=cv2.INTER_CUBIC, borderMode=cv2.BORDER_REPLICATE
     )
 
     return best_angle, corrected
+
+
+def downsize(image, max_width=2550, max_height=3300):
+    h, w = image.shape[:2]
+    if w > max_width or h > max_height:
+        scaling_factor = min(max_width / w, max_height / h)
+        new_dimensions = (int(w * scaling_factor), int(h * scaling_factor))
+        image = cv2.resize(image, new_dimensions, interpolation=cv2.INTER_AREA)
+    return image