detect when a PDF has images at the same resolution, just chopped up

dlibrary/dlibrary.py

@@ -470,6 +470,12 @@ def median(items):
     items.sort()
     return items[len(items) // 2]
 
+def mean(items):
+    if len(items) == 0:
+        return None
+
+    return sum(items) / len(items)
+
 def superior_or_equal(a, b):
     return len(a) >= len(b) and all(a[i] >= b[i] for i in range(len(b)))
 
@@ -500,14 +506,29 @@ def try_collate_images_vs_pdf(srcs, dest, start_index, exclude):
 
     pdf_sizes = pdf_image_sizes(inner_pdf)
     standalone_sizes = [standalone_image_size(f) for f in images]
-    if abs(len(pdf_sizes) - len(standalone_sizes)) > 2:
-        return False
 
     median_pdf_size = median(pdf_sizes)
     median_standalone_size = median(standalone_sizes)
     if not (median_pdf_size and median_standalone_size):
         return False
 
+    if abs(len(pdf_sizes) - len(standalone_sizes)) > 2:
+        with fitz.open(inner_pdf) as pdf:
+            pdf_page_count = len(pdf)
+        height_adjusted_pdf_image_count = (
+            len(pdf_sizes) *
+            mean([size[1] for size in pdf_sizes]) / mean([size[1] for size in standalone_sizes])
+        )
+        if (
+                abs(pdf_page_count - len(standalone_sizes)) <= 2 and
+                len(pdf_sizes) > len(standalone_sizes) and
+                median_pdf_size[0] == median_standalone_size[0] and
+                abs(height_adjusted_pdf_image_count - len(standalone_sizes)) <= 2
+        ):
+            return collate_from_paths(non_pdf_srcs, dest, start_index, exclude)
+        else:
+            return False
+
     if superior_or_equal(median_standalone_size, median_pdf_size):
         return collate_from_paths(non_pdf_srcs, dest, start_index, exclude)
     elif superior_or_equal(median_pdf_size, median_standalone_size):