Interlaced PNGs cannot be directly embedded but have to be re-encoded

src/img2pdf.py

@@ -759,6 +759,8 @@ def read_images(rawdata, colorspace, first_frame_only=False):
 
     # depending on the input format, determine whether to pass the raw
     # image or the zlib compressed color information
+
+    # JPEG and JPEG2000 can be embedded into the PDF as-is
     if imgformat == ImageFormat.JPEG or imgformat == ImageFormat.JPEG2000:
         color, ndpi, imgwidthpx, imgheightpx = get_imgmetadata(
                 imgdata, imgformat, default_dpi, colorspace, rawdata)
@@ -770,12 +772,22 @@ def read_images(rawdata, colorspace, first_frame_only=False):
             raise JpegColorspaceError("jpeg can't have an alpha channel")
         im.close()
         return [(color, ndpi, imgformat, rawdata, imgwidthpx, imgheightpx, [])]
-    elif imgformat == ImageFormat.PNG:
+
+    # We can directly embed the IDAT chunk of PNG images if the PNG is not
+    # interlaced
+    #
+    # PIL does not provide the information whether a PNG was stored interlaced
+    # or not. Thus, we retrieve that info manually by looking at byte 13 in the
+    # IHDR chunk. We know where to find that in the file because the IHDR chunk
+    # must be the first chunk.
+    if imgformat == ImageFormat.PNG and rawdata[28] == 0:
         color, ndpi, imgwidthpx, imgheightpx = get_imgmetadata(
                 imgdata, imgformat, default_dpi, colorspace, rawdata)
         pngidat, palette = parse_png(rawdata)
         return [(color, ndpi, imgformat, pngidat, imgwidthpx, imgheightpx, palette)]
-    else:
+
+    # Everything else has to be encoded
+
     result = []
     img_page_count = 0
     # loop through all frames of the image (example: multipage TIFF)