12.5k jpeg not converted to progressive

[GoogleCodeExporter]

Site is http://aamjanata.com/, running ngx_pagespeed 1.6.29.5-3346

WPT gives us an "F" for progressive jpegs.  We can't fix all of them due to 
third party ads.  But there's one that looks like our heuristic is getting 
wrong.  I don't know if the heuristic should be tweaked or whether it is 
inevitable we'll miss some.

See WPT report: 
http://www.webpagetest.org/result/131002_A5_YQ1/1/performance_optimization/#prog
ressive_jpeg

FAILED (12.7 KB) - 
http://aj2-cdn.appspot.com/wp-content/uploads/200x140xcrack-on-wall-near-tracks-
Sandhurst-Road-200x140.jpg.pagespeed.ic.8nH_TftDk5.jpg

Original issue reported on code.google.com by jmara...@google.com on 3 Oct 2013 at 12:37

[GoogleCodeExporter]

Our heuristic doesn't work well for this image. We're predicting that the 
recompressed image will have 6,689 bytes. Because the default threshold for 
switching to progressive format is set to 10,240 bytes, it is compressed as 
non-progressive.

In fact, the sizes for the recompressed image is 11,948 bytes for progressive, 
and 12,857 bytes for non-progressive.

It's difficult to predict the recompressed size accurately because it depends 
on the image content. However, we may be able to improve the prediction 
formula. 

Original comment by hui...@google.com on 3 Oct 2013 at 1:23

[GoogleCodeExporter]

Original comment by hui...@google.com on 3 Oct 2013 at 1:23

• Changed state: Accepted

[GoogleCodeExporter]

Attaching a slurp of the site as of right now, so we can debug it without 
hitting the live site.

Original comment by jmara...@google.com on 3 Oct 2013 at 1:46

Attachments:

• aamjanata.com.slurp.tar.gz

[GoogleCodeExporter]

Also, http://www.webpagetest.org/result/131003_7T_K3E/ has response bodies 
saved.

Original comment by jmara...@google.com on 3 Oct 2013 at 1:48