Panoramic stitching is easier said than done. When Microsoft shipped the first release of Windows Live Photo Gallery last November, I must admit I didn’t think too highly of the panoramic stitching feature they included in the application. I mean this is a free application, how good can it be?
To find out, I’ve put it under some rigorous testing with what I think are some extreme stitching scenarios. To my surprise, it easily surpassed my expectations and could even give the professional applications a run for its money. Have a look at just how well it did and didn’t.
The alternate stitching application I’m using is called Autopano Pro. It retails for about US$145 (that’s $145 more than Live Photo Gallery for those of you without a calculator) and is considered one of the most advanced professional panorama creation tools today. It holds the prize for stitching the world’s second largest digital panorama of 13 gigapixels.
The first test is a 8-piece panorama of a complicated beach scene with over an hundred boats being disassembled after a race. There’s people moving around, objects moving or being disassembled, boats moving in the water and waves splashing on the sand. So basically it’s one hell of a challenge. Each photo was exposed at different shutter speeds but with the same aperture and ISO, so the software will also have to compensate varying exposures.
Here is the result of Autopano Pro (8931x3026px).
Here is the result of Windows Live Photo Gallery (8112x2970px).
In the big picture (pun), both results are excellent considering the circumstances. In detail, each reveals a few different minor flaws in the stitching. For example in the Autopano picture, a few sails were distort with sand showing through. A small misaligned seam also appears on the horizon. In Windows Live Photo Gallery (WLPG), a sail’s mast is misaligned with its bottom as well as a small seam on the horizon.
The second test taken from the beach is similar to the first in regards to moving people and objects, but has 13-segments in total which increases the complexity dramatically since nearly in each shot, every object is changing in some way or other. Each photo was also exposed differently.
Here is the result of Autopano Pro (7745x2409px).
Here is the result of Windows Live Photo Gallery (7492x2878px).
In this example the difference between Autopano and WLPG is much more obvious. Apart from the critical flaw in WLPG’s picture where half a sail is missing on the boat in the left, WLPG also has problems compensating for the varying exposure amongst the different pictures. Whilst the WLPG is brighter overall, in the middle of the picture parts of the sky is obviously inconsistent with the color around it. On the other hand the Autopano picture is much darker, but has more consistent brightness between the pictures.
The last test is 6-piece panorama of an ocean beach with big rocks and waves. There’s not a lot of moving people, but water and especially rolling waves is always tough since not only are they moving rapidly, its almost impossible to find similarities waves between shots so it’s all up to the magic of software. Again, each photo was exposed differently.
Here is the result of Autopano Pro (6575x2767px).
Here is the result of Windows Live Photo Gallery (6265x2809px).
In this example the result from WLPG is actually more appealing that Autopano’s. The exposure on the big rock at the left is much more visible yet there are no inconsistencies in the color of the sky. However unfortunately WLPG seems to suffer from a few more obvious misalignments in both the horizon and a small wave coming towards the middle.
By now you’re probably wondering how WLPG can practically come out of the blue and do so well in its first release. The answer is actually quite simple. Both WLPG and Autopano benefit from some breakthrough image-matching technologies developed by Matthew Brown at University of British Columbia, now working at Microsoft Research. In 2003, Matthew developed the first fully automatic 2D image stitching algorithm which is now the foundation of many commercial stitching applications. Of course more magic happens behind the scenes than just image matching, and those details can be found here.
If you have a digital camera and haven’t ever tried to make a panorama, then I urge you to try it because it can be lots of fun (even if you fail). Having one of the best image-stitching technology on the market today built into a free Windows application is just a dream come true for amateur photographers all over. Plus, with the amount of talent Microsoft Research has in this field, the technology is bound to get even better.
For more examples of WLPG stitching, Brandon LeBlanc has a few recent panoramas from CES in Las Vegas.