Comments:"Unbelievable - The Hacker Factor Blog"
URL:http://www.hackerfactor.com/blog/index.php?/archives/549-Unbelievable.html
Many of these trending picture topics are not new -- they are just new to the various forums. And some of the topics are not even very interesting or controversial (unless you actually actually care about Justin Bieber or follow that boy band One Direction -- which is really just a remake of Menudo but with less talent). Then again, maybe I just don't understand the controversy.
However, there are two currently trending pictures that got my attention...
Eddie Murphy
There's a picture that is supposed to be an old photo of an Eddie Murphy lookalike. This picture has been making its rounds for years. Back in 2010, BuzzFeed said it was Eddie Murphy's great-grandfather. In 2012, BuzzFeed featured it again in their Time-Traveling Celebrities montage.
The forum that began to make it popular at FotoForensics is at Godlike Productions -- a conspiracy site. The original posting on this forum (by a guy called "Ford Perfect") interpreted the ELA result as denoting a modified image.
Ford Perfect was correct -- this old Eddie photo is fake.
With Error Level Analysis (ELA), we need to make sure that it isn't identifying a false-positive result from some resave or poor image handling. The way to check this is to compare all surfaces with surfaces and all edges with edges.
In this picture, the surfaces (non-edges) include the background, his jacket, and the large, uniform portions of his hat. They are all at the same error level. Well, mostly. His stripped jacket is at the same error level as the non-striped portion (implying a consistent low-quality), but the middle band on the hat is at a higher error level than expected.
Meanwhile, the edges are inconsistent. Specifically, his eyes, nostrils, and mustache are at a higher error level than everything else. For example, his eyes have an extreme edge -- dark iris next to white sclera (the white of his eyes). In contrast, his shirt and jacket form a similar high-contrast area (white shirt next to dark jacket) but is at a lower error level potential. This flags the image as a likely modified picture.
I rarely rely on one analysis algorithm to evaluate an image. Instead, I use many different algorithms. I've previously written how photo analysis can reveal slight information variances. It can approximate an image's quality and can even reveal a face under a face. (Artists typically do something like a 99% blend for alignment, and we can identify the 1% residue from the original face.) In the Eddie Murphy picture, there is no face under the face -- it's a 100% overlay. However, the overlaid portion really stands out:
Not all of the face is from Eddie Murphy, but everything below the eyes has been replaced.
And the Award goes to....
My friend Xenon sent me a link to a growing controversy among news photographers.
This photo is supposed to show mourners in Gaza City carrying children who died in an Israeli air strike (Nov. 2012). The photographer, Paul Hansen, recently received the World Press Photo Award for this picture. The controversy is that the picture may not be a photo, or was significantly edited. The controversy deepened when Hansen "forgot" to bring the camera-RAW image as proof of the originality. As Spiegel Online wrote:
I have seen variations of this photo at FotoForensics since February. In my opinion, it has been significantly altered.
For analysis, it is best to start with the highest-quality and largest image possible. FotoForensics has versions that range from 1600x1067 to 460x276. The first thing that stands out is that none of the sizes are native to digital cameras. At bare minimum, the picture was significantly cropped and/or scaled.
The biggest picture at FotoForensics is also the only one that contains any useful metadata. The metadata says that it was captured by a Canon EOS 5D Mark III (a good digital camera) on 2012-11-20 at 9:39:38 am GMT+01:00. (The time is important... you'll see why in a moment.) It also says that it was modified with Adobe Photoshop CS6 for the Macintosh.
The thing about Photoshop is that it natively includes an XMP block that details the save history. In this case, it says the picture was:
The January 4th date is interesting because the deadline for submissions to the World Photo Awards was 17 January 2013. So although the base photo was taken in November 2012, it wasn't edited until about two weeks before the contest deadline. Moreover, the final date is the day after the international jury concluded their meeting (on 14 February 2013) and announced the winner.
I cannot tell you about the original picture(s), but I can tell you that the controversial picture is definitely not original. Moreover, it appears to have been modified specifically for this contest.
Knowing the history of the picture allows us to set a baseline. ELA should have rainbowing (faint red/blue patches), and edges and textures should have a higher ELA value due to Photoshop's unintentional auto-sharpening.
The rainbowing is clearly visible on the sky, walls, and people. And high contrast edges have a bright ELA result. This is consistent with the handling described in the metadata.
However... (There's always a "however"...) notice how the near wall on the far left is significantly brighter than the near wall on the far right. And notice how the middle people are much brighter than the other people. Those are either due to splices or touch-ups.
Finally, there's the sun. According to the metadata, the photo was taken in the morning in November in the Northern hemisphere. The sun should be low on the horizon. The strong shadows on the left building allow us to draw lines that intersect with the sun.
The shadows from the left wall line up with a consistent sun location. The sun isn't exactly low but maybe the reported time is wrong. At least the sky brightens in the direction of the sun. Unfortunately, the lighting on the people does not match the sun's position. The people should have dark shadows on their right sides (photo-left), but their facial lighting does not match the available lighting.
So here's what likely happened... The photographer took a series of photos. However, the sun's position made everyone dark and in silhouette. So, he combined a few pictures and altered the people so you could see their faces.
Finally, I found other photos of the samepeople. What stands out to me is that the foreground child in Hansen's "photo" has much more dirt on his face than the same child in the other photos.
Hansen's picture is a composite. This year's "World Press Photo Award" wasn't given for a photograph. It was awarded to a digital composite that was significantly reworked. According to the contest site, the World Press Photo organizes the leading international contest in visual journalism. However, the modifications made by Hansen fail to adhere to the acceptable journalism standards used by Reuters, Associated Press, Getty Images, National Press Photographer's Association, and other media outlets.
Now I understand the controversy.