Inconsistencies

[nekitamo]

After seeing constant attacks and accusations that the research done here is biased towards proving the drones are real (and I assure you that is not the case with my work), I believe it would be in order for us to create a new board/thread with all of the inconsistencies noted so far in various elements of this case. By this I don't mean it should include all those numerous unsubstantiated speculations, but only elements for which the inconsistency can be proven beyond reasonable doubt. I'm going to elaborate on some of the inconsistencies that I'm aware of, and if nobody objects, this could be the first content for the new board/thread (without this introduction and a bit more detailed, of course):
----------------------------------- snip ----------------------------------------------



1. Raj's images are inconsistent with a Minolta DiMageX camera

Overall hue / vignetting / lightness comparison:



Detailed sharpness / video-look / exposure comparison:



CA / PF comparison:




More Minolta DiMageX sample images: http://www.pbase.com/cameras/minolta/dimagex



2. Noise patterns of variable size in Raj's images

Variable size of the noise in patterns extracted from Raj's images indicates they were created from resized originals:



The patterns were produced with high pass+autocontrast functions in PS from the top left image corners showing clear blue sky in all of Raj's images except PICT0018, which is from top-middle part. Real Minolta DiMageX samples (top row) are random clips of clear blue sky from two different cameras of the same model.


3. Disturbed background noise in Stephen's images

The same procedure as above (high pass+autocontrast) was used for producing the following noise patterns from Stephen's images:







Example showing that lens-produced blur doesn't affect the background CCD noise:

[tomi]

What do you mean by "resized" originals exactly?  Does that mean that a raw photo was saved as a jpeg or that a high resolution jpeg was resized automatically by a piece of software before emailing it out?  Or does it mean something else entirely?
Thanks for so much detailed research as well.
Cheers,

[nekitamo]

First, I'll clarify some more on what I did to get those noise patterns:



As you can see, in first three images I used exactly the same group of pixels (256x256), which means they should represent exactly the same area of the CCD sensor chip in Raj's camera. Then I applied high pass filter (radius 2.0 pixels) to eliminate everything but high-frequency CCD noise. In this way, eventual clouds, vignetting, or similar low-freq transitions don't affect the resulting noise pattern. Lastly, auto-contrast function was applied to make noise more visible.

What seems suspicious here is obviously different noise grain size in each image, even though the first three noise patterns should have been produced by exactly the same CCD elements, and thus very similar under similar lighting levels. AFAIK, a proper noise grain size in original, unresized image should match the size of pixels as that's what it is by definition - a slight, unwanted variation in response of each pixel. Here's a similar example from my camera, and although they all come from different images, you can hardly tell the difference between the second and the third pattern:



Like I said, one possible explanation for this difference in noise grain size is that original images of various unknown resolutions were resized (along with the noise they contain) and clipped to 1600x1200 we're looking at here, and the first inconsistency that I mentioned (that images weren't produced with a Minolta DiMageX) certainly does not help to disprove this.

It would be great if someone double-checks this by using different methods/software.

[elevenaugust]

Hi Neki.
Great work!
We already knew your point n°1; remember Jtp's analysis:

The following Fourier transforms are all created from images taken with a real Minolta DiMAGE X camera except for the second one which is from Rajman's PICT0016:


And to be complete, here's a demonstration on how JPEG compression in the Capitola images affects the analysis. The Fourier transform on the left is from an image compressed with the standard compression of a DiMAGE X camera, and the Fourier transform on the right is from the same image, but it has been recompressed in Photoshop using the same settings as in the Capitola images:

[10538]

Good job Neki.  I'm with you!  Our research can only be taken seriously if we remain unbiased.  I am here to get the truth.  I know that it has been said the purpose of this website is to study the "reality" of the drones.  Please keep in mind that this reality can take many forms.  I think we are doing a good job so far.  Let's study the evidence and see where it leads us.  Not form a conclusion and then seek evidence to prove it.

About the hue difference, I see the Raj pics have much more red in them.  Have you checked exif for firmware differences or exposure setting differences?  Could Raj or his brother have tweaked the pics with their PS software?  Maybe some kind of auto leveling function?  Could Raj's camera have been malfunctioning? How about a misconfigured white balance or other setting?

I found this in a review for the Dimage X

Special Features

Movie mode with sound.
Continuous Shooting mode.
10-second Self-Timer for delayed shutter release.
Voice Memo mode for recording captions.
White balance (color) adjustment with five modes.
DPOF (Digital Print Order Format) compatibility.
USB cable for connection to a computer (driver software included).

Recommendation
The Dimage X is very small and compact. Its full automatic exposure control suits standard point-and-shoot operations though the user does have a few options for lightening or darkening the image, or adjusting the color balance.

[nekitamo]

About the hue difference, I see the Raj pics have much more red in them.  Have you checked exif for firmware differences or exposure setting differences?  Could Raj or his brother have tweaked the pics with their PS software?  Maybe some kind of auto leveling function?  Could Raj's camera have been malfunctioning? How about a misconfigured white balance or other setting?

Well, those are all valid points, but after browsing through a good part out of more than 15000 dimagex images behind the link mentioned in my opening post, I've found none with similar characteristics. Raj's images somehow look too good for dimagex, too perfect for such simple camera with known flaws present in almost every image. I even considered a possibility that they were shot using RAW mode (dimagex does have this feature), but such images take at least 8 seconds for saving and use a lot of space - which would mean yet another inconsistency with Raj's story.

[10538]

One thing I'm noticing is it's very hard to find a picture with the exact same exif data as the Raj pics.  Most have the tag exif version 210 instead of 220.  Some have software version V100-02 but most don't.  Try to find a pic with all the same tags and has been run through PS is about impossible.  I still wonder if there could be a valid explanation for the differences and I think it's possible.  It may be that we have not yet stumbled onto it.

I have to admit I was very disappointed when the Raj pole could not be found.  I kept running it through my mind and could not put it all together with the original Craigslist post.

One question I never figure out was how (and who) the craigslist post was first discovered. 

[tomi]

Here is a picture I found from that link you posted with the DiMage X photos.  I found a few with some curious overtones and similar to the raj colouring.

[10538]

Could CCD temperature play a role in hue reproduction?  That is, if the camera had been laying on a table in direct sunlight for an extended period, would that cause an increased red level?

[spf33]

excellent presentation, nekitamo.

you say the likely way to introduce the noise differences in raj's photos is for them to have been scaled from the originals, which seems unlikely.

and although jpg compression can some effect on the noise, the extreme differences seen in the fft's makes jpg compression an unlikely culprit for the change in noise patterns.

what possibilities does this leave?  i can't see how color or brightness adjustments would have and effect.  what about some sort of sharpening command,  could raj or his brother have attempted to clean these images up in ps?

also, a common technique when compositing is to unite the composited object with the background photo by adding a bit of film grain to enhance photorealism and soften the edges of the composited object blending them into the background photo.

have you tried adding ps noise to any dimage images and comparing the results to raj's?

[elevenaugust]

Spf, Neki, here's what I find by doing some tries with my own camera:

Here's the PICT0013 I used for my study


And the Fourier Transform of a 256*256 square of it:


The original photo I took with my camera (Canon Powershot A720 IS)

To this photo, I applied the following process:
1- A center motion blur of -80° and 6 pixels
2- Add 20% of noise with 0 colours saturation
3- Add JPEG compression artifact by saving it at 80%

The result:


A direct comparison:

Even if the camera model is quite different (in fact, we know nothing about the REAL camera used by Raj!!), I think I have here a good match.
Not a smoking gun neither real nor fake, but a confirmation of what I thought: the odd aspect of the photos is a combinaison of multiple factors:

• Motion blur
• Noise
• JPEG artefact

A real deeper study of the EXIF/XMP data could surely be useful....

[spf33]

11, can we really compare fft's and noise using 2 different cameras as you have done?  maybe i'm misunderstanding.

didn't raj confirm at omf that he used the minolta dimage?  along with the exif data, i'm not sure i know what you mean "we know nothing about the REAL camera used by Raj!!".

[elevenaugust]

I'am pretty sure that Raj never used a Minolta dIMAGEx, for several reasons:
1- As Neki exposed it above, Raj images look too good for such a camera.
2- The study of "real" Minolta dIMAGEx Fourier Transform shows now match:
The following Fourier transforms are all created from images taken with a real Minolta DiMAGE X camera except for the second one which is from Rajman's PICT0016:


Unless Raj or his brother-in-law made some manipulations under Photoshop to clear up the photos, but I don't think these simples manipulations are able to modify the fft so much.
I'm still trying to find another camera that could produce such a fft without elaborate manipulation.
My theory is that Raj could have made a mistake about the Camera model.

[nekitamo]

My theory is that Raj could have made a mistake about the Camera model.

Actually, he mentioned it only as "an old Minolta", the exact model was deduced from EXIF data.

Btw, here's what can be told about the timeline from embedded data:

- per EXIF metadata, i.e. PICT0013 was taken at:

Code: [Select]

  <exif:DateTimeOriginal>2007-05-16T17:41:11+01:00</exif:DateTimeOriginal>
  <exif:DateTimeDigitized>2007-05-16T17:41:11+01:00</exif:DateTimeDigitized>

- edit info from xmp metadata embedded into images by photoshop (also for PICT13):

Code: [Select]

- <rdf:Description rdf:about="" xmlns:xap="http://ns.adobe.com/xap/1.0/">
  <xap:CreatorTool>Adobe Photoshop CS2 Windows</xap:CreatorTool>
  <xap:CreateDate>2007-05-20T13:04:39-07:00</xap:CreateDate>
  <xap:ModifyDate>2007-05-20T13:04:39-07:00</xap:ModifyDate>
  <xap:MetadataDate>2007-05-20T13:04:39-07:00</xap:MetadataDate>
  </rdf:Description>

Note that camera timezone is shown as GMT+1, though this doesn't mean it wasn't set for correct local time. We can see that GMT-7 timezone (PDT) is set at the computer used for editing, so it's probably in the area. Also, note that the images were edited in the following order:

PICT13: 2007-05-20T13:04:39-07:00
PICT15: 2007-05-20T13:05:03-07:00
PICT16: 2007-05-20T13:05:16-07:00
PICT17: 2007-05-20T13:05:32-07:00
PICT14: 2007-05-20T13:48:32-07:00
PICT18: 2007-05-20T13:59:32-07:00

Last two images seem almost like an afterthought, as they were saved 40 minutes later. And finally, Raj's Craigslist post is dated 2007-05-20, 3:18PM PDT, advertising the images that were already uploaded to flickr in the meantime. This leads me to the conclusion that the same computer was used for editing, upload to flickr and posting to Craigslist, and judging from Raj's story it should be his, so - hopefully - it was also used for posting to OM forum ten days later.

It's probably too late now, but is it possible for the detectives to use Raj's IP from OM forum and  ask his ISP about its actual user for the exact time and date of his posting? Probably not w/o the court order, but maybe there is a way? 

[elevenaugust]

Ok, so here's the possibilities:

1- Raj used a Minolta DimageX, but he (or his brother-in-law) made some improvements after taking the pictures. The question is which improvement? And which is able to modify so much the Fourier transform?
We already know that the JPEG compression artefact doesn't alter so much the fft.
So is it rescaling, contrast or luminosity improvement.....or what?
I will make some test on some photographs shown on the link Neki gave.

2- Raj never used a Minolta, but another camera; means that the EXIF has been modified intentionnally; however, I wasn't able to find yet any camera who can produce a similar Fourier transform. Moreover, I don't see the point for Raj to hide his original camera model.