Inconsistencies

[elevenaugust]

I made some tries with an original Minolta DimageX picture:

http://www.divshare.com/download/4720126-e34
Can't access to our ftp server!! 

I took a 256*256 square of the sky and applied various basic modifications, before creating the Fourier transform:



I can't make match yet the same circular halo shape as original Raj's ft picture, I think that, with more tries, a good match of the other characteristics could be find with a combination of some basic improvements.
But what about this circular halo shape Any idea?

[nekitamo]

I've noticed that a slight gaussian blur tends to create the effect you want, if I understood you correctly?

[elevenaugust]

The shadows studies of the Raj drone have been moved here:
http://droneteam.com/drt/index.php?topic=130.0

[nekitamo]

After working with elevenagust in the FFT thread, I can't help but conclude my research about noise inconsistencies in Raj's and Stephen's images the following way: they are resulting from someone's attempt to hide camera noise signature(s). I can't tell if it was done only to prevent identification of the author(s) or also to hide image manipulation, but this whole case seems like someone was following the instructions described here to the letter:

step 1: clean/replace EXIF data to prevent camera identification
step 2: clean the lens to avoid repeating smears/dust/etc.
step 3: crop and resize images to displace noise and eventual "stuck" pixels
step 4&5: remove image noise, or even better - add more noise (confirmed in Raj's images)
step 6: use different accounts (identities?) to distribute images, carefully hide your (network) tracks
step 7: use classic film camera and scan your images with various scanners.

All this steps can be recognized in various aspects of this case.

[spf33]

really appreciate all the lengths you have gone to nekitamo and 11a. 

just a couple thoughts to add here.

is it possible to do any sort of comparison between raj and steven noise to determine software specific addition of noise, i.e. both show use of photoshop's noise filter?

can we take into account the possible emf radiation from the drone which might be affecting the camera image sensors? can a real life example be cited?

[elevenaugust]

I'll try to apply the same process I applied in the fft thread to Stephen photos, to see I there is any match between the noises added.

We can say that there are two possibilities:
- The fake one: as explained by Neki, the faker hide camera noise signature, adding noise and save as jpeg at 70%. It seems that this process can hide image manipulation. Moreover, the motion blur, as we can see it on picture n°0017, can't be fake, UNLESS it has been done BEFORE add noise and save as jpeg 70%:

For a 256*256 blue sky square:




For a 256*256 landscape square, with motion blur added:




- The real one:

• Is there a "natural" process that can add noise to an image?
• Can the drones alter the CCD captor to give such an odd signature?

[elevenaugust]

Another inconsistencie is the difference between the thumbnail size and the original pictures size, in Raj's pictures.
This could be correlate to what Neki said about the noise differences between the pictures.

Does the thumbnail size match the size of the pictures ?
If we recompose thumbnails from the photographs we have, do we get a bytes result corresponding to an image '' weight'' in conformity with that indicated in the EXIF data?
Not certain, as we know, the JPEG format is getting heavier as more information is present in the picture. The presence of the drone should increase the weight of the labels. (Thumbnails)
In the hypothesis where the drone was added to blue sky pictures, there should be a difference.

Unfortunately, it cannot be used as a means of checking: a tool such as '' Exifer '' enables the visualisation of the label registered with the picture, but also to export it.
With such tools, it is possible for a forger to totally modify data EXIF and also to change the label associated with the photograph to have a match.
I extracted the labels from the 6 pictures below; we note they are strongly degraded. The format is JPEG as well. The labels' size is 120x160, which is nothing compared to the size of a LCD screen of 110 000 pixels used on this camera.
 
 

• (PICT0013) thumb: 3 473 octets
  originale: 548 064 octets
  Ratio: 157   
• (PICT0014) thumb: 2 002 octets
  originale: 462 564 octets
  Ratio: 231   
• (PICT0015) thumb: 1 389 octets
  originale: 277 177 octets
  Ratio: 199
• (PICT0016) thumb: 2 868 octets
  originale: 466 017 octets
  Ratio: 162   
• (PICT0017) thumb: 2 409 octets
  originale: 388 766 octets
  Ratio: 161   
• (PICT0018) thumb: 3 233 octets
  originale: 451 241 octets
  Ratio: 139

The ratio differences (weight of the original image divided by the thumbnail's weight) between the pictures are surprising. (From 140 to 230) Why such variations?
Certain thresholds on the level of granularity's scale composing the picture's elements may be crossed, that would cause variations. In order to be sure, it would be necessary to start from the original pictures with Photoshop and reduce them in the same way.

[spf33]

Moreover, the motion blur, as we can see it on picture n°0017, can't be fake, UNLESS it has been done BEFORE add noise and save as jpeg 70%:

motion blur>add noise>save as jpeg 70% would be logical steps for any compositing happening in pic17.

is there a way to determine if raj's photos were  file>save as>jpg or file>save for web>jpg? 
not sure how this could matter, just curious.

[nekitamo]

is there a way to determine if raj's photos were  file>save as>jpg or file>save for web>jpg? 
not sure how this could matter, just curious.

AFAIK, "Save for web..." option completely removes EXIF and other metadata, while "Save as..." removes some EXIF fields, but adds PS xmp metadata. As for your previous question:

is it possible to do any sort of comparison between raj and steven noise to determine software specific addition of noise, i.e. both show use of photoshop's noise filter?

Hard to tell was it really photoshop or some other software, or even which type of photoshop noise was eventually used as the difference is very small even with untouched images. Here's how it looks:



Both noise types cover full frequency spectrum, but with slightly different amplitude distribution. Judging from the histogram and slightly denser central area of the noise samples from Raj's images, I'd say gaussian noise was used. This seems to be the case also with Stephen's images, but even without additional tampering the noise in nearly overexposed areas of the sky is hardly usable. However, samples I did manage to extract look almost exactly like Raj's and are nowhere near this EOS350D example from dpreview.com with the similar over-exposed look of the sky:

[elevenaugust]

Here you can find a very complete analysis and description of the Minolta DimageX.

At section 11 (Conclusion Pro/Con), there are few interesting words:

• Low noise in most images
• Vignetting of images shot between wide and mid-zoom focal lengths

I digged into the pbase site you provides us, Neki, and haven't find any example of Minolta DimageX with such noise as Raj's pictures.
However, previous versions can produce similar noise, like this one, about Minolta Dimage 7i (third example).

So we have two possibilities:

• The original camera used is a Minolta DimageX, but the photos were modified to correct all the typicals defects of this camera:
  
  • Video Camera look. (Although i really wonder how it's possible!)
  • Vignetting. It's really possible to correct vignetting under Photoshop (see here). Extract: "One way I have eliminated vignetting in my digiscoped images with Adobe Photoshop is to use the Render Lighting Effects filter. I don't know if Elements has this filter or not, but for the sake of this reply I'll run through the exercise.
    I select "5DOWN" style and change the light type to OMNI. Generally, I only need four lights, one for each corner, so I turn one off. The remaining four I position around the border of the picture, move them carefully until I have the vignetting eliminated and overall picture as balanced as I can, or whatever just looks good to me."
    
  • Sharpness, lightness, overhall hue, exposure, reddish appearance Lots of things easy to do under PSP, but the result of such modifications is that the output image has absolutly anything anymore common with the original one.
    The question about the noise could be: "Is there an improvement process, such as described above, that could add noise?
• The original camera used is NOT a Minolta DimageX, and there are some clues showing this possibility (put aside the above photographs inconsistencies)
  
  • Minolta Dimage X is supposed to produce data EXIF in version 0210, not 0220. However, this could be explained by:
    
    • The use of a later version of the DimageX (Xt, Xi, X1?? But wasn't supposed to appears in the EXIFs??)
    • The version has been improved by a third application, or by the author himself (Why?)
    • The used of a Dimage x with a special firmware
  • The data is recorded in ''Big endian'' (EXIF Byte Order)
    We have here another contradiction with data produced by Dimage x since this camera creates â??â??little endian '' images. This particular data does not tolerate exception since it is related to the architecture of the processor used.
    ''Big endian'' corresponds to a reading of the bytes from left to right and ''Little endian'' from right to left. (The files are recorded in such a way to facilitate the job of reading ''words'' made of several bytes.)
    These changes cannot be done unless there is creation of a new file. It does not work on an original file, even if transformed since it is the way of recording the information which was modified. Some tests show that this change is effective if one creates a new document starting from Photoshop CS2 and gets the work developed from this one. Photoshop works in ''Big endian''.
    Here again, it would be necessary to check if:
    
    • The use of a later version of the DimageX, or
    • The improvement of the version by a third application, or
    • The used of a Dimage x with a special firmware
    can't modify the data recording from "Little endian" to "Big endian"
    
  If the camera used is not a Minolta DimageX, then we are 100% sure that the EXIF has been modified, at least to modify the name of the camera.
  This is really confusing (and doesn't make any sense), since it's really easy, as we saw it, to recognize other inconsistencies in the EXIF and in the photographs.
  
  
  And I don't want to add more confusion, but I would like to add something that almost everyone forgot and that, IMO, is VERY important:
  From FORE, on OMFRT board:
  
  The creation date of rajman's first upload was on May 16, 2007
  http://www.flickr.com/photos/8418528@N06/sets/72157600236430072/
  (Scroll all the way to the end of the testimony)
  ------------------------------------
  6 photos | 41,484 views
  Photos are from 16 May 07.
  ------------------------------------
  Then it was renamed or reuploaded on May 20th, 2007
  Compare:
  http://www.flickr.com/photos/8418528@N06/archives/date-taken/2007/05/16/
  and
  http://www.flickr.com/photos/8418528@N06/archives/date-posted/2007/05/calendar/
  
  Means that the photos were uploaded FIRSTLY ON MAY 16th (!!) and not on May 20th, like everyone believe it.
  Is there a possibility that someone saw and saved these first uploads between  May 16th and 20th?
  Curious to see them.....
  

[/list]

[nekitamo]

Not so sure about the pictures being uploaded on May 16th... I think what may have lead to this mix-up is because flickr is also showing dates from EXIF data in some places, not just upload dates. Unfortunately, there is no previous record to May 25th in the wayback machine, but note that both dates are present at this preserved page for PICT0013:

http://web.archive.org/web/20070525151221/www.flickr.com/photos/8418528@N06/506533380/in/set-72157600236430072/

The first page that Fore is talking about (with the "Photos are from 16 May 07" note) is also preserved:

http://web.archive.org/web/20070525082730/www.flickr.com/photos/8418528@N06/sets/72157600236430072/

And, finally, the page where upload dates are clearly shown:

http://web.archive.org/web/20070525151232/http://www.flickr.com/photos/8418528@N06

Note the "Bizzare Thing in sky" set at the right side? That leads to the page with "Photos are from May 16th" note. Now, check the similar page of the following (random) flickr user:

http://www.flickr.com/photos/madison_guy/sets/72157605386861521/

It says May 31st 2008, but the images were uploaded on June 1st, so it is clear that the date shown  on group page is extracted from the EXIF data and not the upload date. Also, if images are not all taken on the same date, a date range is shown in the same line. Try some other flickr users, you'll see what I mean.

But if Raj's images from earlier date really exist, it would be interesting to see them - that would mean they are still not edited in PS CS2 on May 20th like those that we have. Btw, I've just noticed that it is also possible to browse images per camera type on flickr:

http://www.flickr.com/cameras/minolta/dimage_x/

[10538]

It would be good if we knew the history of how the Raj report became public.  Here is OMF's Bren with his intial post about it.  Question is how did he hear about it?
http://lucianarchy.proboards21.com/index.cgi?board=cali1&action=display&thread=2018&page=16

[elevenaugust]

I have something very interesting.
Take a look at this picture:

She's very special (not for the model ) because:

• Same reddish hue as Raj's pictures
• Lots of noise

EXIF shows with Exiftool:

• Same EXIF version as Raj's pictures (V.0220)
• Data recorded in "Big Endian", like in Raj's pictures

Moreover, in this other photo taken with THE SAME Minolta DimageX, you have:

• NO vignetting
• Exif version 0210 (!!)
• Data recorded in "Little Endian"(!!)
• No reddish hue

These pictures come from Flickr, here
The user, "actionhero", took lots of shoots with his camera and some interesting things are shown by the study of the EXIF, with Exiftool:

Here's a comparative table:


We can here only make assumptions, but it's possible that:

• Either the "model" picture wasn't a picture, but a video (maybe produce this grainy appearance) and that Flickr user "actionhero" used a Mac software tool to get pictures of this video. The Minolta DimageX specification say:
  "Movie clip            320 x 240, with audio, 35 sec max". However, the size doesn't match.....
• Or it was a picture and the Mac software used after the shoots to get the pictures out of the camera and improve them modify the EXIF, at least the EXIF version and the encoding endian......and the noise??

I'll try to contact "actionhero" to see what is the process he used to improve the "model" picture, and which software he used.

Special thanks to 10538 and Karen for their brilliant ideas and supports

PS: Jeffrey's EXIF say:

[nekitamo]

I believe the lack of vignetting for low-res images is easily explained with the following:



But there is also this - Minolta implemented an image enhancement process in DiMageX which can't be turned off, so it affects all images:



Maybe we could also use the changes introduced with this process for better identification of DiMageX images (sharpening and saturation boost)?

[spf33]

what about the possibility that raj originally uploaded a tif to flickr which was converted to jpg by flickrs own software?

still looking, but cant' find any info on the flickr algorithms or processes used to convert jpg to tif other than flickr might not support exif import on tifs and the jpg compression they use to convert jpgs to tifs is not very good and uploaded tiffs look grainy compared to jpg.  guess this doesn't help explain steven's anomalies, though...