Image PAC Files

I wouldn’t be surprised if you have never heard of an Image PAC file. You may know it by the more common name Kodak Photo CD Image. Kodak’s PhotoCD format actually refers to the system and Disc format used to store images for compatibility with other hardware. The Kodak PhotoCD format was pretty advanced for its time, it original purpose was to store scanned 35mm film to a disc which was playable on computers and other hardware. In fact, because it was meant to store 35mm rolls as they were scanned it was the first use of the linked Multi-session CD format made standard by the orange book specification. The format was widely adopted at first, but eventually lost favor and was abandoned by 2004.

The Kodak PhotoCD format was also used on many commercial CD-ROM products. One example was the Corel Professional CD series. Below is a photo of a case of 200 CD’s I recently acquired. Each has around a hundred PCD images and viewing software on disc. Most discs can be viewed here. Or you can view their “Sampler” CD-ROM.

The actual PCD image file format was referred to as an Image PAC File. The format was unique in the fact it has multiple resolutions built into a single file. It also stored the raster data in a format called Photo YCC color encoding metric, developed by Kodak. This requires conversion to RGB for many uses. Adobe Photoshop for many years had an import filter for the format built in which included ICC profiles for properly converting the source to a destination colorspace, but support was dropped in CS3 of their products.

Photoshop Kodak PCD import

The Image PAC PCD format was a proprietary format which Kodak protected aggressively, even to the point of threatening legal action to those who attempted to reverse engineer the format. This frustrated developers and was probably part of the reason the format was abandoned. Of course this didn’t deter some curious developers and was partially reversed engineered and is available in the NetPBM library formally knows as PBMPlus. The tool hpcdtoppm was developed to convert PCD to PBM.

The trick in preserving older obsolete formats is to find a way to first identify them, gather significant properties, then migrate to a modern format if appropriate with minimal loss of data. Luckily most PCD files have the ascii string “PCD_IPI” starting around offset 2048. This is basically how the PRONOM registry identifies the format and has assigned it fmt/211. Exiftool also supports the format in identifying some of the significant properties.

ExifTool Version Number         : 12.62
File Name                       : 136009.PCD
Directory                       : /Users/thorsted/Desktop/blog/Kodak/PCD
File Size                       : 3.6 MB
File Modification Date/Time     : 2023:06:23 10:48:55-06:00
File Access Date/Time           : 2023:06:26 23:43:50-06:00
File Inode Change Date/Time     : 2023:06:27 11:18:38-06:00
File Permissions                : -rwx------
File Type                       : PCD
File Type Extension             : pcd
MIME Type                       : image/x-photo-cd
Specification Version           : 0.6
Authoring Software Release      : 3.0
Image Magnification Descriptor  : 1.0
Create Date                     : 1993:09:20 07:35:34-06:00
Image Medium                    : Color reversal
Product Type                    : 116/01 SPD 0064  #00
Scanner Vendor ID               : KODAK
Scanner Product ID              : FilmScanner 2000
Scanner Firmware Version        : 2.21
Scanner Firmware Date           : 
Scanner Serial Number           : 0296
Scanner Pixel Size              : 0b.30 micrometers
Image Workstation Make          : Eastman Kodak
Character Set                   : 95 characters ISO 646
Photo Finisher Name             : HADWEN GRAPHICS
Scene Balance Algorithm Revision: 3.1
Scene Balance Algorithm Command : Neutral SBA On, Color SBA On
Scene Balance Algorithm Film ID : Unknown (131)
Copyright Status                : Restrictions apply
Copyright File Name             : RIGHTS.USE
Orientation                     : Horizontal (normal)
Image Width                     : 3072
Image Height                    : 2048
Compression Class               : Class 1 - 35mm film; Pictoral hard copy
Image Size                      : 3072x2048
Megapixels                      : 6.3

Exiftool is able to gather much of the important properties including an original creation date and the pixel dimensions. It would be nice if was able to mention each of the resolution options as some later Pro versions of PCD had a 64 base for resolutions of 4096 x 6144.

Migration to a more modern open format is a common preservation strategy. The National Archives and Records Administration has the format NF00224 listed as needing to migrate to JPG, while others prefer migration to TIFF. Others have learned valuable lessons attempting to find the right method for migration. There is a right way and a wrong way as the Center for Digital Archaeology learned. The easiest method is to use the popular ImageMagick command-line tool.

thorsted$ identify 136009.PCD 
136009.PCD PCD 768x512 768x512+0+0 8-bit YCC 3.44727MiB 0.020u 0:00.006
thorsted$ convert 136009.PCD[5] -colorspace sRGB +compress 136009.tif
thorsted$ identify 136009.tif
136009.tif TIFF 3072x2048 3072x2048+0+0 8-bit sRGB 18.0004MiB 0.000u 0:00.000

ImageMagick along with most other tools like IrfranView and XnView only see the base resolution of 768 x 512, but with an extra little addition to the command by adding “[5]” after the filename if forces the conversion to use the “Fifth” 16 Base resolution which is the highest resolution on most PCD files, the Pro versions may have higher. The other issue is the colorspace conversion. It is known there could be a loss of highlights. This webpage illustrates different tools and the issues with highlights. You can see the difference if I use -colorspace RGB instead of sRGB.

ImageMagick conversion using RGB vs sRGB colorspace setting.

Other tools such as the open source pcdtojpeg and paid pcdMagic both work well, but the only tool I have tested so far which keeps the original metadata is pcdMagic.

ExifTool Version Number         : 12.62
File Name                       : 136009_1.tif
Directory                       : .
File Size                       : 38 MB
File Modification Date/Time     : 2023:06:27 12:06:26-06:00
File Access Date/Time           : 2023:06:27 12:06:29-06:00
File Inode Change Date/Time     : 2023:06:27 12:06:27-06:00
File Permissions                : -rw-r--r--
File Type                       : TIFF
File Type Extension             : tif
MIME Type                       : image/tiff
Exif Byte Order                 : Little-endian (Intel, II)
Subfile Type                    : Full-resolution image
Image Width                     : 3072
Image Height                    : 2048
Bits Per Sample                 : 16 16 16
Compression                     : Uncompressed
Photometric Interpretation      : RGB
Image Description               : color reversal: Unknown film. SBA settings neutral SBA on, color SBA on
Make                            : KODAK
Camera Model Name               : FilmScanner 2000
Strip Offsets                   : 1622
Samples Per Pixel               : 3
Rows Per Strip                  : 2048
Strip Byte Counts               : 37748736
Planar Configuration            : Chunky
Software                        : pcdMagic V1.4.19
Modify Date                     : 2023:06:27 12:06:26
Copyright                       : Copyright restrictions apply - see copyright file on original CD-ROM for details
Exif Version                    : 0231
Date/Time Original              : 1993:09:20 07:35:34
Create Date                     : 1993:09:20 07:35:34
Offset Time                     : -06:00
User Comment                    : color reversal: Unknown film. SBA settings neutral SBA on, color SBA on
Color Space                     : Uncalibrated
File Source                     : Film Scanner
Profile CMM Type                : Unknown (KCMS)
Profile Version                 : 2.1.0
Profile Class                   : Display Device Profile
Color Space Data                : RGB
Profile Connection Space        : XYZ
Profile Date Time               : 1998:12:01 18:58:21
Profile File Signature          : acsp
Primary Platform                : Microsoft Corporation
CMM Flags                       : Not Embedded, Independent
Device Manufacturer             : Kodak
Device Model                    : ROMM
Device Attributes               : Reflective, Glossy, Positive, Color
Rendering Intent                : Perceptual
Connection Space Illuminant     : 0.9642 1 0.82487
Profile Creator                 : Kodak
Profile ID                      : 0
Profile Copyright               : Copyright (c) Eastman Kodak Company, 1999, all rights reserved.
Profile Description             : ProPhoto RGB
Media White Point               : 0.9642 1 0.82489
Red Tone Reproduction Curve     : (Binary data 14 bytes, use -b option to extract)
Green Tone Reproduction Curve   : (Binary data 14 bytes, use -b option to extract)
Blue Tone Reproduction Curve    : (Binary data 14 bytes, use -b option to extract)
Red Matrix Column               : 0.79767 0.28804 0
Green Matrix Column             : 0.13519 0.71188 0
Blue Matrix Column              : 0.03134 9e-05 0.82491
Device Mfg Desc                 : KODAK
Device Model Desc               : Reference Output Medium Metric(ROMM)
Make And Model                  : (Binary data 40 bytes, use -b option to extract)
Image Size                      : 3072x2048
Megapixels                      : 6.3
Modify Date                     : 2023:06:27 12:06:26-06:00

There is a way to convert the PCD to TIF using ImageMagick, then using Exiftool to map some of the metadata over to the new TIFF file. It would look something like this:

exiftool -addtagsfromfile 136009.PCD '-EXIF:DateTimeOriginal<PhotoCD:CreateDate' '-EXIF:CreateDate<PhotoCD:CreateDate' '-ExifIFD:SerialNumber<PhotoCD:ScannerSerialNumber' '-ExifIFD:ExifImageWidth<PhotoCD:ImageWidth' '-ExifIFD:ExifImageHeight<PhotoCD:ImageHeight' '-IFD0:Make<PhotoCD:ScannerVendorID' '-IFD0:Model<PhotoCD:ScannerProductID' '-IFD0:Orientation<PhotoCD:Orientation' '-IFD0:Copyright<PhotoCD:CopyrightStatus' 136009.tif

JPG Structure

If you hadn’t been over to see the posters made by Ange Albertini, head over now. Below is his poster on the JPG image file format. This is the basic JFIF file format, which stands for JPEG File Interchange Format. There are also raw JPEG streams and Exif, Exchangeable Image File Format.

The basic format is pretty straight forward. There is a start of image marker FFD8 some format information, then the raster compressed data, then an end of image marker FFD9. Identification of a JPEG file should be pretty straight forward. Knowing the start and end marker values and then the type of JPEG based on the Application data, can be very specific. That is until some software engineers start playing fast and loose with the format specifications.

A while back I received a JPG file which didn’t identify using the latest PRONOM signature. It’s happened before, some new phones came out and started using a newer version of the exif specification so I submitted an update to PRONOM for JPG’s using exif 2.3 and greater. But also may need to submit another signature soon for the newly released Exif 3.0 specification! But this JPG I received wasn’t a new version, it should have been identified with the current PRONOM signature. It started with FFD8 and when I went to look at the end of the file for the end of image marker FFD9, it wasn’t where I expected it to be.

This JPG file had an additional 9632 bytes after the FFD9 end of image marker. But why? The image rendered just fine in multiple JPG viewers. The only warning from Exiftool was for “Unrecognized MakerNotes”, which is not too uncommon. So I went to the JPG Exif specification.

EOI, Recording this marker is mandatory. It shall be recorded in this position.

But reading a little further we see…..

Moreover, Exif/DCF readers should be implemented to operate without interruption even if certain kinds of data have been recorded after EOI of the primary image defined in the Exif standard. Specifically, unknown data after EOI of the primary image should be skipped. (see section 4.7.1)

So the extra data is allowed by specification. Any readers should ignore or skip any data after the EOI (End of Image). Well that makes identification more difficult. All the PRONOM signatures are based on having the EOI marker at the “End”. Some have allowance for padding, but not enough for the worst offenders……

The image referenced above was created on a Huawei MHA-L29 cameraphone. But since finding this image, I have also found many Samsung phones do the same thing. Here is one from a Samsung SM-G975U1. Much less padding but enough to throw off identification.

Apple iPhones are also not exempt from this “feature” either. When using the MacOS ImageCapture tool with the HEIC format, a bug can add an excessive amount of empty data at the end of the converted JPG file.

So, when it comes to identification, if your JPG files don’t seem to identify correctly, look closer at the end of the file, it may have some “extra” data.

Universal Scene Description

A few years ago I became obsessed with creating 3D models from physical objects. There was an app on my iPhone called 123D Catch by AutoDesk and it allowed you to take a series of photos with your iPhone camera, then combine them to create a 3D Model. This lead me down a path to eventually take a course on Photogrammetry and develop a process for capturing objects in our Museum.

Autodesk eventually discontinued the app and built the technology into their paid products. This is when we started seeing lidar introduced with handheld devices. The first one I tried was using my XBOX360 kinect sensor with the skanect software. The quality was horrible, but was fun to learn about depth sensors and structure from motion. When the iPhone finally came out with lidar sensor it was like Apple had read my mind. I love having the ability to capture objects I find into 3D models. The quality is pretty good, not as good as taking the time to capture image sets for photogrammetry and use tools like Aigisoft metashape, but apps like Scaniverse do a fantastic job. You can check out some of the models I have captured on my Sketchfab page.

With any new technology comes new file formats, and 3D formats are definitely no exception. It seems every software developer has to come up with their own proprietary format leaving the digital preservation folks scrambling to keep up. The DPC and Archivematica published a report a couple years ago and state:

“There are many challenges in preserving 3D data. As well as the complexity of the data itself, there is
a lack of interoperability between the different (often proprietary) systems that are used to create
and manipulate 3D models. Relationships to other data, software and hardware also need to be
captured and managed effectively.”

https://www.dpconline.org/docs/technology-watch-reports/2479-preserving-3d/file

With my new iPhone in hand I found myself with new file format I was unfamiliar with. Universal Scene Description is a framework to exchanging 3D data between different software developed by Pixar. The relationship between Apple and Pixar goes way back so it was no surprise the Apple iPhone has support built in for this new format and I found myself capturing and sending 3D models to others with an iPhone. The USDZ format is a ZIP package format for containing a USD 3D model and is perfect for sharing and preserving.

There is no current PRONOM signatures for identifying USD formats, so I wanted to look into creating one. This is where I ran into a problem. The current PRONOM signature syntax has no way of properly identifying the USDZ format. Let me explain.

When DROID or Siegfried is used to identify a container format such as USDZ. It will first identify the format as a ZIP file, which technically it is. This triggers the software to then refer to the container signature to see if any patterns from the files internal to the ZIP match to a known format. This is done by pointing to a specific file and a hex pattern or ascii string within the file. In the case of a USDZ the internal structure may look like this:

Listing archive: scaniverse-20210928-113055.usdz

--
Path = scaniverse-20210928-113055.usdz
Type = zip
Physical Size = 5702256

   Date      Time    Attr         Size   Compressed  Name
------------------- ----- ------------ ------------  ------------------------
2021-09-28 11:47:36 .....       297999       297999  scaniverse-20210928-113055.usdc
2021-09-28 11:47:36 .....      5403849      5403849  0/texgen_0.jpg
------------------- ----- ------------ ------------  ------------------------
2021-09-28 11:47:36            5701848      5701848  2 files

In this sample file the name of the USDZ is the same name as the internal USDC file. So the name of the USDC is variable and DROID needs a static name and path to look for patterns. The USDZ specification is clear that the only required file inside a USDZ is a USD model, anything else is ancillary and is not always going to be included. Currently the only format used is USDC, but in the future may allow a simple USD or USDA format. In addition, some of the other sample files show a very nested USDC file, making identification even more difficult.

Listing archive: Scan.usdz

--
Path = Scan.usdz
Type = zip
Physical Size = 19155195
Characteristics = Minor_Extra_ERROR

   Date      Time    Attr         Size   Compressed  Name
------------------- ----- ------------ ------------  ------------------------
2021-03-09 09:22:36 .....     19154773     19154773  /private/var/mobile/Containers/Data/Application/EFD09E66-32FB-4B08-8BED-B7E3D78FE1A8/tmp/Scan.usdc
------------------- ----- ------------ ------------  ------------------------
2021-03-09 09:22:36           19154773     19154773  1 files

The USDZ format is not the only file format which makes identification difficult through variable names and non-static patterns. An issue on GitHub has been raised to address this problem. One potential fix is to use glob patterns as suggested by the amazing Richard Lehane, creator of Siegfried. This way we could use wildcard to ignore the variable names and find any file with an extension of .USDC for example. The USDC file format has a nice 8 byte header “PXR-USDC” which is perfectly suited for identification so our container signature might look like this:

<ContainerSignature Id="1000" ContainerType="ZIP">
  <Description>USDZ 3D Package</Description>
    <Files>
	<File>
          <Path>*.usdc</Path>
              <BinarySignatures>
                  <InternalSignatureCollection>                    
	             <InternalSignature ID="300">
	                 <ByteSequence Reference="BOFoffset">
	                     <SubSequence Position="1" SubSeqMinOffset="0" SubSeqMaxOffset="0">
	                         <Sequence>50 58 52 2D 55 53 44 43</Sequence>
	                     </SubSequence>
	                 </ByteSequence>
	             </InternalSignature>
	          </InternalSignatureCollection>
             </BinarySignatures>
        </File>           
    </Files>
</ContainerSignature>

Update: I was able to get a beta version of Siegfried working with my test signature.

siegfried   : 1.11.0
scandate    : 2023-06-02T08:54:27-06:00
signature   : default.sig
created     : 2023-06-02T08:52:33-06:00
identifiers : 
  - name    : 'pronom'
    details : 'DROID_SignatureFile_V112.xml; container-signature-20230510.xml; extensions: usdz-signature-file-v1.xml; container extensions: usdz-dev1-signaturefile-20230601.xml'
---
filename : 'scaniverse-20210928-113055.usdz'
filesize : 5702256
modified : 2021-09-28T11:47:37-06:00
errors   : 
matches  :
  - ns      : 'pronom'
    id      : 'BYUdev/1'
    format  : 'USDZ 3D Package'
    version : 
    mime    : 'model/vnd.usdz+zip'
    class   : 
    basis   : 'extension match usdz; container name scaniverse-20210928-113055.usdc with byte match at 0, 8 (signature 1/2)'
    warning : 

I am still in the process of testing some beta versions of Siegfried in hopes of getting the glob matching to work, but still have more to do. Stay tuned!

Embedded WAVE, thanks HP 👋

Digital Preservation is all about identifying risks. This is done through a process which includes identification, validation, and metadata extraction. The more you know about the digital data you need to preserve over time, the more you can do to minimize those risks with the goal of making the data accessible over time.

Many formats are pretty straight forward, they are identifiable through a header and then have some binary bits or plain text that is readable by certain software. Others are more complicated. A common practice for more complex needs is to use a container. Word processing programs started out with plain text with maybe some formatting codes mixed in, then many moved to the Microsoft OLE container so you could have additional content embedded in a single file. Today file formats such as DOCX use a ZIP container, which houses all the text, images, formatting and anything else the format supports. Knowing what the format is and knowing what it may contain is important to preservation.

IM000959.JPG

I collect older digital cameras, specifically cameras with unique file formats, raw and otherwise. When I picked up a HP (Hewlett-Packard) point and shoot camera awhile back, I was initially unimpressed as it would only capture in a JPEG format and only 3 quality settings. While looking at a copy of the manual, I saw the camera was capable of capturing audio clips or voice memos for each photo taken. This can be handy when taking many photos and need a reminder about the context. This was not unique to HP, as many cameras could do this, normally a JPG was captured and the Audio would have the same name connecting the two. But when I recorded some audio on my little HP, placed the SD card in my computer, I couldn’t find the additional audio file. I also not the only one to ask about this.

There are many types of JPG files. Raw Streams, JPEG File Interchange Format (JFIF), and Exchangeable Image File Format (EXIF). Normally these formats have raster image data sprinkled with metadata. I have seen JPEG files embedded into other formats and containers, such as MP3, PDF, etc, but JPEG’s are not container formats. Or so I thought…..

View of HP Photosmart 433 folder in HP Photo & Imaging Gallery

Lets take a look at an image I took with my HP Photosmart 433. We’ll start with identification:

siegfried   : 1.10.1
scandate    : 2023-05-25T12:27:04-06:00
signature   : default.sig
created     : 2023-05-22T08:43:02-06:00
identifiers : 
  - name    : 'pronom'
    details : 'DROID_SignatureFile_V112.xml; container-signature-20230510.xml'
---
filename : 'GitHub/digicam_corpus/HP/Photosmart 433/IM000959.JPG'
filesize : 178922
modified : 2023-05-25T11:23:32-06:00
errors   : 
matches  :
  - ns      : 'pronom'
    id      : 'x-fmt/391'
    format  : 'Exchangeable Image File Format (Compressed)'
    version : '2.2'
    mime    : 'image/jpeg'
    class   : 'Image (Raster)'
    basis   : 'extension match jpg; byte match at [[0 16] [366 12] [178907 2]] (signature 2/2)'
    warning : 

IM000959.JPG was identified as x-fmt/391 which is a compressed Exchangeable Image File Format. version 2.2. Pretty straight forward. Next lets look at validation:

Jhove (Rel. 1.28.0, 2023-05-18)
 Date: 2023-05-25 12:35:16 MDT
 RepresentationInformation: GitHub/digicam_corpus/HP/Photosmart 433/IM000959.JPG
  ReportingModule: JPEG-hul, Rel. 1.5.4 (2023-03-16)
  LastModified: 2023-05-25 11:23:32 MDT
  Size: 178922
  Format: JPEG
  Status: Well-Formed and valid
  SignatureMatches:
   JPEG-hul
  ErrorMessage: Tag 41492 out of sequence
   ID: TIFF-HUL-2
   Offset: 606
  MIMEtype: image/jpeg
  JPEGMetadata: 
   CompressionType: Huffman coding, Baseline DCT
   Images: 
    Number: 1
    Image: 
     NisoImageMetadata: 
      FormatName: image/jpeg
      ByteOrder: big_endian
      CompressionScheme: JPEG
      ImageWidth: 640
      ImageHeight: 480
      ColorSpace: YCbCr
      DateTimeCreated: 2021-11-16T09:04:04
      ScannerManufacturer: Hewlett-Packard
      ScannerModelName: hp PhotoSmart 43x series
      DigitalCameraManufacturer: Hewlett-Packard
      DigitalCameraModelName: hp PhotoSmart 43x series
      FNumber: 4
      ................................
     Exif: 
      ExifVersion: 0220
      FlashpixVersion: 0100
      ColorSpace: sRGB
      ComponentsConfiguration: 1, 2, 3, 0
      CompressedBitsPerPixel: 1.568
      PixelXDimension: 640
      PixelYDimension: 480
      MakerNote: 0, 97, 48, 101, 114, 32, 78, 111, 116, 101, 115, 0, 0, 0, 0, 0
      DateTimeOriginal: 2021:11:16 09:04:04
      DateTimeDigitized: 2021:11:16 09:04:04
   ApplicationSegments: APP1, APP2, APP2, APP2, APP2, APP2, APP2, APP2, APP2, APP2, APP2, APP2, APP2, APP2, APP2, APP2

I removed a few lines to show important parts, but we get some similar information about the format, a JPEG with EXIF version 2.2. We also learn that HP improperly ordered their tags and put Tag 41492 out of sequence, but we can ignore that for now. Looking close at the output does not give us any indication of audio formats. There is a clue when we see the mention of a Flashpix version and additional Application Segments.

Since this is an image with EXIF data, lets also take a look at the output of Exiftool.

ExifTool Version Number         : 12.62
File Name                       : IM000959.JPG
Directory                       : .
File Size                       : 179 kB
File Modification Date/Time     : 2023:05:25 11:23:32-06:00
File Access Date/Time           : 2023:05:25 11:24:42-06:00
File Inode Change Date/Time     : 2023:05:25 11:24:39-06:00
File Permissions                : -rwxr-xr-x
File Type                       : JPEG
File Type Extension             : jpg
MIME Type                       : image/jpeg
Exif Byte Order                 : Little-endian (Intel, II)
Image Description               : IM000959.JPG
Make                            : Hewlett-Packard
Camera Model Name               : hp PhotoSmart 43x series
Orientation                     : Horizontal (normal)
X Resolution                    : 72
Y Resolution                    : 72
Resolution Unit                 : inches
Software                        : 1.400
Modify Date                     : 2021:11:16 09:04:04
Y Cb Cr Positioning             : Co-sited
Copyright                       : Copyright 2002-2003
Exposure Time                   : 1/29
F Number                        : 4.0
ISO                             : 100
Exif Version                    : 0220
Date/Time Original              : 2021:11:16 09:04:04
Create Date                     : 2021:11:16 09:04:04
Components Configuration        : Y, Cb, Cr, -
Compressed Bits Per Pixel       : 1.567552083
Shutter Speed Value             : 1/30
Aperture Value                  : 4.0
Exposure Compensation           : 0
Max Aperture Value              : 4.0
Subject Distance                : 1 m
Metering Mode                   : Average
Light Source                    : Unknown
Flash                           : Auto, Did not fire
Focal Length                    : 5.7 mm
Warning                         : [minor] Unrecognized MakerNotes
Flashpix Version                : 0100
Color Space                     : sRGB
Exif Image Width                : 640
Exif Image Height               : 480
Interoperability Index          : R98 - DCF basic file (sRGB)
Interoperability Version        : 0100
Digital Zoom Ratio              : 1
Subject Location                : 0
Compression                     : JPEG (old-style)
Thumbnail Offset                : 2046
Thumbnail Length                : 7112
Code Page                       : Unicode UTF-16, little endian
Used Extension Numbers          : 1, 31
Extension Name                  : Audio
Extension Class ID              : 10000100-6FC0-11D0-BD01-00609719A180
Extension Persistence           : Always Valid
Audio Stream                    : (Binary data 117820 bytes, use -b option to extract)
Image Width                     : 640
Image Height                    : 480
Encoding Process                : Baseline DCT, Huffman coding
Bits Per Sample                 : 8
Color Components                : 3
Y Cb Cr Sub Sampling            : YCbCr4:2:2 (2 1)
Aperture                        : 4.0
Image Size                      : 640x480
Megapixels                      : 0.307
Shutter Speed                   : 1/29
Thumbnail Image                 : (Binary data 7112 bytes, use -b option to extract)
Focal Length                    : 5.7 mm
Light Value                     : 8.9

Ohh, what do we have here? Exiftool mentions an audio stream. An audio stream inside the JPEG? How is this possible? The Flashpix format was originally developed by Kodak in which collaborated with HP. This was later added to the EXIF specifications. Below is an screenshot from the Exif Version 2.2 spec.

Exiftool mentioned Flashpix and additional APP2 segments. Lets take a look at the raw file in a hex editor.

Ahhh….. In one of the App2 segments we can see something familiar. A RIFF WAVE header! Lets see if we can extract the WAVE file.

exiftool -b -AudioStream IM000959.JPG > IM000959.WAV

mediainfo IM000959.WAV
General
Complete name                            : IM000959.WAV
Format                                   : Wave
Format settings                          : WaveFormatEx
File size                                : 115 KiB
Duration                                 : 10 s 681 ms
Overall bit rate mode                    : Constant
Overall bit rate                         : 88.2 kb/s

Audio
Format                                   : ADPCM
Codec ID                                 : 11
Codec ID/Hint                            : Intel
Duration                                 : 10 s 681 ms
Bit rate mode                            : Constant
Bit rate                                 : 88.2 kb/s
Channel(s)                               : 1 channel
Sampling rate                            : 22.05 kHz
Bit depth                                : 4 bits
Stream size                              : 115 KiB (100%)

MediaInfo can give us details on the embedded WAVE file, which is pretty terrible quality but is a PCM audio stream.

Embedded audio inside a raster image is not common. Most software which can render a JPEG image will most likely ignore the embedded WAVE and not even give a warning it exists. IM000959.JPG opens fine in Adobe Photoshop, but saving to a new format or making any edits will delete the WAVE file. Imagemagick also will remove the WAVE with any editing with no warning.

In order to ensure the embedded audio stream is preserved we first need to know it is there, this is where tools like exiftool can be used to extract metadata from the file and the image can be associated with having an audio stream and handled differently than any other JPEG file. More work is needed, Exiftool may mention an Audio Stream, but currently does not have the ability to pull any data from the stream.