file/magic/Magdir/jpeg

#------------------------------------------------------------------------------
# $File: jpeg,v 1.15 2009/09/19 16:28:10 christos Exp $
# JPEG images
# SunOS 5.5.1 had
#
#	0	string		\377\330\377\340	JPEG file
#	0	string		\377\330\377\356	JPG file
#
# both of which turn into "JPEG image data" here.
#
0	beshort		0xffd8		JPEG image data
!:mime	image/jpeg
!:apple	8BIMJPEG
!:strength +1
>6	string		JFIF		\b, JFIF standard
# The following added by Erik Rossen <rossen@freesurf.ch> 1999-09-06
# in a vain attempt to add image size reporting for JFIF.  Note that these
# tests are not fool-proof since some perfectly valid JPEGs are currently
# impossible to specify in magic(4) format.
# First, a little JFIF version info:
>>11	byte		x		\b %d.
>>12	byte		x		\b%02d
# Next, the resolution or aspect ratio of the image:
#>>13	byte		0		\b, aspect ratio
#>>13	byte		1		\b, resolution (DPI)
#>>13	byte		2		\b, resolution (DPCM)
#>>4	beshort		x		\b, segment length %d
# Next, show thumbnail info, if it exists:
>>18	byte		!0		\b, thumbnail %dx
>>>19	byte		x		\b%d

# EXIF moved down here to avoid reporting a bogus version number,
# and EXIF version number printing added.
#   - Patrik R=E5dman <patrik+file-magic@iki.fi>
>6	string		Exif		\b, EXIF standard
# Look for EXIF IFD offset in IFD 0, and then look for EXIF version tag in EXIF IFD.
# All possible combinations of entries have to be enumerated, since no looping
# is possible. And both endians are possible...
# The combinations included below are from real-world JPEGs.
# Little-endian
>>12	string		II		
# IFD 0 Entry #5:
>>>70	leshort		0x8769          
# EXIF IFD Entry #1:
>>>>(78.l+14)	leshort	0x9000		
>>>>>(78.l+23)	byte	x		%c
>>>>>(78.l+24)	byte	x		\b.%c
>>>>>(78.l+25)	byte	!0x30		\b%c
# IFD 0 Entry #9:
>>>118	leshort		0x8769          
# EXIF IFD Entry #3:
>>>>(126.l+38)	leshort	0x9000		
>>>>>(126.l+47)	byte	x		%c
>>>>>(126.l+48)	byte	x		\b.%c
>>>>>(126.l+49)	byte	!0x30		\b%c
# IFD 0 Entry #10
>>>130	leshort		0x8769          
# EXIF IFD Entry #3:
>>>>(138.l+38)	leshort	0x9000		
>>>>>(138.l+47)	byte	x		%c
>>>>>(138.l+48)	byte	x		\b.%c
>>>>>(138.l+49)	byte	!0x30		\b%c
# EXIF IFD Entry #4:
>>>>(138.l+50)	leshort	0x9000		
>>>>>(138.l+59)	byte	x		%c
>>>>>(138.l+60)	byte	x		\b.%c
>>>>>(138.l+61)	byte	!0x30		\b%c
# EXIF IFD Entry #5:
>>>>(138.l+62)	leshort	0x9000		
>>>>>(138.l+71)	byte	x		%c
>>>>>(138.l+72)	byte	x		\b.%c
>>>>>(138.l+73)	byte	!0x30		\b%c
# IFD 0 Entry #11
>>>142	leshort		0x8769          
# EXIF IFD Entry #3:
>>>>(150.l+38)	leshort	0x9000		
>>>>>(150.l+47)	byte	x		%c
>>>>>(150.l+48)	byte	x		\b.%c
>>>>>(150.l+49)	byte	!0x30		\b%c
# EXIF IFD Entry #4:
>>>>(150.l+50)	leshort	0x9000		
>>>>>(150.l+59)	byte	x		%c
>>>>>(150.l+60)	byte	x		\b.%c
>>>>>(150.l+61)	byte	!0x30		\b%c
# EXIF IFD Entry #5:
>>>>(150.l+62)	leshort	0x9000		
>>>>>(150.l+71)	byte	x		%c
>>>>>(150.l+72)	byte	x		\b.%c
>>>>>(150.l+73)	byte	!0x30		\b%c
# Big-endian
>>12	string		MM		
# IFD 0 Entry #9:
>>>118	beshort		0x8769          
# EXIF IFD Entry #1:
>>>>(126.L+14)	beshort	0x9000		
>>>>>(126.L+23)	byte	x		%c
>>>>>(126.L+24)	byte	x		\b.%c
>>>>>(126.L+25)	byte	!0x30		\b%c
# EXIF IFD Entry #3:
>>>>(126.L+38)	beshort	0x9000		
>>>>>(126.L+47)	byte	x		%c
>>>>>(126.L+48)	byte	x		\b.%c
>>>>>(126.L+49)	byte	!0x30		\b%c
# IFD 0 Entry #10
>>>130	beshort		0x8769          
# EXIF IFD Entry #3:
>>>>(138.L+38)	beshort	0x9000		
>>>>>(138.L+47)	byte	x		%c
>>>>>(138.L+48)	byte	x		\b.%c
>>>>>(138.L+49)	byte	!0x30		\b%c
# EXIF IFD Entry #5:
>>>>(138.L+62)	beshort	0x9000		
>>>>>(138.L+71)	byte	x		%c
>>>>>(138.L+72)	byte	x		\b.%c
>>>>>(138.L+73)	byte	!0x30		\b%c
# IFD 0 Entry #11
>>>142	beshort		0x8769          
# EXIF IFD Entry #4:
>>>>(150.L+50)	beshort	0x9000		
>>>>>(150.L+59)	byte	x		%c
>>>>>(150.L+60)	byte	x		\b.%c
>>>>>(150.L+61)	byte	!0x30		\b%c
# Here things get sticky.  We can do ONE MORE marker segment with
# indirect addressing, and that's all.  It would be great if we could
# do pointer arithemetic like in an assembler language.  Christos?
# And if there was some sort of looping construct to do searches, plus a few
# named accumulators, it would be even more effective...
# At least we can show a comment if no other segments got inserted before:
>(4.S+5)	byte		0xFE
>>(4.S+8)	string		>\0		\b, comment: "%s"
# FIXME: When we can do non-byte counted strings, we can use that to get
# the string's count, and fix Debian bug #283760
#>(4.S+5)	byte		0xFE		\b, comment
#>>(4.S+6)	beshort		x		\b length=%d
#>>(4.S+8)	string		>\0		\b, "%s"
# Or, we can show the encoding type (I've included only the three most common)
# and image dimensions if we are lucky and the SOFn (image segment) is here:
>(4.S+5)	byte		0xC0		\b, baseline
>>(4.S+6)	byte		x		\b, precision %d
>>(4.S+7)	beshort		x		\b, %dx
>>(4.S+9)	beshort		x		\b%d
>(4.S+5)	byte		0xC1		\b, extended sequential
>>(4.S+6)	byte		x		\b, precision %d
>>(4.S+7)	beshort		x		\b, %dx
>>(4.S+9)	beshort		x		\b%d
>(4.S+5)	byte		0xC2		\b, progressive
>>(4.S+6)	byte		x		\b, precision %d
>>(4.S+7)	beshort		x		\b, %dx
>>(4.S+9)	beshort		x		\b%d
# I've commented-out quantisation table reporting.  I doubt anyone cares yet.
#>(4.S+5)	byte		0xDB		\b, quantisation table
#>>(4.S+6)	beshort		x		\b length=%d
#>14	beshort		x		\b, %d x
#>16	beshort		x		\b %d

# HSI is Handmade Software's proprietary JPEG encoding scheme
0	string		hsi1		JPEG image data, HSI proprietary

# From: David Santinoli <david@santinoli.com>
0	string		\x00\x00\x00\x0C\x6A\x50\x20\x20\x0D\x0A\x87\x0A	JPEG 2000 image data

# Type: JPEG 2000 codesream
# From: Mathieu Malaterre <mathieu.malaterre@gmail.com>
0	belong		0xff4fff51						JPEG 2000 codestream
45	beshort		0xff52