Beefy Boxes and Bandwidth Generously Provided by pair Networks
XP is just a number
 
PerlMonks  

Re^2: Matching Images

by LanX (Canon)
on Nov 22, 2010 at 13:45 UTC ( #872964=note: print w/ replies, xml ) Need Help??


in reply to Re: Matching Images
in thread Matching Images

> including comparing different image formats provided they contain the same picture

Really? AFAIK png and jpg use compression methods resulting in unpredictable information loss. IMHO converting them to a raw format will not often produce identical data.

Or am I missing a special magic feature?

I think I would rather try to calculate a diff and test the average distance for a threshold passed. Not sure if the result of ->compare can be used for this.

Cheers Rolf


Comment on Re^2: Matching Images
Download Code
Re^3: Matching Images
by BrowserUk (Pope) on Nov 22, 2010 at 14:18 UTC
    Really? IMHO png and jpg use compression methods resulting in unpredictable information loss. IMHO converting them to a raw format will not often produce identical data.

    Really. And to resolve your doubts please try it.

    The "secret" such as it is, is that the raw format used for the comparison is neither jpg nor png, but libgd's own internal format, which is an uncompressed, lossless format. The images of all formats supported by GD are converted fully expanded into this internal format, and it is this binary representation, returned by the $image->gd method that is being compared in the code I posted. As such, even individual pixel differences are detected.

    By way of demonstration, in my post above, I took a jpg image of a mandrill monkey and used an image editor to convert it to a compressed but lossless .png. I then compared those two images and no discrepancies were found.

    By way of further demonstration, using the slightly expanded version below, I then edited the ,png version by changing the color of a single pixel on the mandrill's nose, from it's original near white color to pure red. and then compared that against the original .jpg:

    c:\test>cmpImages mandrill.png mandrill-2.png images are different Process pixels? Difference at (665:482): [248 248 238] .v. [255 0 0]

    As you can see, that single pixel difference was correctly detected by the raw binary compare, and then subsequently isolated using a pixel-by-pixel comparison. Please try it to convince yourself that it is correct.

    But do bear in mind my warning above. If, for example, you were to take a ,png image and convert it to a lossy image format such as .jpg, then although the images may appear superficially similar to the human eye, they may be quite different in actuality. This due to the jpeg algorithms practice of substituting a single color for many, closely aligned but mathematically different colors in order to reduce palette sizes and aid compression.

    Such lossy transforms will be correctly detected as different, even though cursory, by-eye inspections might adjudge them to be the same.

    The expanded code:

    #! perl -slw use strict; use GD; GD::Image->trueColor( 1 ); my $img1 = GD::Image->new( $ARGV[ 0 ] ); my $img2 = GD::Image->new( $ARGV[ 1 ] ); die "Different dimensions" unless $img1->width == $img2->width and $img1->height == $img2->height; my $raw1 = $img1->gd; my $raw2 = $img2->gd; if( $raw1 ne $raw2 ) { print "images are different"; printf "Process pixels? "; <STDIN>; for my $y ( 0 .. $img1->height -1 ) { for my $x ( 0 .. $img1->width -1 ) { my $p1 = $img1->getPixel( $x, $y ); my $p2 = $img2->getPixel( $x, $y ); if( $p1 != $p2 ) { print "Difference at ($x:$y): [@{[ $img1->rgb( $p1 ) ] +}] .v. [@{[ $img2->rgb( $p2 ) ]}]"; } } } } else { print "Images are the same"; } __END__

    Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
    "Science is about questioning the status quo. Questioning authority".
    In the absence of evidence, opinion is indistinguishable from prejudice.
      > and used an image editor to convert it to a compressed but lossless .png.

      my point was that in "reality" lossless conversions are the exception and not the rule. IIRC the compression factor of a JPG is freely choosable.

      Cheers Rolf

        my point

        If information is "lost" through a lossy conversion, the images are different, and the code I posted will detect those differences.

        Note well: The OP didn't ask to find images that "looked the same" or were "substantially similar", or any other form of fuzzy comparison.

        He asked for:

        I would like my script to compare two images and tell me if even one single pixel is slightly different in one of them.

        Which is exactly what the code I posted--as opposed to your theory--does.

        Hence, your "point" is pointless.


        Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
        "Science is about questioning the status quo. Questioning authority".
        In the absence of evidence, opinion is indistinguishable from prejudice.

      "I then edited the ,png version by changing the color of a single pixel on the mandrill's nose, from it's original near white color to pure red."

      Neat!

      Where d'get them paint brushes man.

Re^3: Matching Images
by ikegami (Pope) on Nov 22, 2010 at 16:10 UTC
    PNG is lossless. It's reason for being was to replace GIF, a lossless format.
      A right! Thx, I had wavelet stuff in mind like DjVu...

      So if the PNG is created from a JPG original it can work, the other direction or just another source would IMHO likely fail.

      (Though I have no stats about compression defaults for JPEGs)

      Cheers Rolf

Log In?
Username:
Password:

What's my password?
Create A New User
Node Status?
node history
Node Type: note [id://872964]
help
Chatterbox?
and the web crawler heard nothing...

How do I use this? | Other CB clients
Other Users?
Others examining the Monastery: (5)
As of 2014-07-26 12:28 GMT
Sections?
Information?
Find Nodes?
Leftovers?
    Voting Booth?

    My favorite superfluous repetitious redundant duplicative phrase is:









    Results (176 votes), past polls