Take the 2-minute tour ×
Mathematica Stack Exchange is a question and answer site for users of Mathematica. It's 100% free, no registration required.

I often export images(plots, matrix, arrays, etc..) from mathematica which I end up putting in word documents or uploading to the web. The problem is that I often lose the original code and I am left only with the image representing the original output.

I was thinking/considering of using one of the libraries discusses here http://stackoverflow.com/questions/3335220/embed-text-into-png but was wondering what the Mathematica community new of any functionality built into Mathematica allowing to embed text(the code) into the image.

Compatibility Table For Answers

Edit: For documentation purposes this is the code I personally use. It varies slightly from the other answers because it embeds that data into the images pixels.

share|improve this question
6  
@OleksandrR. Plot[Sin[x], {x, 0, 1}, PlotLabel -> Style[#0, Transparent]] &[] :D –  rm -rf May 24 '13 at 0:18
1  
If image is intended for web and HTML only, you can write its code in the alt or title attribute of img tag. –  BoLe May 24 '13 at 9:15
3  
PNG and JPEG also support metadata, so you could write your code there. Interesting Q by the way:) –  Ajasja May 24 '13 at 9:59
2  
If you want to make such a table I suggest you put it in a summary section at the end of your question. I'd say it's rather uncommon, though sometimes in questions where efficiency is an issue you see timing tables pop up. –  Sjoerd C. de Vries May 26 '13 at 5:38
1  
The current entries in the table are incorrect. As far as I know Photoshop doesn't read or write SVG, so the "yes" in Jens' column is incorrect. SVG can also not be converted to other formats without losing Jens' info. Another incorrect entry. My solution allows for about 4GB of code, which I would say is endless for all means and purposes. So the "no "there would seem incorrect. It can also be converted from docx (see my update). –  Sjoerd C. de Vries May 26 '13 at 18:42

7 Answers 7

up vote 18 down vote accepted

Here a quick hack for PNG images. As its Wikipedia page shows the format works with coded chunks and you can make up and insert chunk types yourself. I'm not sure how safe it is to add beyond the official end of file marker as Simon Woods suggests in his answer. It seems like a breach of the standard to me.

The following code, which more closely seems to follow the PNG standard, inserts a "mmAc" (Mathematica code) chunk before the end of file marker. A chunk consists of a four byte length coding, a four byte chunk name, the content itself and a four byte CRC32 check.

ClearAll[myGraphicsCode];

SetAttributes[myGraphicsCode, HoldFirst];

myGraphicsCode[gfun_, opts__: {}] :=
  Module[{img, pngData, extraData},
    img = Image[gfun, FilterRules[opts, Options[Image]]];
    pngData = Drop[ImportString[ExportString[img, "PNG"], "Binary"], -12];
    extraData = ToCharacterCode@Compress@Defer@gfun;
    Join[pngData, 
         IntegerDigits[Length[extraData], 256, 4], 
         ToCharacterCode@"mmAc", 
         extraData, 
         IntegerDigits[
           Hash[StringJoin["mmAc", FromCharacterCode@extraData], "CRC32"], 
           256, 4
         ], 
         {0, 0, 0, 0, 73, 69, 78, 68, 174, 66, 96, 130}
    ]
  ]

Please note that the specific capitalization of the chunk name used here is essential.

Generating the image:

Export[
  "C:\\Users\\Sjoerd\\Desktop\\Untitled-1.png", 
   myGraphicsCode[
     Plot[Sin[ x^2], {x, -3, 3}], 
     ImageResolution -> 100
   ], 
   "Binary"
]

Posting it here:

enter image description here

Getting the plot information from the image posted above:

Import["http://i.stack.imgur.com/4bEXu.png", "Binary"] /. 
   {___, a : PatternSequence[_, _, _, _], 109, 109, 65, 99, b___} :> 
  Uncompress@FromCharacterCode@Take[{b}, FromDigits[{a}, 256]]

Plot[Sin[x^2], {x, -3, 3}]

Some image editors respect the chunk, others don't. Here is a vandalized version of the above file (done in MS Paint):

enter image description here

It still works:

Import["http://i.stack.imgur.com/eA1CS.png", "Binary"] /. 
    {___, a : PatternSequence[_, _, _, _], 109, 109, 65, 99, b___} :> 
  Uncompress@FromCharacterCode@Take[{b}, FromDigits[{a}, 256]]

Plot[Sin[x^2], {x, -3, 3}]

I tested it in Photoshop 10.0.1, but it unfortunately didn't work there.


UPDATE 1

As requested by Stefan, here a step by step explanation how it's done. I'll use an update version of the above code that I used to investigate ajasja's suggestion of using standard public chunck names instead of custom ones. This to see whether Photoshop respects those (it doesn't either).

Attributes HoldFirst is set so that I can enter plot code without having it evaluated prematurily.

ClearAll[myGraphicsCode];
SetAttributes[myGraphicsCode, HoldFirst];

I want to be able to flexible set the bitmap properties of the plot. So I allowed for the options of Image to be passed through my function.

myGraphicsCode[gfun_, opts__: {}] :=
 Module[{img, pngData, extraData},
  img = Image[gfun, FilterRules[opts, Options[Image]]];

I use ExportString to export the image as a PNG to string data. This saves me temporary file handling. The image is immediately imported again, but now as a list of bytes. Mathematica closes the PNG with a standard 12 byte sequence ({0,0,0,0} (data length)+"IEND"+CRC). I chop it off and will add it back later on.

  pngData = Drop[ImportString[ExportString[img, "PNG"], "Binary"], -12];

Here the stuff for a "iTXt" chunk (see the W3 PNG definition for details):

  extraData = 
   Join[ToCharacterCode@"iTxtMathematica code", {0, 0, 0, 0, 0}, 
    ToCharacterCode@Compress@Defer@gfun];

I wrapped the plot code with Defer so that it won't be evaluated once recovered from a file's meta data. Compress converts it to a safe character range and does some compression.

Putting it all together. IntegerDigits[value, 256, 4] turns value into 4 bytes. 4 is subtracted because the length should not include the chunk name.

  Join[pngData, IntegerDigits[Length[extraData] - 4, 256, 4], 
   extraData, 

Now, the CRC32 hash is calculated and also turned into a four-byte sequence. Note that both Photoshop and MS Paint don't seem to check this. Quicktime's ImageViewer OTOH does check it and can be used therefore to verify your code. Finally, the end marker is added back.

   IntegerDigits[Hash[FromCharacterCode@extraData, "CRC32"], 256, 4], 
   {0, 0, 0, 0, 73, 69, 78, 68, 174, 66, 96, 130}]
  ]

Code for importing the meta data:

codeFinder := {___, a : PatternSequence[_, _, _, _], Sequence @@ 
              ToCharacterCode@"iTXtMathematica code", b___} :> 
  Uncompress@FromCharacterCode@Take[{b}, {5, FromDigits[{a}, 256]}]

Import["C:\\Users\\Sjoerd\\Desktop\\Untitled-1.png", "Binary"] /. codeFinder 

Note that I import as binary. I don't want and need any image conversion. What follows is a bit of pattern matching. The core of which is the chunk name "iTXt" and the keyword "Mathematica code" that I wrote into the file earlier.

The preceding a : PatternSequence[_, _, _, _] is used to catch and name the 4 length bytes. After conversion with FromDigits again, this is used to take a precise bite out of the data from the remainder of the file that was put into b. FromCharacterCode converts it to a string again, which is then returned into readable Mathematica code by Uncompress.


UPDATE 2

I tested importing graphics from Word documents. I added the above picture to a DOCX and used the following:

Import[
  "C:\\Users\\Sjoerd\\Desktop\\Doc1.docx", 
  {"ZIP", "word\\media\\image1.png", "Binary"}
]  /. codeFinder

Plot[Sin[x^2], {x, -3, 3}]

Works without a hitch.

Internal file names used by Word can be found thus:

Import["C:\\Users\\Sjoerd\\Desktop\\Doc1.docx"]

{"[Content_Types].xml", "_rels\.rels", \ "word\_rels\document.xml.rels", "word\document.xml", \ "word\theme\theme1.xml", "word\media\image1.png", \ "word\media\image2.gif", "word\settings.xml", \ "word\webSettings.xml", "word\stylesWithEffects.xml", \ "word\styles.xml", "docProps\core.xml", "word\fontTable.xml", \ "docProps\app.xml"}

Which is where I found my PNG file imported above.

share|improve this answer
    
upvote from me...very well thought out –  Stefan May 24 '13 at 22:21
    
what i don't like is the code. why do you guys do not show the steps of your solution? the code golfing here is just leading to questions that could be answered already. please explain yourself. is there a need for a function when the steps could be solely explained? I'm more interested in steps not in aggregations...to assemble a function is easy enough. –  Stefan May 24 '13 at 22:41
    
@stefan I'll try tomorrow. Off to bed now. –  Sjoerd C. de Vries May 24 '13 at 22:43
1  
Hmm, perhaps (ab)using zTXt or ` tEXt` fields will be preserved in Photoshop as well? w3.org/TR/PNG/#11textinfo –  Ajasja May 25 '13 at 7:17
1  
@ajasja I tried iTXt, but no improvement in Photoshop. It seems to throw away most metadata. MSPaint still works though. –  Sjoerd C. de Vries May 25 '13 at 12:44

When thinking about graphics formats that can be displayed in web browsers and also in Word, the first thing that comes to mind is a rasterized image. However, there is one alternative that makes including comments a complete no-brainer: SVG (scalable vector graphics).

The way you do it is similar to what cormullion suggested for EPS, except that EPS of course can't be used on the web. I don't have Word, but according to Google SVG can be displayed in Word. And SVG can be displayed in a resolution-independent fashion by most modern web browsers, too (because it's a vector format).

Since SVG files are plain text XML documents, comments can be included anywhere in the same way you add them to a web page, by simply enclosing them between <!-- and -->. So I do that below, to embed the plot command Plot[Sin[x], {x, 0, 2 Pi}] in the exported file.

The advantage of this format is that the inclusion of comments conforms fully to the official standard. I don't have to hack anything, or append stuff behind the end of the data stream, etc.

Edit

If you insist on exporting an Image (e.g., because the plot is a complicated Graphicds3D object), then you can still use the SVG format by simply replacing the definition of the plot p above with

p = Rasterize[Plot[Sin[x], {x, 0, 2 Pi}], "Image"];

Explanation:

The way this works is that ExportString creates the exact string representing the SVG content. Before writing this string to an SVG file with Export["filename.svg", ..., "Text"] I use StringReplace to insert the comment string directly after the SVG start tag <svg >. The comment always starts with the word ***Exported Comment*** so that it can be distinguished from other possible comments (although I don't hink that Mathematica would on its own add such comments). The end of the comment is uniquely identified by a comment end tag that directly follows ***Exported Comment***. That is how I recognize the included code in the StringCases command.

Updated code as a function with additional options

Options[svgExport] = {"CommentString" -> "Created by Mathematica", 
   AspectRatio -> Automatic, Background -> Automatic};

Clear[svgExport];
svgExport[name_String, gr_, opts : OptionsPattern[]] := Module[
  {
   svgCode =
    StringReplace[
       ExportString[
        First@ImportString[
          ExportString[gr, "PDF", 
           Background -> OptionValue[Background]], "PDF"],
        "SVG", Background -> OptionValue[Background]
        ],
       "<svg " ->
        "<svg viewBox='0 0 " <> StringJoin[
          Riffle[#, {" ", "' "}]] <> "width='" <> #[[1]] <> 
         "' height='" <> #[[2]] <> If[
          OptionValue[AspectRatio] === Full,
          "' preserveAspectRatio='none' ", "' "
          ]
       ] &[
     ToString /@ 
      ImageDimensions[
       Rasterize[Show[gr, ImagePadding -> 0], "Image"]]]
   },
  Export[
   name,
   StringReplace[svgCode, 
    RegularExpression["(<svg\\b[^>]*>)"] :> 
     "$1" <> "\n<!-- ***Exported Comment***\n" <> 
      OptionValue["CommentString"] <> 
      "\n***Exported Comment*** -->"],
   "Text"
   ]
  ]

In order to make the exported SVG compatible with external viewers, I handle the dimensions of the graphics more explicitly by adding them to the <svg> tag in two forms: as a viewBox and as a width/height attribute. This allows the SVG to be scaled better in web browsers, and it also helps tools like ImageMagick convert recognize the image dimensions. These things are just added for convenience, and for the same reason I now treat the included "Code" as an option named "CommentString".

So the usage of this export function is as follows:

p = Plot[Sin[x], {x, 0, 2 Pi}];

svgExport["plot.svg", p, 
 "CommentString" -> "Plot[Sin[x],{x,0,2 Pi}]"]

(* ==> "plot.svg" *)

StringCases[Import["plot.svg", "Text"], 
 "<!-- ***Exported Comment***" ~~ Shortest[code__] ~~ 
   "***Exported Comment*** -->" -> code]

(*
==> {"
 Plot[Sin[x],{x,0,2 Pi}]
 "}
*)

Additional options for exportSVG are

  • AspectRatio -> Full to allow the SVG to scale independently in horizontal and vertical directions
  • Background -> None to suppress the default (white) background of the exported plot
share|improve this answer
    
uff. sometimes the most most obvious is so near...i pretty like your solution, because it is so simple and again i've to confess to missed the wood from the trees. upvote! –  Stefan May 25 '13 at 6:29
    
I always forget about SVG - it never seemed to take off (probably due to Internet Explorer not supporting it). –  cormullion May 25 '13 at 6:52
    
I would consider this to be the best solution, except the exported svg's don't seem to be able to be imported by Inkscape. Basically I am importing and exporting images(svgs into to Inkscape) so they can be printed out. Keeping track of all the code for each image is impossible though. –  Liam William May 25 '13 at 17:03
    
On my Mac, when I drag any SVG onto Inkscape, it opens fine, and I can print it too. Sometimes there are font problems, so you may want to adapt this font outlining approach to your graphics, or use Rasterize before exporting (as I mentioned above). Inkscape development seems to have been quite slow lately, so I don't know if there's a Windows specific problem with its import abilities. But you can also drag an SVG onto Firefox and print from there... –  Jens May 25 '13 at 17:12
1  
Sure, that sounds reasonable. Can I leave that modification up to you, or do you want me to edit the answer? Ultimately, I'd say if someone edits an exported file externally without knowing what they're doing, there's no way to guarantee that the comments will be preserved anyway. It's the same exact thing with all the other answers. So at some point this becomes an open-ended problem. –  Jens May 25 '13 at 20:30

EPS files are easily hacked in the same way that Simon's answer did for PNG. The passenger text happily sits at the end, after the %EOF marker.

text = ExampleData[{"Text", "AliceInWonderland"}];

SetOptions[OpenWrite, PageWidth -> Infinity, FormatType -> OutputForm];

StringReplace[
  ExportString[l, "EPS"], 
    {"\t" -> " ", 
     "%EOF" -> "%EOF\n%" <> text}] 
  >> "test2.eps"

and later retrieved with something like:

StringCases[Import["test2.eps", "Text"], "%EOF\n%" ~~ text__ -> text]

This is similar to a technique used by applications such as Macromedia's FreeHand, which embedded the source of the document inside the EPS rendition of it.

share|improve this answer
1  
it would be more natural for this data to be just a comment in the metadata section in the beginning of the eps. –  joojaa Jul 23 at 6:42

Some image file formats such as PNG and JPG can tolerate extra data at the end of the file, so you could simply write your original code into the image file. Of course this won't survive unless the file is completely unaltered - so for example loading the file into an image viewer and saving under a new name will wipe out the extra data.

I have confirmed that a Word 2010 .docx file will preserve the original file, though to get it back you have to extract it from the Word file using a zip utility rather than exporting it from within Word. I don't know if its possible with the older .doc format.

Anyway, despite the limitations I thought might be worth sharing:

attachCode[file_String, expr_] := Block[{stream},
  stream = OpenAppend[file];
  WriteString[stream, "Embedded Code:", Compress @ expr];
  Close[stream]]

extractCode[file_String] := Uncompress @ Last @ StringSplit[
    FromCharacterCode @ BinaryReadList @ file, "Embedded Code:"]

For example, attaching the evaluation notebook to a PNG image:

Plot[Sin[x], {x, 0, 10}]

enter image description here

Export["test.png", %];

attachCode[%, NotebookGet @ EvaluationNotebook[]]

If you download the picture above, you should be able to do this:

extractCode["AVRgu.png"] // CreateDocument
share|improve this answer
    
Seems to work well - could be useful! –  cormullion May 24 '13 at 10:32
    
Another way to extract the original files from word, and probably without changing them, is to do a save as web page (a folder is created with all the images) –  P. Fonseca May 24 '13 at 11:29
    
@P.Fonseca You can import the graphics files hidden in a docx file straight from within Mathematica. See update 2 in my answer. –  Sjoerd C. de Vries May 26 '13 at 18:32

Here's a way to actually embed the code as part of the image itself. First, let's find some important code we need to save:

expr = x^2 + 3;

Now compress it and turn it into a string of numbers.

cf = Compress@expr;
data = ToCharacterCode[cf]

Load in an image and embed the compressed data into the image, then save as a new image.

lena = ColorConvert[Import["ExampleData/lena.tif"], "Grayscale"];
imgData = ImageData[lena];
{r, c} = Dimensions[imgData];
imgData[[All, 1]] = N[PadRight[data, r]/255];
newLena = Image[imgData];
Export["lenaPlus.png", newLena]

If you look closely, you'll see the first column (on the left) has been changed, this is the data.

enter image description here

To get the data back, do it in reverse:

imported = Import["lenaPlus.png"];
imgData = ImageData[imported];
getCode = imgData[[All, 1]];
Uncompress@FromCharacterCode[Round[255 getCode]]

and the answer is:

3 + x^2

Of course, this isn't really ready for prime time.. to make it generally useful you'd want to embed in color images (rather than just grayscale) and you'd want to add columns to the image (rather than replacing them as above) but the essential idea is here. I would anticipate that this kind of procedure should be robust to simple reading and saving operations in any literal image format (.tif, .png, .gif, .bmp), but it would surely be destroyed in any kind of lossy compression scheme like jpg.

share|improve this answer
    
I don't really see any significant advantage of this over taking the least significant bit of each pixel as in: blog.wolfram.com/2010/07/08/doing-spy-stuff-with-mathematica –  Ajasja May 24 '13 at 14:17
    
Ajasja - one possibile advantage would be that you do not need to change the picture (assuming you implemented adding columns instead of throwing away the first column). You could get the exact picture back by simply throwing away the added code, whereas in the link, it destroys the LSB of the image. Probably not a large advantage though. –  bill s May 24 '13 at 14:23
1  
@Ajasja I don't know enough about the formats, but this solution could probably be modified in some way so that the data could like be extracted even if the image is exported as a different format then is was originally imported. For example the data might likely stay intact if it is converted form png->jpeg or png->gif because they are physical pixels –  Liam William May 25 '13 at 17:06
1  
@liam the data will most probably not survive a png -> jpg conversion as this is a lossy transformation. –  Sjoerd C. de Vries May 26 '13 at 5:40

Embedding Code into an Image

(Thank god there is Mr. McLoone with his creative mind)

Say for instance we have a plot:

Plot[Sin[x], {x, 0, 2 Pi}]

enter image description here

To generate an image file:

Export["~/plot.png", Plot[Sin[x], {x, 0, 2 Pi}]]; 
image = Import["~/plot.png"]; DeleteFile["~/plot.png"];

(is there a shortcut for this?)

Edit (thank you to rm -rf)

miss the wood for the trees:

image = Rasterize[Plot[Sin[x], {x, 0, 2 Pi}]];

Edit end

If we say now, that we use for every color channel the least significant bit to embed the code, we first have to create a truncated variant of the original image forcing strictly 8-bits per channel:

truncImage = BitAnd[ImageData[image, "Byte"], 2^^11111110];

Next we have to convert the code into a sequence of bits and insert each of them into those empty bits; padding the rest with zeros:

code = PadRight[
    Flatten[IntegerDigits[ToCharacterCode@Compress["Plot[Sin[x],{x,0,2Pi}]"], 2,8]],     
    Apply[Times, Dimensions[ImageData[image, "Byte"]]]];

Now we have to merge the truncated image with code:

codeImage = Image[truncImage + Fold[Partition, code, 
 Reverse@Rest[Dimensions[ImageData[image, "Byte"]]]], "Byte"];

Now the code is embedded into that image. In order to extract the code, we've to reverse the process:

secCode = FromDigits[#, 2] & /@ (Partition[
 Flatten@BitAnd[ImageData[codeImage, "Byte"], 1], 8]);

and then:

Uncompress@FromCharacterCode[secCode]

=> Plot[Sin[x],{x,0,2Pi}]

or:

Uncompress@FromCharacterCode[secCode] // ToExpression

enter image description here

share|improve this answer

Anyone is welcome to edit this answer or the wiki question as necessary.

Compatibility Table For Answers

--------------Jens ..|..Bill ..|..Sjoerd ..|..Simon ..|..corm ..|..Stefan

no-end  | Y    | _   | Y      | Y      | _     | _ 
MSPaint | .    | Y   | Y      | .      | .     | Y  
 psd    | .    | Y   | .      | .      | .     | Y  
 svg    | Y    | .   | .      | .      | .     | _
 doc    | Y    | Y   | Y      | P      | Y     | Y  
 odp    | .    | .   | .      | .      | .     | _
convert | .    | .   | .      | .      | .     | _

Y - Yes . - No _- means unknown or uncertain

P - can't export the image using export dialog. Must extract image from format

Sjoerd - It appears Windows strips metadata when you copy and paste an image from something like a webbrowser

Name    | The image can be....
------------------------------------
no-end  | has no-end(size restriction) to the amount of data that can be encoded 
MS Paint| imported/exported from MS Paint without losing info
.psd    | imported/exported from photoshop without losing info
.svg    | imported/exported from inkscape .svg without losing info
.doc    | imported/exported from word .doc without losing info
.odp    | imported/exported form libreoffice .odp format
convert | converted to other formats (lossy or lossless) w/o losing info
share|improve this answer
    
you have to define the conversion type (SVG->JPG, PNG->JPG etc.) otherwise that row doesn't make much sense. Also, what do you mean by Photoshop import/export? Do you want to export to Photoshop's native format (PSD)? –  Sjoerd C. de Vries May 26 '13 at 19:16
    
@SjoerdC.deVries I was originally going to put separate rows for different format types SVG->JPG, PNG->JPG etc. but thought putting the types in the box might work better, or asterisks * with descriptions below the table if the answers are to long. Feel free to do whatever you think we will the best. –  Liam William May 26 '13 at 19:20
    
I don't get the P in the .doc line. I can right-click the picture and do a save as... from the context menu. The metadata of the resulting picture can be extracted straight away. The direct import from within Mathematica that I demonstrated in my answer is an extra convenience, not a kludge. –  Sjoerd C. de Vries May 26 '13 at 21:22
    
@SjoerdC.deVries That is what I get for just glancing over the questions. :P Fixed. –  Liam William May 26 '13 at 21:28
    
IMO it is better to put this information into the question and then flag it to be un-CWed, rather than to write an answer that isn't. BTW, I cannot understand what "no-end" means, nor "edit" (what is an "appropriate" program?). –  Oleksandr R. May 27 '13 at 16:41

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.