How can I convert a svg file to matrix or image in Matlab?

Question

Simone Cotta Ramusino 2022-11-3

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1842708-how-can-i-convert-a-svg-file-to-matrix-or-image-in-matlab

评论： Simone Cotta Ramusino 2022-11-6

I have a .svg file containing the slicing of a model. How can I import and read the slices in Matlab? They are binary slices and I need to loop over each of them to obtain the corresponding matrix. I'd rather not use Exchange functions, but that is not mandatory.

3 个评论
显示 1更早的评论隐藏 1更早的评论

Simone Cotta Ramusino 2022-11-3

I want to extract the content (93 binary slices) from the SVG file.

DGM 2022-11-4

Please provide an example of one such file.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

DGM 2022-11-4

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1842708-how-can-i-convert-a-svg-file-to-matrix-or-image-in-matlab#answer_1090998

在 MATLAB Online 中打开

multiobject.svg.fakeextension.txt

If you have raster images embedded in an SVG file, you may be able to extract them something like this:

% filename of svg file
fname = 'multiobject.svg.fakeextension.txt';
% extract image extensions and data from file
% ignore all other objects in the file
str = fileread(fname);
blockinfo = regexp(str,'data:image/(.*?);base64,([^"]*)','tokens');
for k = 1:numel(blockinfo)
    % get the info for this image
    thisext = blockinfo{k}{1};
    thisdata = blockinfo{k}{2};
    
    % decode the image file
    thisdata = matlab.net.base64decode(thisdata);
    
    % write the file using the original format extension
    fname = sprintf('myextractedfile_%04d.%s',k,thisext);
    fid = fopen(fname,'w');
    fwrite(fid,thisdata,'uint8');
    fclose(fid);
end

Now all the files can simply be read from disk.

A = imread('myextractedfile_0001.png');

imshow(A)

Without knowing what exactly is in the file, I can't know for sure if that's what you need.

13 个评论
显示 11更早的评论隐藏 11更早的评论

DGM 2022-11-4

编辑：DGM 2022-11-4

在 MATLAB Online 中打开

A screenshot of the shape doesn't tell me anything about what's actually in the file. While I gave a suggestion for reading embedded raster images, I have my doubts that the objects in the file are embedded raster images at all. I don't know why a slicer would render output as raster data in an SVG container. I would expect the output to be vector. If that's the case, then the above code won't work. Since I don't know what errors occured, I have nothing else to go on.

It's possible that they are raster images, but my shallow familiarity with the format conventions may mean that the text processing won't work on SVG files generated by applications other than the one that I used. That might be an easy thing to fix, but I'd hate to just keep guessing.

EDIT: I said I'd hate to keep guessing, but here's a guess:

blockinfo = regexp(str,'data:image/(.*?);\s*base64,\s*([^"]*)','tokens');

If it turns out that the objects are vector data, it may be a minor challenge to get it converted within MATLAB. There may be other options. I don't know yet.

If you can use your existing methods to generate a simplified version of the file or a version of the file using non-confidential information, then you could upload that. It really doesn't need more than one or two slices. The editor won't let you attach a .svg file directly, but you can either zip it or do what I did above and just put a fake extension on it.

DGM 2022-11-5

编辑：DGM 2022-11-5

在 MATLAB Online 中打开

Ah, well.

For what it's worth, this is what I came up with.

% filename of svg file
fname = 'Type_IV.svg';
% image scaling
scale = 20; % px/mm (i'm assuming these units are mm)
% extract image extensions and data from file
% ignore all other objects in the file
str = fileread(fname);
% get page geometry
% example:  <svg width="114.999998" height="19"
pagesize = regexp(str,'<svg width="(\d+\.?\d*)" height="(\d+\.?\d*)"','tokens');
pagesize = fliplr(str2double(pagesize{1})); % [y x]
% get z-height for each polygon
% i'm currently assuming that these are in sequential order
% example:  <g id="layer0" slic3r:z="1.35e-008">
zpos = regexp(str,'<g id="layer\d+" slic3r:z="(\d+\.?\d*[eE+-]*\d*)">','tokens');
zpos = str2double(vertcat(zpos{:}));
nframes = numel(zpos);
% get point list for each polygon
% example:  <polygon slic3r:type="contour" points="..."
pointdata = regexp(str,'<polygon slic3r:type="contour" points="([^"]*)"','tokens');
for c = 1:nframes
    thisblock = pointdata{c}{:};
    thisblock = strrep(thisblock,',',' ');
    thisblock = str2num(thisblock); %#ok<ST2NM>
    pointdata{c} = reshape(thisblock,2,[]).'; % [x y]
end
% build image stack using ROI tools
% this is strictly 1 page per slice, so z-geometry is currently ignored!
outpict = false([ceil(scale*pagesize) nframes]);
imshow(outpict(:,:,1))
ROI = images.roi.Polygon(gca);
for c = 1:nframes
    ROI.Position = scale*pointdata{c};
    outpict(:,:,c) = createMask(ROI);
end
% assuming that z-steps are uniform, loft z by rescaling
% this won't really help in this case, since the part is so thin
% if z shares the same resolution as x and y, it reduces to a single page
%thick = scale*range(zpos);
%outpict = logical(imresize3(single(outpict),[ceil(scale*pagesize) thick],'nearest'));
volshow(outpict)

The above will generate a binary volumetric image. As given, the z-resolution isn't respected; it's just one page per layer (thicker than in reality). In this case, that makes the effective resolution [20 20 1/(2.7E-08)] px/mm (assuming x,y,and z are all in mm), given that the z-step is 2.7E-08.

The commented code will do lofting, but I'm thinking that it may best to represent these things with a z-scale that's independent of the x,y scale.

DGM 2022-11-6

在 MATLAB Online 中打开

You should be able to open an SVG file with a text editor to see what's in it. As you can see, sometimes different software writes things in ways that a naive parser (like the one I wrote) won't expect. Note the fact that slic3r uses its own tags.

Like I said, my parsing attempt is pretty naive. it's strictly based on the file at hand and my recollection of other files -- as opposed to an in-depth understanding of the format specification and best practices for robust decoding. My use of regex to match things so literally makes this example inapplicable for files generated by other software.

I expanded the expressions a bit here. I'm not great with regex, but it's a start.

% filename of svg file
fname = 'Type_IV.svg';
% output image size
outsize = [224 2717]; % [y x]
% read the entire file
str = fileread(fname);
% get page geometry
% example:  <svg width="114.999998" height="19"
% portions in () are the tokens to extract
% (\d+\.?\d*)
% \d+		one or more of any numeric digit
% \.?		zero or one literal . character
% \d*		zero or more of any digit
pagesize = regexp(str,'<svg width="(\d+\.?\d*)" height="(\d+\.?\d*)"','tokens');
pagesize = fliplr(str2double(pagesize{1})); % [y x]
scale = outsize./pagesize; % calculate scale from output size
% get z-height for each polygon
% i'm currently assuming that these are in sequential order
% example:  <g id="layer0" slic3r:z="1.35e-008">
% (\d+\.?\d*[eE+-]*\d*)
% \d+		one or more of any digit
% \.?		zero or one literal . character
% \d*		zero or more of any digit
% [eE+-]*	zero or more of the characters within the brackets
% \d*		zero or more of any digit
zpos = regexp(str,'<g id="layer\d+" slic3r:z="(\d+\.?\d*[eE+-]*\d*)">','tokens');
zpos = str2double(vertcat(zpos{:}));
nframes = numel(zpos);
% get point list for each polygon
% example:  <polygon slic3r:type="contour" points="..."
% ([^"]*)
% [^"]*		zero or more characters which are not "
pointdata = regexp(str,'<polygon slic3r:type="contour" points="([^"]*)"','tokens');
for c = 1:nframes
    thisblock = pointdata{c}{:};
    thisblock = strrep(thisblock,',',' ');
    thisblock = str2num(thisblock); %#ok<ST2NM>
    pointdata{c} = reshape(thisblock,2,[]).'; % [x y]
end
% build image stack using ROI tools
% this is strictly 1 page per slice, so z-geometry is currently ignored!
outpict = false([ceil(scale.*pagesize) nframes]);
imshow(outpict(:,:,1)) % displaying one page to set the axes geometry
ROI = images.roi.Polygon(gca); % create a polygon ROI object in the current axes
for c = 1:nframes
    ROI.Position = fliplr(scale).*pointdata{c}; % set the vertex list of the polygon
    outpict(:,:,c) = createMask(ROI); % convert the polygon to a 2D binary image
end

I rearranged the use of the scale parameter. In this case, you specify the output geometry instead of the scale. The result should be 224x2717x93.

Simone Cotta Ramusino 2022-11-6

在 MATLAB Online 中打开

Thank you @DGM, your patience is really appreciated.

So, if I get this right, you "simply" created a syntax that materially recognizes and reads the tokens of this .svg file; and even if it could not be multipurpose, should it always work for these kind of .svg files created with slic3r?

Then you define the page geometry and you "plot" the needed shape over it. I will always have a loss of information, like an aliasing-effect, right? This is due to the attempt to define cruved lines with pixels, I assume..even if I try:

ROI=image.roi.FreeHand(gca);

nothing changes.

Finally, the last real doubt: why did you define the scale array? I mean, at line 42, couldn't one write this:

outpict = false([ceil(outsize) nframes]);

and, in the following loop:

for c = 1:nframes
    ROI.Position = fliplr(outsize./pagesize).*pointdata{c}; % set the vertex list of the polygon
    outpict(:,:,c) = createMask(ROI); % convert the polygon to a 2D binary image
end

The outsize array is not fixed and I always know how to fill it, so scale looks unnecessary..or am I misinterpreting something?

DGM 2022-11-6

Yeah, I basically looked at the file and assumed that other files generated by the same software would have the same formatting. Unless slic3r happens to add something other than polygon objects to some files, I tentatively assume that it should work. That's an approximation based on a single observation.

Swapping images.roi.Polygon() with images.roi.Freehand() doesn't really change anything, since their position properties are compatibly-oriented. The only real difference between the two is in how they're created when doing so interactively (with the mouse). Since they're being used programmatically, they behave the same.

Yes, I suppose you could rewrite it that way as well. I just saw potential value in knowing the factors which map between part geometry and image geometry. That, and it's a bit of a holdover from the prior code. You're free to adapt it however you feel suits your needs.

Simone Cotta Ramusino 2022-11-6