loading images without compression

Question

Thijs Boeree 2023-11-12

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2046100-loading-images-without-compression

评论： DGM 2023-11-21

Hello there,

I've made a for loop for filling a struct with png images, this because i need to recall the same images a view times in my application. If i recall my images from the struct i get the images but there seems to be a jpg compression on the png file? This is how i load the struct:

  imageDir = 'Edwin\files\';  % Map met de afbeeldingen
                imageExt = '.png';         % Extensie van de afbeeldingen
                % Lijst van afbeeldingsnamen zonder extensie
                imageNames = {'edammer', 'edammer_selected', 'emmentaler', 'emmentaler_selected', 'beemster', 'beemster_selected', ...
                    'gorgonzola', 'gorgonzola_selected', 'goudse', 'goudse_selected', 'gruyere', 'gruyere_selected', 'manchego', ...
                    'manchego_selected', 'noordwoudse', 'noordwoudse_selected', 'roquefort', 'roquefort_selected', ...
                    'kaasfondue3','kaasfondue4','kaasfondue5','kaasfondue6','kaasfondue7','kaasfondue8','kaasfondue9', ...
                    'edammer_vlag', 'edammer_vlag_selected', 'gorgonzola_vlag', 'gorgonzola_vlag_selected', 'manchego_vlag', ...
                    'manchego_vlag_selected', 'roquefort_vlag', 'roquefort_vlag_selected', 'haarlem_vlag', 'haarlem_vlag_selected', ...
                    'basel_vlag', 'basel_vlag_selected', 'gruyere_vlag', 'gruyere_vlag_selected', 'goudse_vlag', 'goudse_vlag_selected', ...
                    'emmentaler_vlag', 'emmentaler_vlag_selected', 'kaasfondue33','kaasfondue34','kaasfondue35','kaasfondue36','erik0','erik1', ...
                    'erik2','erik3','erik4','erik5','erik6','erik7'};
                % Loop door de lijst van afbeeldingsnamen en laad de afbeeldingen
                for i = 1:length(imageNames)
                    imageName = imageNames{i};
                    imagePath = [imageDir, imageName, imageExt];
                    app.imageStruct.(imageName) = imread(imagePath,"png", 'BackgroundColor', [0.94 0.94 0.94]);
                end
             imageStructfile = app.imageStruct;
             save("Edwin\files\imageStruct.mat",'imageStructfile');

And when i load an image from the struct it looks like this:

app.image_Erik.ImageSource = app.imageStruct.erik0;

Can anyone tell me if i am doing something wrong.

If i load the image with app.image_Erik.ImageSource = imread('files/erik0.png'); then i don't see any compression...

Best regards,

Thijs

2 个评论
显示无隐藏无

Matt J 2023-11-12

编辑：Matt J 2023-11-12

there seems to be a jpg compression on the png file?

Please show the indications of that to us. Ideally, you would run your code here online so we can see the result.

DGM 2023-11-12

编辑：DGM 2023-11-12

It may also help to know how the images were created. For example, if they're color photographs, then chances are that they were JPG to begin with.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Walter Roberson 2023-11-12

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2046100-loading-images-without-compression#answer_1351320

在 MATLAB Online 中打开

                    app.imageStruct.(imageName) = imread(imagePath,"png", 'BackgroundColor', [0.94 0.94 0.94]);

The fact that you are specifying the backgroundcolor tells us that your png images are not rgb images. They are probably indexed images -- and you are failing to read and apply the colormap.

29 个评论
显示 27更早的评论隐藏 27更早的评论

Walter Roberson 2023-11-12

在 MATLAB Online 中打开

  imageDir = 'Edwin\files\';  % Map met de afbeeldingen
                imageExt = '.png';         % Extensie van de afbeeldingen
% Lijst van afbeeldingsnamen zonder extensie
imageNames = {'edammer', 'edammer_selected', 'emmentaler', 'emmentaler_selected', 'beemster', 'beemster_selected', ...
    'gorgonzola', 'gorgonzola_selected', 'goudse', 'goudse_selected', 'gruyere', 'gruyere_selected', 'manchego', ...
    'manchego_selected', 'noordwoudse', 'noordwoudse_selected', 'roquefort', 'roquefort_selected', ...
    'kaasfondue3','kaasfondue4','kaasfondue5','kaasfondue6','kaasfondue7','kaasfondue8','kaasfondue9', ...
    'edammer_vlag', 'edammer_vlag_selected', 'gorgonzola_vlag', 'gorgonzola_vlag_selected', 'manchego_vlag', ...
    'manchego_vlag_selected', 'roquefort_vlag', 'roquefort_vlag_selected', 'haarlem_vlag', 'haarlem_vlag_selected', ...
    'basel_vlag', 'basel_vlag_selected', 'gruyere_vlag', 'gruyere_vlag_selected', 'goudse_vlag', 'goudse_vlag_selected', ...
    'emmentaler_vlag', 'emmentaler_vlag_selected', 'kaasfondue33','kaasfondue34','kaasfondue35','kaasfondue36','erik0','erik1', ...
    'erik2','erik3','erik4','erik5','erik6','erik7'};
num_images = length(imageNames);
num_with_alpha = 0;
% Loop door de lijst van afbeeldingsnamen en laad de afbeeldingen
for i = 1:num_images
    imageName = imageNames{i};
    imagePath = [imageDir, imageName, imageExt];
    [thisimg, thismap, thisalpha] = imread(imagePath,"png", 'BackgroundColor', [0.94 0.94 0.94]);
    if isempty(thismap)
        rgbimg = thisimg;
    else
        rgbimg = ind2rgb(thisimg, thismap);
    end
    if ~isempty(thisalpha); num_with_alpha = num_with_alpha + 1; end
    app.imageStruct.(imageName) = rgbimg;
end
if num_with_alpha ~= 0
    fprintf('Note: ignored alpha properties on %d of %d images\n', num_with_alpha, num_images );
end
imageStructfile = app.imageStruct;
save("Edwin\files\imageStruct.mat",'imageStructfile');

Thijs Boeree 2023-11-13

移动：Stephen23 2023-11-14

在 MATLAB Online 中打开

My example picture,

And my example code:

classdef app2 < matlab.apps.AppBase
    % Properties that correspond to app components
    properties (Access = public)
        UIFigure       matlab.ui.Figure
        image_Hanna_6  matlab.ui.control.Image
        image_Hanna_5  matlab.ui.control.Image
        image_Hanna_4  matlab.ui.control.Image
    end
    % Callbacks that handle component events
    methods (Access = private)
        % Code that executes after component creation
        function startupFcn(app)
            app2.imageStruct = struct();
            app.image_Hanna_5.ImageSource = imread("Edwin\files\Hanna_complete.png");
            imageDir = 'Edwin\files\';  % Map met de afbeeldingen
            imageExt = '.png';         % Extensie van de afbeeldingen
                % Lijst van afbeeldingsnamen zonder extensie
                imageNames = {'Hanna_complete'};
                % Loop door de lijst van afbeeldingsnamen en laad de afbeeldingen
                for i = 1:length(imageNames)
                    imageName = imageNames{i};
                    imagePath = [imageDir, imageName, imageExt];
                    app2.imageStruct.(imageName) = imread(imagePath,"png", 'BackgroundColor', [0.94 0.94 0.94]);
                end
            
            
            
            app.image_Hanna_6.ImageSource = app2.imageStruct.Hanna_complete;
        end
    end
    % Component initialization
    methods (Access = private)
        % Create UIFigure and components
        function createComponents(app)
            % Create UIFigure and hide until all components are created
            app.UIFigure = uifigure('Visible', 'off');
            app.UIFigure.Position = [100 100 1308 491];
            app.UIFigure.Name = 'MATLAB App';
            % Create image_Hanna_4
            app.image_Hanna_4 = uiimage(app.UIFigure);
            app.image_Hanna_4.ScaleMethod = 'none';
            app.image_Hanna_4.Position = [9 4 417 482];
            app.image_Hanna_4.ImageSource = 'Hanna_complete.png';
            % Create image_Hanna_5
            app.image_Hanna_5 = uiimage(app.UIFigure);
            app.image_Hanna_5.ScaleMethod = 'none';
            app.image_Hanna_5.Position = [436 4 417 482];
            % Create image_Hanna_6
            app.image_Hanna_6 = uiimage(app.UIFigure);
            app.image_Hanna_6.ScaleMethod = 'none';
            app.image_Hanna_6.Position = [862 4 417 482];
            % Show the figure after all components are created
            app.UIFigure.Visible = 'on';
        end
    end
    % App creation and deletion
    methods (Access = public)
        % Construct app
        function app = app2
            % Create UIFigure and components
            createComponents(app)
            % Register the app with App Designer
            registerApp(app, app.UIFigure)
            % Execute the startup function
            runStartupFcn(app, @startupFcn)
            if nargout == 0
                clear app
            end
        end
        % Code that executes before app deletion
        function delete(app)
            % Delete UIFigure when app is deleted
            delete(app.UIFigure)
        end
    end
end

In the first imagebox, the file is loaded from the app designer, the second one via imread, the third one from a struct file.

Best regards,

Thijs

DGM 2023-11-13

在 MATLAB Online 中打开

That's correct. If imread() is called with a single output argument, and it's asked to load an RGBA PNG file, it will not return the original RGB content. It will reduce the RGBA image to RGB by composing it with a solid-color matting. The default color will be black, though the color can be specified explicitly as you're doing here.

% collapse the image by composing it with a colored matting

inpict = imread('image.png','backgroundcolor',[1 0.8 0.3]);

imshow(inpict,'border','tight')

The question is whether that's intended. I see in some places you're calling imread() with one output, and some places, you're using three. In the case you're using three output arguments, you're discarding alpha and using the RGB content alone. That's a huge gamble, because you're assuming that the hidden RGB content is meant to be seen. There's no telling what's in the transparent regions. It's often garbage.

% preserve the original RGB content, but ignore alpha

[inpict,~,alpha] = imread('image.png');

imshow(inpict,'border','tight')

Consider another example. This is a green ellipse with transparency. What's hidden in the transparent regions?

In this case, there aren't any transparent regions left anymore.

% collapse the image by composing it with a colored matting

inpict = imread('tiltellipse.png','backgroundcolor',[1 0.8 0.3]);

imshow(inpict,'border','tight')

In this case, we find that there isn't necessarily any correlation between the apparent object content and the content of the RGB channels alone. The object content exists entirely in alpha. Discarding alpha without composition destroys the image.

% preserve the original RGB content, but ignore alpha

[inpict,~,alpha] = imread('tiltellipse.png');

imshow(inpict,'border','tight')

If you want to preserve the original RGB content without alteration, you must call imread() with all three outputs. If you want to preserve transparency information, you will then have to contend with the fact that nothing in base MATLAB or IPT makes concession for images with transparency. Imshow() can't display them without jumping through hoops, and transporting monolithic RGBA images becomes a problem because of the fact that the few ways they can be used (e.g. by imshow()/imwrite()) requires the alpha to be split into a separate array.

So the question is what you want to do with the image, and whether this is all supposed to be permanent alteration or just something that's done for sake of viewing it. If you're discarding alpha as in the second example, you're essentially destroying the image by throwing away necessary information. Operating on it or saving it would likely be a waste of time. If you want the image composed with a fixed, solid color background, as in the first example, then you can do that via the call to imread(), or you could do it after the fact -- but know that you're still losing transparency information.

Maybe you're just trying to display images without alteration and need to find some way to represent transparent content for temporary visualization purposes?

DGM 2023-11-13

编辑：DGM 2023-11-13

在 MATLAB Online 中打开

image.png

I'm not familiar with appdesigner stuff, since it won't even work properly in my environment. I don't really know how the image displayer behaves. That said, I recalled that uiimage objects behave inconsistently depending on how they're used. See:

https://www.mathworks.com/matlabcentral/answers/1734305-how-to-show-an-image-with-transparent-background-using-uiimage#answer_979565

To summarize, uiimage() can only handle grayscale or transparency if it reads it directly from a file. If the image comes from the workspace instead, it will not handle either properly, and there is no accessible alphadata property that could be used as one would with image(). If using uiimage() with images from the workspace, any composition must be performed prior to feeding it to uiimage(), since uiimage() only supports RGB, not I/IA/RGBA.

How would composition be handled? That depends what's adequate for your needs. Backing up for a moment, consider what was happening before when the image was loaded with imread():

% let imread() do the matting composition

bgc = [0.94 0.94 0.94]; % approximate figure color

composed = imread('image.png','backgroundcolor',bgc); % RGB

imshow(composed,'border','tight')

In this case, we're basically doing a composition with a solid color matting. We could do the same thing by doing the matting composition outside of imread().

% do the composition manually using base tools

bgc = [0.94 0.94 0.94]; % approximate figure color

[FG,~,FGa] = imread('image.png'); % RGB+A

FGa = im2double(FGa);

composed = FGa.*im2double(FG) + (1-FGa).*permute(bgc,[1 3 2]);

composed = im2uint8(composed); % RGB

imshow(composed,'border','tight')

If we wanted to display the image with its transparency, composited against the figure background (or whatever is rendered beneath it in the uistack, we could do it with image objects like so. This is what happens internally when you call uiimage(filename), but as far as I know, there's no other way to do this with uiimage() due to the inconsistency in its behavior and the lack of accessible AlphaData properties. That said, I'm not super familiar with using uiimage(), so maybe I'm missing something.

% do the composition in-figure using image() objects (not uiimage())

% normally, the default figure background is uint8(240), but not on the forum

[FG,~,FGa] = imread('image.png'); % RGB+A

hi = imshow(FG,'border','tight');

hi.AlphaData = FGa;

For other ideas like the checkerboard matting, the thread I linked above has some examples. See also:

In-figure composition, external solid-color and checkerboard matting; base/IPT or MIMT tools

https://www.mathworks.com/matlabcentral/answers/1773920-displaying-alpha-values-written-into-an-image-file#answer_1308026

Checkerboard matting using MIMT joinalpha(), alphasafe(), or imshow2()

https://www.mathworks.com/matlabcentral/answers/1814745-i-have-an-image-that-is-234x432x4-size-wise-i-tried-imshow-and-image-but-it-doesn-t-work-what#answer_1064560

Mask generation and checkerboard matting using base/IPT tools only

https://www.mathworks.com/matlabcentral/answers/890417-i-want-to-remove-black-color-from-my-image-and-make-the-background-transparent-so-that-i-can-paste-i

Some of those answers use MIMT tools, but I also cover base/IPT alternatives.

DGM 2023-11-14

编辑：DGM 2023-11-14

在 MATLAB Online 中打开

In $MLROOT/toolbox/matlab/uitools/uicomponents/components/+matlab/+ui/+internal/IconUtils.m

function iconString = getURLFromCData(cdata)
    % create ImageComp directory under tempdir           
    tmpDir = fullfile(tempdir, 'ImageComp');            
    [~,~,~] = mkdir(tmpDir);
    % create tempfile for image writing
    tmpFile = fullfile([tempname(tmpDir), '.jpg']);
    % convert CData to JPG file
    imwrite(cdata, tmpFile);
    
    % get URL for the tmpFile created
    iconString = matlab.ui.internal.URLUtils.getURLToUserFile(tmpFile, false);
end

When fed raster image data, it literally writes it to a garbage-tier 70% 4:2:0 downsampled temporary JPG and reads it back. What the hell?

I guess that means if you want to display raster image data with uiimage(), you have to add an extra layer of what should already be completely unnecessary levels of babysitting:

Make sure any single-channel images are expanded to RGB
Make sure any logical or signed integer images are converted to uint8 or double
Write the modified image to a temporary file using some lossless format (e.g. PNG)
Feed the temp file name to uiimage()
Deal with cleaning up piles of temporary images as necessary

I guess if you're forced into a workflow using temporary files, that does give you the opportunity to use any alpha content in the temporary PNG -- since uiimage() seems to only support alpha when reading from a file. That's not really much of a silver lining.

I suppose you could alternatively change that part of IconUtils.m to use PNG instead of JPG for the temp file, but that wouldn't give you the opportunity to pass through any available alpha content. There is mention in some of the documentation that RGBA inputs are supported, but I haven't traced through and figured out which cases (if any) actually do, or where any attached alpha winds up. Obviously, if it uses a JPG temp file in certain cases, it can't support alpha, and external observation confirms that. Also consider that modifying IconUtils.m wouldn't fix the problem for anyone else who might use your code.

Unless someone else knows of a good explanation/excuse or a better workaround, I'd recommend filing a bug report. It won't fix the problem in time for your needs (or ever), but I don't know how anyone could consider this to be a feature.

DGM 2023-11-14

编辑：DGM 2023-11-14

It's internal to uiimage() stuff, not imread(). I haven't sussed out everything that it does. If you tell uiimage() to open a filename, it uses a particular routine to read that image, handle its potentially various forms (depth,class,alpha) and render it. If you have raster image data (e.g. from imread()) and you feed that to uiimage(), it converts it to a JPG file and then somehow uses the tempfile that it created. I'm assuming that it probably ends up using the same code as it would normally use when reading an image file in the prior case.

I'm calling it a bug because I don't think it's acceptable to be reading/writing low-quality JPG temp files as part of a graphics rendering routine. It seems so absurd, but it's apparently been this way long enough that I have to ask if I'm the crazy one.

One thing you should know is that whenever you're writing a JPG in MATLAB (e.g. imwrite(), saveas(), etc), it's probably a lower-quality JPG than you expect. Default is either 70% or 75% quality (don't remember). While that could be changed in direct calls to imwrite(), you don't have that opportunity when using saveas(), or internal calls like the one in IconUtils.m. Furthermore, the default is 4:2:0 chroma subsampled, meaning you lose 75% of your color information. That really shows up. As far as I know, that's simply not something that can be configured. While the lossy nature of JPG needs to be weighed in any context, MATLAB makes it more costly than I think most people expect. I think default JPG settings in GIMP are 90% with 4:4:4 downsampling (i.e. none). I imagine Photoshop is similar.

DGM 2023-11-14

编辑：DGM 2023-11-14

在 MATLAB Online 中打开

Consider this example of exploiting the direct file reading to handle images in the workspace. I've attached a few test files.

app.UIFigure = uifigure('Visible', 'on');
app.UIFigure.Position = [100 100 940 320];
app.UIFigure.Name = 'MATLAB App';
%fname = 'peppers.png'; % uint8, RGB
fname = 'peppers_rgba.png'; % uint8, RGBA
%fname = 'cmania.png'; % uint8, IA
%fname = 'grid_301.png'; % logical, I
%fname = 'indexedblobs.png'; % uint8, indexed
% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% direct file reading
% this handles I/IA/RGB/RGBA/indexed images and multiframe images
% handles a broader range of classes than example #2 does
% but IA/RGBA inputs can only be rendered against the figure background
% there are no other ways to control the matting
app.image1 = uiimage(app.UIFigure);
app.image1.ScaleMethod = 'none';
app.image1.Position = [10 10 300 300];
app.image1.ImageSource = fname;
% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% direct raster image data
% need to do extra babysitting because uiimage()
% can't handle raster cdata inputs sensibly
% no I/IA/RGBA/indexed/indexed+transparency or multiframe inputs
% no logical or signed integer inputs
[cdata,CT,adata] = imread(fname); % read file
if ~isempty(CT)
    cdata = ind2rgb(cdata,CT);
end
inpict = joinalpha(cdata,adata); % attach any alpha if present
inpict = alphasafe(inpict); % discard alpha by doing checkerboard matting comp
inpict = im2uint8(gray2rgb(inpict)); % make sure image is RGB and a supported class
app.image2 = uiimage(app.UIFigure);
app.image2.ScaleMethod = 'none';
app.image2.Position = [20+300 10 300 300];
app.image2.ImageSource = inpict;
% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% raster data -> tempfile -> direct file read
% This creates temp files the same way (and in the same place) that IconUtils.m does.  
% Just like IconUtils.m, unused temp files are never cleaned up, even if MATLAB exits.
% Assuming tempdir is on a ramdisk or something, it'll just be wasted memory until you reboot.
% Should be able to handle I/IA/RGB/RGBA/indexed, 
% but not indexed+transparency, and no multiframe
[cdata,CT,adata] = imread(fname);
inpict = joinalpha(cdata,adata);
tempfname = createtempfile(inpict,CT,'discardalpha'); % see options
app.image3 = uiimage(app.UIFigure);
app.image3.ScaleMethod = 'none';
app.image3.Position = [30+600 10 300 300];
app.image3.ImageSource = tempfname;
function tempfname = createtempfile(inpict,CT,alphaoption)
%  TEMPFILENAME = CREATETEMPFILE(INPICT,[],ALPHAOPTION)
%  TEMPFILENAME = CREATETEMPFILE(INPICT,CT,ALPHAOPTION)
%
%  INPICT is an I/IA/RGB/RGBA or indexed-color image of any 
%    standard image class.  Multiframe inputs are not supported.
%  CT is a Mx3 unit-scale color table associated with 
%    indexed-color image inputs. When CT is non-empty, INPICT
%    must be a single-channel integer-valued index array.
%  ALPHAOPTION specifies how IA/RGBA images are handled
%    'keepalpha' will preserve the given alpha and let uiimage()
%       handle the compositing with the figure background or 
%       whatever is beneath it in the uistack. 
%    'discardalpha' will flatten the image by composing it 
%       with a gray checkerboard background.
% create ImageComp directory under tempdir           
tmpDir = fullfile(tempdir, 'ImageComp');            
[~,~,~] = mkdir(tmpDir);
% create tempfile for image writing
tempfname = fullfile([tempname(tmpDir), '.png']);
% prepare and write temp file
if isempty(CT)
    switch lower(alphaoption)
        case 'keepalpha'
            % keep any attached alpha and let uiimage() handle the compositing
            [cdata adata] = splitalpha(inpict);
            if isempty(adata)
                imwrite(cdata,tempfname);
            else
                imwrite(cdata,tempfname,'alpha',adata);
            end
        case 'discardalpha'
            % discard alpha by doing checkerboard matting composition
            cdata = alphasafe(inpict);
            imwrite(cdata,tempfname);
    end
else
    imwrite(inpict,CT,tempfname);
end
end

The third example avoids the JPG-ification problem, and it works around a bunch of the class/depth/alpha limitations.

For my own convenience, I wrote this using some MIMT tools. I also didn't consider multiframe images and some types of indexed images.

DGM 2023-11-17

编辑：DGM 2023-11-17

Yes, the direct file reading method (the first uiimage() example) does seem to load the file nondestructively, and it preserves alpha content. It will simply compose the image over the figure background (i.e. uint8(240)), or whatever is under it in the uistack.

The problem comes if you need to display an image from the workspace (e.g. one that is being worked on). The second uiimage() example demonstrates the problem. Uiimage() just simply does not acceptably handle inputs from the workspace. Not only is it destructive, it creates an inconsistency in the image types that are supported.

The third example allows images in the workspace to be displayed using the direct file reading method. There are some limitations, given that the temp file is a PNG (e.g. can't support multiframe images), but it does allow the opportunity to control how any alpha is handled.

'keepalpha' lets uiimage() compose the transparent image as in the first example.

'discardalpha' gets rid of the transparency by composing with checkerboard matting

So the third example can do it whichever way you want, without ruining the image with JPG crust.

Thijs Boeree 2023-11-19

I have an app with a possibility to select and deselect several (27) images. I use the same images more than once, for the sake of speed i wanted to put all the images in a struct (memory) and then load and change (unload and load) within the app. Now the app has to read the image every time from the hard drive.

DGM 2023-11-21

That's basically the gist of it. Needing to rely upon direct file reads makes uiimage() significantly less practical in my opinion. I've filed a bug report, but that doesn't really fix things for current needs.

请先登录，再进行评论。