Steve on Image Processing

Image overlay using transparency 61

Posted by Steve Eddins,

Last August I posted an example showing how to display patch objects transparently over an image. I meant to follow that up with another post showing a couple of ways to display one image transparently over another. I was embarrassed to discover recently that I had completely forgotten to post the follow-up.

So here it is!

Handle Graphics image objects can be displayed transparently. In fact, each individual pixel can be assigned a different level of transparency. This can be used in various ways to view one image on top of another. In my first example for today, I'll use a "checkerboard" transparency pattern to view a gray-scale image on top of the original color image.

First, display the color image and the gray-scale image together, in the same place.

rgb = imread('peppers.png');
imshow(rgb);
I = rgb2gray(rgb);
hold on
h = imshow(I); % Save the handle; we'll need it later
hold off

Not too surprisingly, only the gray-scale is visible. That's because it's "covering up" the color image. Let's give it a "checkerboard" transparency pattern, so that some of the pixels are fully opaque, and others are fully transparent.

[M,N] = size(I);
block_size = 50;
P = ceil(M / block_size);
Q = ceil(N / block_size);
alpha_data = checkerboard(block_size, P, Q) > 0;
alpha_data = alpha_data(1:M, 1:N);
set(h, 'AlphaData', alpha_data);

Now we can see some of both images. This visualization technique is often used to evaluate image fusion algorithms.

Using image data to control transparency

My second transparency example gets a bit more creative. I'll display an image that's a solid color, but I'll use another data set to vary the solid color image's transparency on a pixel-by-pixel basis.

Here's a digital elevation model (DEM) of Peppercorn Hill and North Pond in Massachusetts.

E = imread('peppercorn_hill.png');
imshow(E, 'InitialMag', 'fit')

The bright blob at the upper left is Peppercorn Hill, and the flat, dark plateau in the upper middle is North Pond.

Below is an "influence map." This is a visualization of down-hill water flow, starting from the peak of Peppercorn Hill.

I = imread('peppercorn_hill_influence_map.png');
imshow(I, 'InitialMag', 'fit')

It's difficult to interpret the influence map image on its own, apart from the original DEM. Let's visualize the two images together as follows:

  1. Display the original DEM image.
  2. Display a solid green "image" on top of the original image.
  3. Use the influence map pixels to control the transparency of each pixel of the green image.
imshow(E, 'InitialMag', 'fit')

% Make a truecolor all-green image.
green = cat(3, zeros(size(E)), ones(size(E)), zeros(size(E)));
hold on
h = imshow(green);
hold off
% Use our influence map image as the AlphaData for the solid
% green image.
set(h, 'AlphaData', I)

Now it's easy to understand the water flow in the context of the original DEM image. We can see that the water flows from the peak into the pond, then out the southern end of the pond.

So there you go, better late than never. Two more image visualization techniques to add to your bag of tricks.


Get the MATLAB code

Published with MATLAB® 7.7

61 CommentsOldest to Newest

Hi
I have error message when I execute this code. It says “imshow (ParseInputs)
Unknown option string “fit”.. “.
I’m using matlab 6.5.01.
Thanks,

Zul—You are using a version of MATLAB from almost seven years ago. The ‘fit’ option wasn’t in the imshow back then.

Thanks for the great tutorial.

What would a convenient way to save the processing result? I want to save the result as an image of the same size as the original image (i.e. background), but not sure how to do it from the figure (so the image could be used for further processing).
General, saving the figures in Matlab seems poorly done (when saving a graph or an image from the figure, say as tif, the pixel size of the saved file varies depending on the size of the window, i.e. screen resolution). Any advice on how to save figures with less hassle would be greatly appreciated.

Second, is there an alternative way to do the same thing but without using imshow (i.e without invoking the figure)? Besides the above issue with saving figures, I need to process big batches of files in this manner, and was wondering if the opening/closing figure can be avoided?
Thanks so much.

Dusan—If we had designed it the other way, someone would have posted “saving figures in MATLAB seems poorly done; no matter how big the figure is on the screen, it always saves with the same resolution.” :-) I understand you, though. There are ways to control what you get. I recommend using the print function with the -dtiff and -rnum options, where num is the desired resolution in pixels per inch.

You should be able to do all of this in an invisible figure. That is, create a figure with 'Visible', 'off'. If that isn’t satisfactory, you’ll have to do a little programming yourself to implement the operation. Do an internet search for “image compositing.”

What can I do if I need to compare two image file?
one image is the image that I draw from a raw data, another image is from the pdf ( I can crop it). I hope I can compare this two image files with the second file are transparency?
Is it possible?

Nick—Your question is a little vague, so it’s hard to give a precise answer. This post shows you how to overlay one image transparently over another. MATLAB does not have a function for reading from a PDF file, so you’ll need to find another tool to get that image.

Let me ask the question in another way.
How can I put one image file transparent on top of another image file?

I’m wondering if this tip puts me on the right track for what I’m trying to do. I’m trying to display a picture of a scene with a certain object in it “blacked out”. Then, one bit (or pixel, or unit, or whatever) at a time, I want to reveal the hidden object to see how fast users can recognize it. Mr. Eddins states “each individual pixel can be assigned a different level of transparency”, so should I be adjusting the transparency of individual pixels (or groups of them) to achieve the effect I desire? I’m brand new to Matlab, so any pointers I can get would really help.

when I use the h = imshow(green);
it shown error:
??? Error using ==> imageDisplayParseInputs>validateCData at 286
Unsupported dimension

Error in ==> imageDisplayParseInputs at 207
common_args.CData = validateCData(common_args.CData,image_type);

Error in ==> imshow at 199
[common_args,specific_args] = …
What is this about?
I’m really new to MATLAB

another error is:
??? Error using ==> set
Bad value for image property: ‘AlphaData’
so what value is suitable for ‘AlphaData’ ?
thank you very much.

Hi Steve,

I am trying to use your method to overlay say correlation scores which is a matrix on a BIG image. So, first i try to copy the pixel value to cover that sub-image region that maps a correlation score to entire sub-image. Is there an easier way to do that, repmat wouldnt work since it would replicate entire matrix not pixel dilation of sorts which i need?
Anyways, main question concerning overlay is that how do i overlay the correlation score map with a colormap set on it? the usual way wouldnt work.

Thanks so much

Hi Steve,

I realized one major issue concerning this approach is that the overlay image has to be as big as the image under. Is there any way around it ? i am always runnign out of memory.

Thanks.

Tom—I think you could the image overlay transparency technique to do what you’re describing. You can change the transparency of individual pixels in order to reveal the hidden object a little bit at a time.

Tom and Rishabh—Your overlay image does not need to be as large as the image underneath. By setting the XData and YData properties of the overlay image, you can position it to display over the desired location of the image underneath.

Tom—You’ll need to learn something about Handle Graphics and the properties of graphics objects in MATLAB to program what you’re trying to do. There’s a lot of documentation on this. Look for general information about how to set and get object properties, and then look at specific information about the image object and its CData, AlphaData, XData, and YData properties.

NSJ—Make sure that your green variable is M-by-N-by-3. Look in the MATLAB documentation for information about the AlphaData property of the image object. Here’s a link to get you started.

Hi Steve,

I’m having trouble saving my final image. I’m getting the following error with the print command-

Warning: Files produced by the ‘tiff’ driver cannot be sent to printer.
File saved to disk under name ‘figure1.tif’.
> In graphics\private\name at 64
In print at 202
Warning: Problems in UIW_SetUpGLPrinting
> In C:\Program Files\toolbox\matlab\graphics\private\render.p>render at 129
In print at 263

Any help would be appreciated.

Thanks,
Ben

Benjamin—When you save to TIFF using print, you need to give it a filename, like this:

print -dtiff myfile.tif

Hi Steve,

Sorry, yes I forgot to give the image title.

I’m still getting the following error-
Warning: Problems in UIW_SetUpGLPrinting
> In graphics\private\render at 142
In print at 267

I’m using Matlab R2007b

Thank you for your time

Hi, Steve

I can not understand how do it in 2nd example. In fact, i can not repeat your result. can you list these commands one by one? sorry about that.

I am trying to do something similar to this, but I want to overlay a plot onto the image. When I try to save it with the print command, I get a new image that is a different resolution. Do you know of a way to print to a specified resolution (say 1600×1200 pixels)?
Thanks

Jay—Set the figure to have the desired size and units, set the figure’s PaperPositionMode to ‘auto’, and then use the -r switch with print.

I tried to use your second example to overlay a picture with a transparent box. But I get an unknow mistake:”??? Error using ==> imageDisplayParseInputs>validateCData at 249
Unsupported dimension

Error in ==> imageDisplayParseInputs at 202
common_args.CData = validateCData(common_args.CData,image_type);

Error in ==> imshow at 173
[common_args,specific_args] = …

Error in ==> plot at 84
h = imshow(green);” What is the Problem. I use Matlab 7.5.0

Hello Steve,

I’ve got another problem concerning the first example.
After processing the line “set(h, ‘AlphaData’, alpha_data);” my figure shows only the a few rows of the image. The Alpha_data and the image have the same dimension. This problem will also appear, if I run the script as shown in this article.
Is there an opportunity to overlay some picture, add text and save the “figure” as png or jpg without displaying it on the screen?

Kai—Use dbstop if error to get MATLAB to stop when the error occurs. Then examine the variables in your code to see what might be the problem. For example, look at the size of the variable green in the call to imshow.

Also, I encourage you to use the “Get the MATLAB code” link at the bottom of the blog post to make sure you are using the same code that I posted.

Regarding saving a figure without it appearing on screen—create an invisible figure:

fig = figure('Visible', 'off')

Then plot into the figure and use print -dpng to print the figure.

Hallo Steve,

thanks for your help. The problem was, that my images are imported as rgb colored pics. After converting with rgb2gray, the program run perfectly.

Another question. If I use print to save the current figure, the image will be saved bigger than I need. At the moment I reload the picture cut it and save it again. How can I replace this second step. Can I set a size for my figure. Or is there an option to save the overlayed axes without margin.

Thanks for help
Kai

Kai—You can control the size of the figure explicity by setting the figure properties. You can also control the size of the axes object in which the image is drawn by setting the axes properties. See the MATLAB Graphics doc for information about the ‘Units’ and ‘Position’ properties for figure and axes objects.

Hello!
i want to ask you that how can I add a small figure on a particular pixel(as its center) of another large image?

Thanks!

Amina—I don’t understand. Can you be more specific? Do you mean superimpose plotted 2-D graphics over an image?

yeah Im using a map which is in 2-D and now I want to show the client position with the help of a small icon.
anything a square or a circle.

Hey Steve,

I really appreciate your blog. I am very new to image processing. I work in the MRI field, and I am trying to overlay data using pcolor and contour. It doesn’t look like these commands work in the same way as imshow, as when I use your first technique above, nothing really happens. But when I use imshow, it works just as you suggest.

Are contour and pcolor just different from imshow in some way?

I want a grayscale image in the background with a default colors contour plot in front of the image. Can you give me a strategy for accomplishing this?

Thanks

Matt—I’m not sure exactly what effect you are trying to accomplish. Here’s an example adapted from the reference page for pcolor:

H = hadamard(20);
pcolor(H)
colormap(gray(2))
axis ij
axis square
hold on
contour(1.5:20.5, 1.5:20.5, H, [.5 .5], 'b')

Note that I had to supply the X and Y coordinates explicitly in the call to contour. That’s because pcolor and contour assume slightly different coordinate systems by default.

Also, I explicitly set the desired color for the contour lines. That’s because the contour line colors are usually drawn from the figure colormap, but the figure colormap has been set to grayscale in order to draw the grayscale background image.

If you really want the contour line colors to be set automatically from the figure colormap and for that colormap not to be gray, then you’ll need to display the background image in truecolor format, whether you use imshow or pcolor.

Hello Steve,
I am having problems with obtaining the transparency. I understand that all components must have the same size of elements and matrix dimension. However when I do

I get a N x M x 9 matrix, and I want a N x M x 3 as my background picture is, I tried to reshape it but was unsuccessful. Can you kindly shed some light on this matter.
Thank you
ZA

ZA—Sounds like your variable “E” is N-by-M-by-3 already. So when you concatenate three things that size along the third dimension, you get N-by-M-by-9.

Dear Steve,
Thank you for the tutorial, I am overlaying matrix ff onto hh according to your instruction successfully. However, I cannot export the image onto word or save as a jpg or tiff. For example, if I try to print it, it will become all black: http://docs.google.com/leaf?id=0B4swH7ucq9FuMDhiMzAzNzAtMWZmYy00N2M4LTgzMTctYWU3MGNiZjliZGNl&sort=name&layout=list&num=50

I have tried print -dtiff also. Even I changed colordef white, it’s still all black. (BTW, exportfig.m also makes all black output in my case.)

Dear Steve, the code presented here works like a dream, but for some reason if I’m using axes (of any kind) strange things will happen, e.g. the ticks stay where they should be, but the tick labels turn upside down. I’m working with Matlab 7.5.0 (2007b). I assume that something goes wrong with the handles when the images are overlaid, but can’t figure out what.

Answering to my own question, by calling

opengl software

before doing anything seems to solve the problem, god knows why.

In reference to posts 5-6, 24-25, and 29-30: I had the same problem trying to save images at a particular size, and this worked for me.

1. Set ‘PaperPositionMode’ to ‘auto’ and ‘Position’ to [doesn't matter, doesn't matter, desired output width in pixels, desired output height in pixels] in your figure.
2. Create axes within the figure with some arbitrary ‘XLim’ and ‘YLim’ properties that you specify. Specify the axes ‘Position’ property to be [0, 0, 1, 1], so the image takes up the whole figure window.
3. Display the image with the function discussed at the bottom of this post using the same values for ‘XLim’ and ‘YLim’.
4. Lookup the resolution of your monitor with:

get(0, 'ScreenPixelsPerInch')

5. Print using your monitor’s resolution with the ‘-r’ parameter.

Let me know if this doesn’t work for you.

In reference to posts 31-33:

Dear Amina,

I think you’re looking for this function:

image(xLim, yLim, yourImageAsArray)

xLim and yLim are two-element row vectors specifying the bounding rectangle into which the image should be placed. Does that help?

Hi Steve,

great post. I was wondering if there might be a way to do this using images created using the image command. Heres what my code might look like:

%%data is a bunch of images

h=image(data(:,:,1),'cdatamapping','scaled');
colormap(gray)
hold on;
h2=image(data(:,:,2),'cdatamapping','scaled');
colormap(jet)
index=find(data(:,:,2)<1);
alphadata=zeros(size(data(:,:,1)))
alphadata(index)=1;
set(h2,AlphaData,alphadata,AlphaDataMapping,'none');

Unfortunately this code above doesn’t work because I cant set two different color maps for the images in the same figure? Is there a work around besides just making a color map with ala colormap([grey(64),jet(64)]). I cant quite seem to get that to work. Any suggestions / help would be greatly appreciated.

hi steve,

a (quick) question on transparency: is it possible to display certain values in a matrix transparent, for example nans? i need to overlay two matrices; the top matrix is a diagonal matrix and i only want the diagonal to be displayed, everything else should be transparent.

thanks,
fabian

Fabian—You can’t specify a particular matrix value to be transparent, but you can use the AlphaData property of the image to achieve the same effect.

Your very nice code displays the transparent image; but how can I get this transparent image as a color image in a standalone variable name, say MyIm.

Thanks so much for your reply; what you said works very well. However, I need to get the whole overlayed image in a variable name but without displaying the figure. Because, I need to pass this variable name to some function.

Sorry Steve, I knew how to solve it. it is via set(0, ‘DefaultFigureVisible’, ‘off’). However, the real problem is that the returned image from get frame has a different size and aspect ratio from the passed image!!

set(0, 'DefaultFigureVisible', 'off')
imshow(GrayLevelImage);
Mask = cat(3, RedComponent+ zeros(size(GrayLevelImage)), GreenComponent+zeros(size(GrayLevelImage)), BlueComponent+zeros(size(GrayLevelImage)));
hold on
h = imshow(Mask);
hold off
set(h, 'AlphaData', TransperancyLevel*BWImage);
fig=get(0,'CurrentFigure');
Temp=getframe(fig);
Temp=Temp.cdata;
close(fig);
set(0, 'DefaultFigureVisible', 'on')

Waleed—You may have better luck doing the composition calculation directly instead of letting the MATLAB graphics system do it. That will give you direct control over the size. Do a search for “image composition.” I’ll add this to my list of potential blog topics.

Hi steve,
I have been reading the above post bt I am having the problem that my green matrix is N-by-M-9, for
green = cat(3, zeros(size(E)), ones(size(E)), zeros(size(E)));

How do i solve this error???
thanx in anticipation

Nishant—That’s because the image you downloaded from this web page is in RGB truecolor format rather than grayscale. Convert it to grayscale first.

hello steve
i liked your program vry much.i have seen your other programs also.i always like your work.
in the firs example u superimposed a grayscale image onto a rgb image. i want to superimpose rgb on an rgb image .if i run the same code,its not working.Could you please help me in this.

sir
actually m doing sign language recognition.m using two gloves for that.one is red and the other is blue.i made seperate programs to segment red hand and blue hand.but when i combine the codes in one program,i hold one segmented image and then show the second one,the second one superimposes on the first one.i am not able to show both both the hands
please help me out how to show both hands in one image

Hi Steve

Is there any way to do this on a surf plot?
I am trying to push to 3D the good results I get with this FEX function
http://www.mathworks.com/matlabcentral/fileexchange/14157-shaded-pseudo-color

I tried the code below but I get same result with last three lines commented off/on
Thank you, and happy holidays.

Matteo

black=cat(3,zeros(size(data)),zeros(size(data)),zeros(size(data)));
surf(data,’FaceColor’,'interp’,'EdgeColor’,'none’);
colormap(jet);colorbar;
set(gca,’YDir’,'reverse’);
view(-18,61);
material shiny;
axis tight
axis equal
axis off
hold on
kk=imagesc(black);
hold off
set(kk, ‘AlphaData’, data_slope);

Matteo—Handle Graphics image objects don’t display unless the axes is in a 2-D view. Try instead texture-mapping the data onto a flat surface object.

Not sure if to say Dear Steve because when I was reading all those posts Dear Steve, Steve, ….. the Bruce Almighty movie came to mind. =)

I have a question and I am not sure if even possible (although I think in Matlab everything is possible). I am working in augmented reality and I am trying to exactly what you explained here vey well. However, I noticed you use hold on and off and lose the handle of the first image. In augmented reality you have a “video” on real time of your environment (which would be the first image in your problem here but changing every few miliseconds) and basically I want to overlay icons (which would be your second image here in your problem) on real time and at the same time those icons should be using an AlphaData per pixel of 0.5 and they don’t change or update as the video does. They should stay there. How I am getting the video? I have a webcam and I am using a timer function to display at a given rate. To update the image (I am using image not imshow), I use the property CData. When I run the program it doesn’t show the icon I am intending to see on top of the time-variant picture/video. For the icon I created the object, got the handle, and then update inside the timer right after I update the video. I have been trying to find a way around to make it work but I was actually wondering if it was even possible because I did a search in google and nobody have tried or commented about this topic. Am I in the right track at least?

Entertaining blog. Thanks!

Hey, never mind! I just needed to try one more thing, I did, and it works like charm! Matlab is awesome! Thanks for you blog!! It was the key!

These postings are the author's and don't necessarily represent the opinions of MathWorks.