MATLAB Answers

0

How to save the image WITHOUT the white background using imwrite (advanced problem)

Asked by Salad Box on 3 Sep 2019
Latest activity Edited by Salad Box on 3 Sep 2019
Hi
RGB = imread('bird1.png');
figure
imshow(RGB)
Then I get:
Then:
s = size(RGB);
rectangle('position',[1 1 s(2) s(1)], 'edgecolor', [1 0 0])
rectangle('position',[0 0 s(2) s(1)], 'edgecolor', [1 0 0])
I get:
Now I would like to save this image WITHOUT the white background, only the content within the red box (including the red box).
Imwrite needs to use 'RGB', however after the red box been added onto the image, RGB is no longer the original RGB. It WON'T be correct just to use
imwrite(RGB, 'bird1.png');
Could anyone provide correct solution to this question please?

  3 Comments

You can turn all the edge pixels in the image to red easily if that is what you want, but your question also talks about saving without the white background. An image is saved (in most formats I am aware of) as a 2d matrix of values. If the background isn't white what are you wanting to replace it with? You can't just have nothing there.
Hi Adam,
The reason that I need to remove the border/background of the image is I have to put the image on to my own background (dark-grey or black). See example below.
bird4.png
One more requirement is that I have to add box around the image (in this case it is red box). It would look extremely odd/ inappropriate to present a white border between the red box and the black background as the left one in the above image.
Above bird image is just an example image.
In reality all my images are small.
iptsetpref('ImshowBorder','tight');
would work on large images but won't work on small images as stated in the official documents.
bird5.png
Please help.
Ah well, that is different! None of the images you previously showed had any white background outside the red rectangle so I assumed you meant the white within it.

Sign in to comment.

2 Answers

Answer by Image Analyst
on 3 Sep 2019
 Accepted Answer

Try this:
% To burn the color into the image itself.
[rows, columns, numberOfCOlorChannels] = size(RGB)
RGB(:, 1, 1:3) = repmat([255, 0, 0], rows, 1);
RGB(:, end, 1:3) = repmat([255, 0, 0], rows, 1);
RGB(1, :, 1:3) = repmat([255, 0, 0], columns, 1);
RGB(end, :, 1:3) = repmat([255, 0, 0], columns, 1);
imshow(RGB);
imwrite(RGB, filename);
It's much more robust than saving the bitmap in the overlay, which is subject to change whenever you resize the figure window.

  1 Comment

I can confirm this method is a lot faster and safer.

Sign in to comment.


Answer by Johannes Fischer on 3 Sep 2019

% read image as NxMx3 rgb matrix
RGB = imread('bird1.png');
imshow(RGB)
s = size(RGB);
rectangle('position',[1 1 s(2) s(1)], 'edgecolor', [1 0 0])
rectangle('position',[0 0 s(2) s(1)], 'edgecolor', [1 0 0])
% get a handle of the axis
F = getframe(gca);
% and save the color information of the axis, which is stored as RGB in
% the field 'cdata'
imwrite(F.cdata, 'bird2.png')

  5 Comments

Salad Box, I recommend that you do not use Johannes solution. The reason is it's basically grabbing the bitmap from the displayed image. It is not the same size as the original -- I checked -- it's one pixel bigger. It's even worse (more the wrong size) should you ever resize your figure. I recommend you use my Answer instead because it will maintain the original image size. Just execute these two lines to show it:
[rows, columns, numberOfCOlorChannels] = size(RGB)
[rowsF, columnsF, numberOfCOlorChannelsF] = size(F.cdata)
rows =
305
columns =
200
numberOfCOlorChannels =
3
rowsF =
305
columnsF =
201
numberOfCOlorChannelsF =
3
Actually as I used Johannes Fischer's method on batch size (approximate 300 images), my Matlab crashed.
I restarted my PC and Matlab, I got this message:
"Warning: MATLAB previously crashed due to a low-level graphics error. To
prevent another crash in this session, MATLAB is using software OpenGL instead
of using your graphics hardware. To save this setting for future sessions, use
the opengl('save', 'software') command. For more information, see Resolving
Low-Level Graphics Issues. "
>>
I hope this has got nothing to do with your method Johannes Fischer.
Interesting... it seems to be larger by one pixel in row or column dimension, depending on the input size. I thought it would keep the size as long as no resizing is performed.
But you're right, you're solution is more robust, and in terms of batch processing probably much faster.

Sign in to comment.