Reputation: 319
I would like to calculate the median of each pixel in a set of images or "video". However, when MATLAB starts calculating this, it takes a very long time and finishes randomly with an index error. Why?
This is the code:
V = VideoReader('hall_monitor.avi');
info = get(V);
M = info.Width;
N = info.Height;
nb_frames_bk = 5;
v_pixel = zeros([nb_frames_bk 3]);
IB=zeros([M N 3],'double');
for i=1:M
for j=1:N
for k=1:nb_frames_bk
frm=read(V,k);
v_pixel(k,:)=frm(i,j,:);
end
IB(i,j,:)=median(v_pixel(:,:));
end
end
IB=uint8(IB);
imshow(IB);
Upvotes: 1
Views: 1047
Reputation: 104515
This code can benefit from a lot of refactoring. For one thing, you are re-reading frames when you can just read them once, store them and use them after you're done.
Secondly, iterating over all pixels to compute your median is going to be very slow. From what it looks like in your code, for each spatial position over the first nb_frames_bk
frames, you collect all of the RGB values within these frames and calculate the median RGB value.
Also as a minor note, you are getting a dimension exceeds error because you defined the output matrix wrong. You defined it as M x N
with M
being the width and N
being the height. This needs to be swapped. Remember that matrices are defined as height first, width second. However, this is unnecessary with what I'm going to suggest for implementing this properly.
Instead of reading the frames one at a time, specify a range of frames. This way, you will get a 4D matrix where the first three dimensions references an image, with the fourth dimension representing the frame number. You can then take the median in the fourth dimension to find the median RGB value over all frames.
In other words, simply do this:
V = VideoReader('hall_monitor.avi');
nb_frames_bk = 5;
frms = read(V, [1 nb_frames_bk]);
IB = median(frms, 4);
imshow(IB);
This is much better, to the point and guaranteed to be faster. You also don't need to obtain the width and height of each frame as it is no longer needed as we are no longer looping over each pixel.
Upvotes: 3