How to prevent colors changing when appending new frames to a video

When I read a video and write the frames back into another video the colours change slightly (red turns to orange).

How to prevent this from happening?

Here is code I have used:

Vid = VideoReader('ZV.mp4'); % Video (we want to append to)
v = VideoWriter('ZVC','MPEG-4'); % Create new video file
open(v)
% Iterate on all frames in Video and write one frame at a time
nf=0; %number-of-frames counter
while hasFrame(Vid) 
    Vidf = readFrame(Vid); % read each frame
    writeVideo(v,Vidf); % write each frame
    nf=nf+1;
end

%next we create (31) more frames

figure('Position', [0 0 600 600]);
set(gca,'visible','off')
set(gcf,'color','k')

o=-15/2+nf/ft; %ft is a fixed number
b=o+15;
[X,Y] = meshgrid(l:1/ft:r,o:1/ft:b); %l and r are fixed numbers

for i=0:30
    Y=Y+1/ft;
    o=o+1/ft;
    b=b+1/ft;
    
    Z=Ze(nf+1+i:nf+15*ft+1+i,:); %Ze is a precalculated matrix

    clf
    camroll(-90)
    
    hold on
    contourf(X,Y,Z,L,'LineWidth',1); %L is a list of values
    contour(X,Y,Z,L,'color',[0 0 0],'LineWidth',1);

    colormap(col) %col is a matrix of colours 
    clim([-15/2 25/2])

    xlim([l r])
    ylim([o b])
    hold off

    frame = getframe(gcf);
    writeVideo(v,frame)
end

close(v)

%you can't run this code unless you have ZV.mp4 and set ft,l,r,Ze,L and col
%snapshots of a detail from the video are given below

The original colour is

after 1 read-and-write it becomes

after 2 read-and-writes it becomes

after 3 read-and-writes it becomes

[Benjamin Kraus]

There seem to be two mostly unrelated sections in the code above.

1. The first section is reading one frame at a time from one file ("ZV.mp4") and writing them to another video file ("ZVC.mp4").
2. The second section is generating a contour plot and appending those frames to the same video file ("ZVC.mp4").

Is the issue you are facing with the frames from the first section or frames from the second section?

Based on your wording, and the fact that the picture doesn't look like a contour plot, I'm assuming you are talking about the first section of code, but that leaves me wondering why you included the second section of the code (the code starting with "next we create (31) more frames").

Answer 1

[Benjamin Kraus]

I'm not an exepert in MPEG-4 video compression, but my understanding is that MPEG-4 is a lossy video compression technology, so every time you write your video file you are losing a bit of information, and the video quality will continue to degrade every time you read/write the video. This is not due to MATLAB, but due to the nature of the compression used by MPEG-4 files.

If you look at the documentation for VideoWriter, you can see the name/value pair called LosslessCompression, and the description below it, which includes: "The LosslessCompression property is available only for objects used for writing Motion JPEG 2000 files."

How much information is lost can be tweaked by adjusting the name/value pairs you specify when you create your VideoWriter object. For example, it looks like the Quality property can be used to specify the quality of MPEG-4 videos. Increasing the quality won't prevent you from having degredation of the picture, but it may reduce how bad the impact is.

If your goal is that your picture is perfectly preserved across read/write cycles, then you will need to switch to a lossless compression. MATLAB's VideoWriter supports a few video profiles that include lossless compression. Looking at the documentation, it explicitly states that Archival is lossless, I'm not sure about the other.

[Peter van der Kamp]

That makes sense. Thanks, I will look into this. The first bit is indeed where the degradation occurs, but I only noticed it because of the second bit (which was also used to create the first bit).

[Peter van der Kamp]

I replaced the first two lines by:

Vid = VideoReader('ZetaVid.mj2'); % Video (we want to append to)
v = VideoWriter('ZetaVid','Archival'); % Overwrite/append to (the same) video file

This solves the problem: the colors in the frames that read from file are the same as the ones newly created, hurray! (I also found I can override the old file, which is also useful.)

However, a new issue arose. I now have numbers along one the vertical edge. Somehow the code

set(gca,'visible','off')

is not working. I also tried

axis off

which also did not work. With the mp4 format this problem did not occur.

[Benjamin Kraus]

I'm surprised that changing the video format changed whether there were numbers along one of the vertical edges. I don't see how one would impact the other in any way.

You mention overwriting the existing file, which seems to me to be more likely the source of this change. Have you tried (as an experiment) not overwriting the old file to see if the issue goes away?

Also: It may help if you explained your workflow at a higher level. For example: Why are you iteratively reading frames from one video file and writing them back to the same video file with new frames appended? Why not just write the entire video file in one go?

[Benjamin Kraus]

Also: Upon further digging into your code:

contourf(X,Y,Z,L,'LineWidth',1); %L is a list of values

contour(X,Y,Z,L,'color',[0 0 0],'LineWidth',1);

From what I can tell, this looks like you are trying to draw the same contour twice (once filled and once not filled). Because the filled contour also has black edge lines, the second call to contour looks like it is just drawing black lines along the exact same contour as the first call to contour. I can't see why the second call to contour would change the picture in any meaningful way. Why are you calling contour twice? Does the picture change at all if you remove the second call to contour?

[Peter van der Kamp]

To start with your last question, the answer is no, it doesn't change.

As to why that was done that way, I don't know. I made the picture some time ago, and am now making a video about it.

Removing the second contour plot speeds up the process quite a bit, roughly 30% faster, so thanks heaps for your dig!

Now the other thing. I have found that there is actually no difference between mp4 and mj2 in that respect, as expected. What happened is that I moved axis off out of the (frame-creating) loop in which I clear the figure (clf), and clf resets the axis back on. The other thing that I moved out is set(gcf,'color','k'), but this is not undone by clf. The reason that I had clf in the loop is that I have seen remnants of lines at edges of videos before. I have now tested the difference (with or without clf in the loop): there is no visual difference, however, having it in the loop is much much faster than not having it in the loop.

Finally, about the workflow at a higher level, this video is (going to be) a long term project. I am visualising zeros of the Riemann zeta function in a kind of arty way (see below). It takes a lot of time to calculate the data, and also a lot of time to create the frames/video. As there are infinitely many zeros, the video is going to be infinitely long (bounded by memory I guess). Although I enjoy watching the video while it is being created, this quite slow-motion, and I want to show finite parts of the infinite video to other people at proper speed.