How to read video frames the fastest.

I want to read in a video file into matlab, find the average color of each frame and store that. I want to do to it as fast as possible.

v = VideoReader('File.mkv');
numFrames = floor(v.Duration*v.FrameRate); %Get number of frames
data = zeros(numFrames,3); %initialize data structure
parfor_progress(numFrames); %Just for tracking parfor progress
parfor cnt = 1:numFrames
    frame = read(v, cnt); %Reading a frame
    for i = 1:3
        data(cnt,i) = mean2(frame(:,:,i)); %Average of each RGB value for each frame
    end
    parfor_progress; %Update parfor progress
end
parfor_progress(0) %end Parfor progress bar
toc

The limit seems to be the "read' function, I have tried using the newer 'readFrame' but I can't use parfor with it so it is ultimatly slower.

[OCDER]

Just curious, how much slower/faster is the code when you use a normal for loop, and without using that parfor_progress?

[Gail Distefano]

Hi, I assume you have been successful reading the .mkv file with VideoReader. I get this error. Do you have any suggestion?

v = VideoReader('C:\users\gaila\Documents\output.mkv')

Error using VideoReader/initReader (line 734)

Unexpected exception in plug-in: 'No Frame Rate for this file Reason: The requested attribute was not found.'

Error in audiovideo.internal.IVideoReader (line 136)

initReader(obj, fileName, currentTime);

Error in VideoReader (line 104)

[e-mail](varargin{:});

Answer 1

[Jan]

Replace:

   for i = 1:3
     data(cnt,i) = mean2(frame(:,:,i)); %Average of each RGB value for each frame
   end

by

N = v.Height * v.Width;  % Before the loop
...

    data(cnt, :) = sum(reshape(frame, [], 3), 1) / N;

The progress might take longer than the actual processing. Do you really need it? Then reduce the number of calls, perhaps by:

if rem(cnt, 100)
  parfor_progress;
end

Answer 2

[Álvaro Sacristán Gonzalez]

Instead of manually calculating the grayscale, Id suggest using the image processing toolbox function:

frame = mat2gray(frame)

[Walter Roberson]

rescale() is recommended now

[Álvaro Sacristán Gonzalez]

What do you mean?

[Walter Roberson]

MATLAB introduced rescale() to replace mat2gray. It has a better name And is part of MATLAB itself not a toolbox