Computer Vision System Toolbox

Scene Change Detection

This example shows how to segment video in time. The example algorithm can be used to detect major changes in video streams, such as when a commercial begins and ends. It can also be useful when editing video or when you want to skip ahead through certain content.


This example illustrates the usage of the EdgeDetection System object to detect the scene change in a video stream. Segmentation of the video is done in the following steps. First, it finds the edges in two consecutive video frames, which makes the algorithm less sensitive to small changes. Based on these edges, the example compares sections (blocks) of the video frames to one another. If the number of different sections exceed a specified threshold, it is inferred that the scene has changed.


Use these next sections of code to initialize the required variables and System objects.

NumTimes = 2;   % Number of times the stream processing loop should run

Create a VideoFileReader System object to read video from a file.

hmfr = vision.VideoFileReader( ...
    'Filename', 'vipscenevideoclip.avi', ...
    'PlayCount',  NumTimes);

% Get the dimensions of each frame.
Info = info(hmfr);
rows = Info.VideoSize(2);  % Height in pixels
cols = Info.VideoSize(1);  % Width in pixels
blk_size = 32;  % Block size

% Create ROI rectangle indices for each block in image.
blk_rows = (1:blk_size:rows-blk_size+1);
blk_cols = (1:blk_size:cols-blk_size+1);
[X, Y] = meshgrid(blk_rows, blk_cols);
block_roi = [X(:)'; Y(:)'];
block_roi(3:4, :) = blk_size;
block_roi = block_roi([2 1 4 3], :)';

Create an EdgeDetector System object to find out the edges in each frame.

hedge = vision.EdgeDetector( ...
    'EdgeThinning' ,true, ...
    'ThresholdScaleFactor', 3);

Create a Mean System object to calculate the mean of every block of the edge detected image.

hmean = vision.Mean;
hmean.ROIProcessing = true;

Create VideoPlayer System objects to display the original video and the scene change detected video.

hVideo1 = vision.VideoPlayer;
hVideo1.Name  = 'Original Video';
% Video window position
hVideo1.Position(1) = round(0.4*hVideo1.Position(1));
hVideo1.Position(2) = round(1.5*(hVideo1.Position(2)));
hVideo1.Position([3 4]) = [400 200]; % video window size

hVideo2 = vision.VideoPlayer;
hVideo2.Name  = 'Sequence of start frames of the video shot.';
% Video window position
hVideo2.Position(1) = hVideo1.Position(1) + 410;
hVideo2.Position(2) = round(1.5* hVideo2.Position(2));
hVideo2.Position([3 4]) = [600 200];  % video window size

Stream Processing Loop

Create a processing loop to perform scene change detection in the input video. This loop uses the System objects you instantiated above.

% Initialize variables.
mean_blks_prev = zeros([numel(X), 1], 'single');

scene_out      = zeros([rows, 3*cols, 3], 'single');
count          = 1;
frameCount     = 0;
shotCount      = 0;

while count <= NumTimes
    I = step(hmfr);              % Read input video

    % Calculate the edge-detected image for one video component.
    I_edge = step(hedge, I(:,:,3));

    % Compute mean of every block of the edge image.
    mean_blks = step(hmean, single(I_edge), block_roi);

    % Compare the absolute difference of means between two consecutive
    % frames against a threshold to detect a scene change.
    edge_diff = abs(mean_blks - mean_blks_prev);
    edge_diff_b = edge_diff > 0.08;
    num_changed_blocks = sum(edge_diff_b(:));
    % It is a scene change if there is more than one changed block.
    scene_chg = num_changed_blocks > 0.5;

    % Display the sequence of identified scene changes along with the edges
    % information. Only the start frames of the scene changes are
    % displayed.
    I_out = cat(2, I, repmat(I_edge, [1,1,3]));

    % Display the number of frames and the number of scene changes detected
    if scene_chg
        shotCount = shotCount + 1;
    txt = sprintf('Frame %3d  Shot %d', frameCount, shotCount);
    I_out = insertText(I_out, [15 100], txt);

    % Generate sequence of scene changes detected
    if scene_chg
        % Shift old shots to left and add new video shot
        scene_out(:, 1:2*cols, :) = scene_out(:, cols+1:end, :);
        scene_out(:, 2*cols+1:end, :) = I;
        step(hVideo2, scene_out); % Display the sequence of scene changes

    step(hVideo1, I_out);         % Display the Original Video.
    mean_blks_prev = mean_blks;      % Save block mean matrix

    if isDone(hmfr)
        count = count+1;

    frameCount = frameCount + 1;


Here you call the release method on the System objects to close any open files and devices.



In the Original Video window you can see the original video and the edge detected version of it. In the window titled 'Sequence of start frames of the video shot' you can see the sequence of the start frames whenever any scene change occurs. The number of shots (scene changes) and the number of frames are displayed in the Original Video window.