Video Boundary Detection–Part 2 Gradual Transition and Its Matlab Implementation

Side Note: First draft on Apr 14 2011.

This article is a follow up of the video boundary detection for abrupt transition.  Intensity histogram measurement used in abrupt transition detection is used for gradual transition detection.

Gradual transitions are more difficult to detect than abrupt transitions. It has a lot of forms. Wipe, dissolve, fade in/fade out, just name a few of them.

The frame to frame difference of gradual transition is not as significant as it is in abrupt transitions. However, the difference between the first transition frame and the subsequent frames tend to increase, and this difference (called accumulated difference) will eventually become as comparably large as the difference seen in abrupt transition.

There is a popular method called twin-comparison method for gradual transition detection.

In this method, a lower thread hold Ts is set to detect the candidate frames that start a transition, and the same threshold used for abrupt detection Tb is used to compare against accumulative difference to test whether there really exist a transition. The end frame of the transition is detected when the consecutive difference is less than Ts, and the accumulated difference has gone beyond Tb.

An illustration of the twin-comparison is presented as the figure below,

The upper half of the figure shows how the lower threshold detects the potential start frame of the transition based on intensity histogram difference. And the lower half of the figure indicates the accumulated difference goes beyond higher threshold Tb and a gradual transition is detected.

Matlab Implementation

The matlab implementation of this method can be downloaded here. In some of the test videos, the neighboring frames doesn’t always have a difference bigger than Ts, so the implementation sets if there’re 1 out of 3 frames that has a intensity histogram difference bigger than Ts, the trend is considered as continuing.