Video transition detection using string matching: preliminary results

Video transition detection is an early step towards its high level segmentation and indexing. The visual rhythm is a technique used for reducing the amount of data in a video: it transforms a 3D data set (a time sequence of 2D frames) into a single 2D image. Then, we can search for patterns in the 2D image corresponding to transitions in the initial video. The concept of longest common subsequence of a string is used to transform the 2D image into a 1D signal. Further, transition detection is performed using this 1D signal whose simpler and reduced representation leads to the definition of fast algorithms identifying the different transitions.