Eliminating the Back-tracking Step in the Longest Common Subsequence (lcss) Algorithm for Video Sequence Matching

Specific distance measures have been proposed in order to identify video sequences that are very similar over time but not identical (e.g. repeated takes). One such measure is based on the Longest Common Subsequence (LCSS) algorithm, a variant of the string edit distance. After building a matching matrix back-tracking is performed to identify the longest common subsequence. The modification for video sequence matching is that all sufficiently long subsequences that have gaps below a certain threshold need to be found. This increases the effort for back-tracking from linear to quadratic in the average case. In this paper we propose to eliminate the back-tracking step and integrate finding of the matching subsequences into the matrix creation with only linear effort.