A Distance Measure for Video Sequences

Abstract

Video is a unique multimedia data type, in that it comes with distinguished spatio-temporal constraints. Content-based video retrieval thus requires methods for video sequence-to-sequence matching, incorporating the temporal ordering inherent in a video sequence, without losing sight of the visual nature of the information in the sequence. Such methods will require reliable measures of similarity between the video sequences. In this paper, we formulate the problem of video sequence-to-sequence matching as a patternmatching problem and propose the vstring edit distance as a suitable distance measure for video sequences.

Counting of Video Clip Repetitions using a Modified BMH Algorithm: Preliminary Results

“…While these works consider edit distance to compute the similarity between videos in which insertion, remotion and substitution are permitted, our method works very well when the frame rate is constant and no operation edition is applied. Also, the method proposed here is theorically faster than methods described in [5,6] thanks to the BMH algorithm…”

“…Our work can be directly compared to the methods developed by Adjeroh et alli [5] and Kim et alli [6]. While these works consider edit distance to compute the similarity between videos in which insertion, remotion and substitution are permitted, our method works very well when the frame rate is constant and no operation edition is applied.…”

“…of information to analyse; (iii) insertion of intensional distortions; and (iv) different frame rates. Generally, the methods that deal with this problem can be subdivided into: (i) computation of video signatures afterward the step of temporal video segmentation, like the methods described in [2,3,4]; and (ii) utilization of string matching algorithms afterward transformation of the video frame content into a feature values, like the methods described in [5,6]. When video signatures are used, we need to apply methods for temporal video segmentation [7].…”

“…When string matching algorithms are used, the developed methods take into account the efficiency of the these algorithms when compared to image/video algorithms to solve the video repetition problem. In [5,6] were used the longest common substring algorithm to deal with problem, however it requires a O(mn) space and time cost, in which m and n represent the size of the query and target video clips, respectively.…”

Real-Time Near-Duplicate Elimination for Web Video Search With Content and Context

“…For these more difficult groups, low-level features at the segment or shot level are helpful to facilitate local matching [1], [19], [22], [27], [38]. Typically the granularity of the segment-level matching, the changes in temporal order, and the insertion/deletion of frames all contribute to the similarity score of videos.…”