Video sequence similarity matching

A Distance Measure for Video Sequences

“…The advantage of the vstring representation is that it is fairly general-we can represent different types of video scene classifications using string sequences. For instance, we can easily classify, based on some semantic descriptions or using quantitative features from the video [1], motion vectors, angles, color, etc. More importantly, the vstring representation provides an intuitive method to model various characteristics and phenomena observed in a video sequence-such as repetitions, reverse, fast forward, and video scene breaks.…”

“…More importantly, the vstring representation provides an intuitive method to model various characteristics and phenomena observed in a video sequence-such as repetitions, reverse, fast forward, and video scene breaks. Example 1 taken from [1] shows how the basic video scene transitions such as fast forward, slow motion, reverse, and partial reverse can be modeled by the vstring representation. in Case 2 are required since the edit operation involved two different scenes.…”

“…The symbols resulting from such a fusion operation are then marked for the purpose of cost and distance computations. Thus, unlike the other edit operations which only need to be applied dynamically on only one of the strings, the fusion operation can be applied 1 Though scaling as used here is related to the notion of scaled matching introduced by Amir et al [5], our definition of scaling is different. Here, scaling is only within the symbols in a string, while in [5], scaling was defined on the entire string, i.e.…”

“…Others have tried to model the temporal information directly by treating the video data in its natural form-as an ordered sequence of frames [1,7,33]. This leads directly to video sequence similarity matching, whereby video is treated as a temporally ordered sequence, rather than as a mere collection of images.…”

“…Yazdani and Ozsoyoglu [33] defined some statistics based on the lengths of image sequences and used them for image sequence matching. Adjeroh, King, and Lee [1] described the general problems in video sequence similarity matching and proposed the vstring as an appropriate representation for the video sequence when the objective is similarity matching. In this paper, we propose a similarity measure for video sequences based on the vstring representation.…”