L. Rueda, D. Mery, and J. Kittler (Eds.): CIARP 2007, LNCS 4756, pp. 311–320, 2007.
© Springer-Verlag Berlin Heidelberg 2007
An Evaluation of Video Cut Detection Techniques
Sandberg Marcel Santos
1
, Díbio Leandro Borges
2
, and Herman Martins Gomes
1
1
Departamento de Sistemas e Computação, Universidade Federal de Campina Grande, Av.
Aprígio Veloso s/n, 58109-970 Campina Grande PB, Brazil
{sandberg,hmg}@dsc.ufcg.edu.br
2
Departamento de Ciência da Computação, Fundação Universidade de Brasília, Campus
Universitário Darcy Ribeiro, 70910-900 Brasília DF, Brazil
dibio@unb.br
Abstract. Accurate detection of shot transitions plays an important role on
automatic analysis of digital video contents, and it is a key issue for video
indexing and summarization, amongst other tasks. This work presents in more
detail a novel strategy, based on the concept of visual rhythm, to automatically
detect sharp transitions or cuts in arbitrary videos. The central part of the work
is a comparative evaluation of this strategy versus three other very competitive
approaches for video cut detection: one based on the visual rhythm concept,
other based on pixel differentiation and a last one based on color histograms.
The evaluation carried out demonstrated that the proposed method achieves, on
average, higher recall rates at a cost of a slightly lower precision.
Keywords: video cut detection, visual rhythm, pixel differentiation, color
histograms, video summarization.
1 Introduction
Digital video applications, such as digital libraries, interactive TV, and multimedia
information systems in general, are growing fast due to the advances in multimedia
encoding and decoding technologies, increase in computing power and the ever-
expanding internet [1]. This has stimulated research in the areas video indexing,
retrieval and summarization. While digital videos can be seen as formed by a
concatenation of 2-D image samples (frames) of a scene, shots can be seen as a basic
functional unit of a video. Shots are defined as uninterrupted sequences of video
frames with graphic, spatial and temporal configurations [3]. The automatic detection
of shots or the transition between two consecutive shots is an essential part of most
video content analysis algorithms.
Gradual and sharp transitions are the two most known types of video transitions
[4]. In this paper, we focus on the problem of detecting sharp transitions (or cuts),
which is usually taken as a simpler problem than that of gradual transition detection.
However the state of the art, as indicated in our literature review, reveals there is still
room for improvements in the accuracy of cut detection techniques. In a previous
work [13], we proposed an algorithm for video cut detection based on the concept of
visual rhythm and compared this algorithm with a previous approach, based on the
same principle (the work by Lu et al. [10]). The visual rhythm concept [4] is