Modified phase-correlation based robust hard-cut detection with application to archive film

Urhan, Oğuzhan; Güllü, M. Kemal; Ertürk, Şükrü Mehmet

doi:10.1109/tcsvt.2006.875210

Cited by 33 publications

(34 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Detection of shot boundary for video segmentation was originally introduced decades ago to detect abrupt cuts in videos [3,9,[12][13]. Since then, many techniques have been developed in both the compressed and the uncompressed domains.…”

Section: Related Workmentioning

confidence: 99%

“…In uncompressed-domain, frame difference is usually measured using pixel difference, histogram similarity, texture/edge and inter-frame correlation, etc., followed by decision making via thresholding, statistical analysis and machine learning [3,5,[8][9][10][13][14]. In Grana and Cucchiara [3], a linear transition model is developed to identify the shot transition centre and length.…”

Section: Related Workmentioning

confidence: 99%

“…A discriminative feature selection process is performed, offline, based on mutual information for the KNN (K-Nearest Neighbours) classification of video shots. In Urhan et al [9], a hard-cut detection system is presented based on modified phase correlation with application to archived films. Video frames are spatially sub-sampled for phase correlation and the generated peaks are detected by double thresholding, i.e.…”

Section: Related Workmentioning

confidence: 99%

“…To detect shot cuts, most of the existing work reported in the literature measures the content difference between the current frame and its preceding one, and then apply a threshold to decide whether the difference measured is large enough to justify a cut detection [6,9,14]. Many algorithms have been developed and reported on how to decide such a threshold, which can be briefly summarized as: (i) statistics based approaches such as Bayesian rules, maximum likelihood based probabilistic modelling etc.…”

Section: Introductionmentioning

confidence: 99%

“…Many algorithms have been developed and reported on how to decide such a threshold, which can be briefly summarized as: (i) statistics based approaches such as Bayesian rules, maximum likelihood based probabilistic modelling etc. [3,10]; (ii) empirical studies [9,14]; and (iii) machine learning approaches such as SVM, neural networks etc to bypass the threshold and make decisions based on the training and learning process [5,7,13]. The fundamental issue here, however, is that such a measured neighbourhood frame difference indicator alone would not provide sufficient information for shot cut detection.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Modelling of content-aware indicators for effective determination of shot boundaries in compressed MPEG videos

Chen

Ren

Jiang

2010

Multimed Tools Appl

View full text Add to dashboard Cite

Abstract. In this paper, a content-aware approach is proposed to design multiple test conditions for shot cut detection, which are organized into a multiple phase decision tree for abrupt cut detection and a finite state machine for dissolve detection. In comparison with existing approaches, our algorithm is characterized with two categories of content difference indicators and testing. While the first category indicates the content changes that are directly used for shot cut detection, the second category indicates the contexts under which the content change occurs. As a result,indications of frame differences are tested with context awareness to make the detection of shot cuts adaptive to both content and context changes. Evaluations announced by TRECVID 2007 indicate that our proposed algorithm achieved comparable performance to those using machine learning approaches, yet using a simpler feature set and straightforward design strategies. This has validated the effectiveness of modelling of content-aware indicators for decision making, which also provides a good alternative to conventional approaches in this topic.

show abstract

Section: Related Workmentioning

confidence: 99%