A Method of Video Shot-Boundary Detection based on Grey Modeling for Histogram Sequence

Abstract

Video Shot-Boundary Detection is important for video analysis, segmentation and retrieval. The detection results of Shot-Boundary are not only relation with two nearest frames of shot cuts, but also their several neighboring frames. Video stream is multistream time sequence. Grey model based on set sequence (SGM) is proposed to analyze and process multi-stream time sequences constructed with sampled histograms in this paper. Consequently, a novel Shot-Boundary Detection method is put forward based on SGM and color histogram, which includes the following steps: (1) When a new video frame coming into the detection system, the image is transformed into intensity image under HSI color model; (2) Histogram of the video frame can be obtained. Then, taking samples for the histogram and constructing set sequence with proximate sampled histograms is implemented;(3) SGM is introduced to simulate the sampled histogram sequence in this detection method. Moreover, absolute mean error (AME) and regulative AME (R AME ) with thresholds are applied to make a detection judgment. Finally, the experiments illuminate that the proposed method is an effective Shot-Boundary Detection method, and it is superior to method of Histogram disparity significantly, especially when it is used to detect cuts with some complex video scenes.

Video Shot-Boundary Detection based on Matrix sequence Grey model

“…The comparison methods are Video Shot-Boundary Detection by MGM in this paper, Video Shot-Boundary Detection by SGM in paper [12], and histogram disparity (Dh) between two of nearest frames I k and I k+1 , shown in formula (5) in which threshold value Th is used to judge the cuts. The computing time of three methods are compared on equivalent calculation condition.…”

“…The following metrics is used to evaluate the performance of the proposed cut detection methods [12,14,15]: (1) …”

“…Grey modeling for matrix sequence [11] and point set sequence [12] can used to simulate or predict multi-stream information. In this paper, matrix sequence is introduced here to illustrate the space relationship of video frames.…”

“…Once a new video frame comes into the detection system, the system will do grey modeling with MGM to compute AME (the absolute mean error) values, and the matrix sequence in the modeling process is constructed by consecutive 5 video frames including the current frame and the previous four ones. Then, another criteria, R AME (regulative AME), will be computed at the same time as formula (4) [12]. Thresholds T and T1 are used to make a judgment whether a frame is the beginning of another shot.…”

“…The steps of proposed Shot-Boundary detection method are shown as Fig. 3 similar to paper [12]. In Fig.…”