Video shot boundary detection based on multi-level features collaboration
- PDF / 1,327,150 Bytes
- 9 Pages / 595.276 x 790.866 pts Page_size
- 8 Downloads / 194 Views
ORIGINAL PAPER
Video shot boundary detection based on multi-level features collaboration Shangbo Zhou1 · Xia Wu1 · Ying Qi1 · Shuyue Luo1 · Xianzhong Xie2 Received: 7 January 2020 / Revised: 1 September 2020 / Accepted: 14 September 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Video shot boundary detection (SBD) is a basic work of content-based video retrieval and analysis. Various SBD methods have been proposed; however, there exist limitations in the complexity of boundary detection process. In this paper, a simple yet efficient SBD method is proposed, and the aim here is to speed up the boundary detection and simplify the detection process without loss of detection recall and accuracy. In our proposed model, we mainly use a top-down zoom rule, the image color feature, and local descriptors and combine a kind of motion area extraction algorithm to achieve shot boundary detection. Firstly, we select candidate transition segments via color histogram and the speeded-up robust features. Then, we perform cut transition detection through uneven slice matching, pixel difference, and color histogram. Finally, we perform gradual transition detection by the motion area extraction, scale-invariant feature transform, and even slice matching. The experiment is evaluated on the TRECVid2001 and the TRECVid2007 video datasets, and the experimental results show that our proposed method improves the recall, accuracy, and the detection speed, compared with some other related SBD methods. Keywords Color feature · SURF · Slice · SIFT · Motion area · Video shot boundary detection
1 Introduction A video usually is composed of several scenes, and a scene represents a complete plot, which is made of one or more shots. The first step of splitting a video into shots is to find the border of adjacent shots, that is video shot boundary detection. Video shot boundary is mainly of two types, i.e., cut transition (CT) and gradual transition (GT). CT is a transition where the scene has a sudden change between two adjacent frames. GT may last for several or even tens of frames, which is an artificial shot transformation effect, including fade in and fade out, dissolve, wipe, swirl, etc. For existing SBD methods, we can class them into two major categories: multiple traditional features-based methods and learning-based methods. Methods based on multiple traditional features have been proposed to extract more infor-
B
Shangbo Zhou [email protected]
1
College of Computer Science, Chongqing University, Chongqing, China
2
Key Laboratory of Computer Network and Communication Technology, Chongqing University of Posts and Telecommunications, Chongqing, China
mation of frames and realize efficient detection, such as color texture moments [1], pixel difference [2], color histogram [3,4], and feature descriptor [5–7]. Some improved image features-based methods have also been reported, such as structure information and wavelet transform [8], the scale-invariant feature transform (SIFT)-point distribution histogram [9],
Data Loading...
