Temporal video segmentation using squared form of Krawtchouk-Tchebichef moments
Rapid growth of multimedia data in cyberspace caused a swift rise in data transmission volume. This growth necessitates to look for superior techniques in processing data content. Video contains a lot of useful information; however, it consumes a vast storage space. Content based video indexin...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2018
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/71467/1/FK%202018%20110%20IR.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Rapid growth of multimedia data in cyberspace caused a swift rise in data transmission
volume. This growth necessitates to look for superior techniques in processing data
content. Video contains a lot of useful information; however, it consumes a vast
storage space. Content based video indexing and retrieval (CBVIR) aims to automate
the management, indexing and retrieval of video. Temporal Video Segmentation
(TVS) process is the essential stage in CBVIR which aims to detect transitions between
consecutive shots of videos. TVS algorithm design is still challenging because most of
the recent methods are unable to achieve robust detection for different types of
transitions: hard transition (HT) and soft transition (ST). In this regard, the aim of this
study is to propose an efficient TVS algorithm with high precision and recall values,
and low computation cost for detecting different types of transitions. In the first part of
the proposed algorithm, unique moments coefficients (features) are extracted using a
new hybrid set of orthogonal polynomials which is derived based on the modified
forms of Krawtchouk and a Tchebichef polynomials. The extracted moments showed
superior energy compaction and localization capabilities. For extracting moments, a
mathematical model of block processing that requires low computational cost is
proposed. Moreover, three different types of moments, namely smoothed moments,
and moments of gradients in x and y directions, form the unique feature vectors using
embedded image kernel model. In the proposed TVS, a modified candidate segment
selection technique is initially employed to determine the candidate segments from the
entire video. The Support Vector Machine (SVM) classifier is trained to detect
transitions. Specifically, the HTs are detected by the trained SVM model and then
refined to eliminate the false-alarm events. The fade transitions are detected based on
the smoothed moments energy and the moments of gradients correlation for the
candidate segments. In addition, the wipe and dissolve transitions are detected using
the change-point detection technique, SVM model, and scale invariant feature
transform (SIFT). For all TVS algorithm stages, the moments are computed only for
region of interest. The proposed algorithm has been evaluated on four datasets:
TRECVID 2001, TRECVID 2005, TRECVID 2006, and TRECVID 2007. The performance of the proposed algorithm is compared to that of several state-of-the-art
TVS algorithms. The improvement results of the proposed algorithm in terms of
precision, recall, and F1-score are within the ranges (0.12 – 10.06), (1.65 – 8.25), and
(0.88 – 13.85), respectively. Moreover, the proposed method shows low computation
cost which is ~5.5% of real-time. The proposed method is found to be useful to tackle
the limitations of the existing methods and serve TVS process efficiently. |
|---|