期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2019
卷号：97
期号：19
页码：5023-5034
出版社：Journal of Theoretical and Applied
摘要：e massive volume of videos is highly demanding for produce an efficient and effective video indexing and retrieving frameworks. Extracting and representation of visual features plays a significant role in the video/image retrieval and computer vision. This paper proposes a new compact descriptor named Global Dominant Scale Invariant Feature Transform (GD-SIFT). The GD-SIFT requires fewer bits (16 bits) to represent each visual feature. Importantly, the proposed descriptor is vocabulary-free, training-free and suitable for online and real-time applications. Also, this paper proposes a new video indexing and retrieving framework based on the proposed GD-SIFT descriptor. The proposed framework is a content-based video indexing and retrieving, which helps to retrieve videos by text (e.g. Video name or metadata), image (video frame) or video clip. The experiments carried out on the standard Stanford I2V dataset. Our experiments demonstrated that, the GD-SIFT descriptor is more efficient (in terms of speed and storage) and achieved high accuracy (about 78%) with respect to the related works. Moreover, the results indicated that, the proposed descriptor is more robust to variations (e.g. Scale, rotation, etc.).
关键词：Video Indexing; Video Search; SIFT; Descriptor; Query-By-Image
