Video Annotation Through Search and Graph Reinforcement Mining

...In recent decades, image annotation [19, 20, 22, 30, 44] has been attracting significant research attention in multimedia and computer vision area...
...Moxley et al. [20] first searched visually similar videos based on multiple modalities and then proposed a graph reinforcement mining approach to filter out meaningful tags for test video...

...Recently, noticing that web video sharing sites (e.g., YouTube) often have certain amount of content redundancy, video annotation by tag propagation along overlapping or near-duplicate videos are shown to be effective [22, 29, 42]...

...Such techniques have recently been evident in [15], [19], [24], and [26], which are also referred to as “annotation by search”...
...In [19], tag propagation technique is developed by crawling tags of similar videos for annotation by using text and global visual features...
...Compared to similarity-based labeling on images and videos as in [15], [19], [24], and [26], near-duplicates searched by local keypoints provide more reliable and accurate information for video annotation...
...In [19], variants of graph reinforcement algorithm are proposed for propagating tags from similar documents to query videos...
...These works [15], [19], [26] consider only global visual features for search...
...In [19] and [26], initial textual keywords or labels are further assumed to be available to guarantee efficient search and effective propagation...
...Similarly, the work in [19] aims for effective propagation of tags from similar videos but scalability is not considered...

...Moxley et al. [28] use a multiple graph reinforcement to tag the test video...
...Our algorithm is compared with other video tagging algorithms, i.e., video tagging based on Graph Reinforcement Mining (Graph-Tag) [28], neighbor-based tagging (NTag) and tag propagation based tagging (TagRank) [39]...
...We compare ICTag with other algorithms, i.e., GraphTag [28], NTag and TagRank [32]...

...Recently, such data-driven techniques have been evident by various purposes, for instance, word similarity measure [8], object recognition [22], and image/video annotation [13, 21, 24]...
...Such techniques, also referred to as “annotation by search”, have also been demonstrated in [13, 21] for video annotation...
...Similar videos are ranked in a multimodal search, and graph reinforcement mining is proposed for propagating tags from similar documents to query videos [13]...

Keywords (6)


	The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.


	BibTex \| RIS \| RefWorks Download