Web Video Research Articles

EVENTS RECOGNITION FOR A SEMI-AUTOMATIC ANNOTATION OF SOCCER VIDEOS: A STUDY BASED DEEP LEARNING

Abstract. In this work, we propose an efficient way of web video annotation in soccer domain. To achieve this, it is necessary to enjoy different architectures of deep learning. We aim at realising a system of annotation able to recognise several events from information of the object that is the ball in our case, in order to fuse them as a part of actions in video. We propose to use Deep Neural Network (DNN) to detect ball and actions. However, Mask R-CNN can play a very important role for features extracted as an output using a training network on ImageNet dataset. The Mask R-CNN is chosen as a method using different CNN as backbone (convolutional Neural Network) ResNet50, ResNet101 and ResNet152, VGG16, VGG 19. We experimentally verify the effectiveness of the proposed method in the test phase.

Social web video clustering based on multi-modal and clustering ensemble

Web videos are rich resources for people to satisfy their information and entertainment needs. Previous studies have applied clustering methods by using textual information of videos tagged by the up-loaders to perform web video categorization, which helps users easily find videos that they really want and then increase the user’s satisfaction. However, web video categorization remains a challenging task due to the difficulties in accuracy measuring the semantic relation between terms in videos. In this paper, a novel framework for social web video clustering is proposed by improving the similarity calculation method of web videos and using the clustering ensemble. It consists of the following steps: 1) A new semantic based on Vector Space Model (VSM) is defined by considering the semantic relation of terms obtained from the lexical reference system (WordNet). 2) Word2vec is used to capture the continuous vectors as semantic information in the form of vector set of terms in a document. 3) The comprehensive extension of Semantic VSM by utilizing the Normalized Google Distance is presented. 4) The linear combining function is embodied to combine the similarity based on the optimal values of the parameter to control the weights of models before applying them to clustering paradigms and the Clustering Ensemble is employed to integrate the results of each clustering with Must-Link constraint. Experimental evaluations on real-world social web video datasets demonstrate that the proposed method effectively facilitates the clustering and achieves promising performance.

Master Gardener Satisfaction and Intent to Remain After Training via Remote Delivery

One year after using remote delivery via video web conferencing (VWC) in training, we surveyed Delaware master gardeners to assess the remote delivery trainees' satisfaction and intent to remain as compared to those characteristics in volunteers who had had face-to-face training. Although the remote delivery trainees were satisfied overall, they were significantly less satisfied than those who had participated in face-to-face trainings. A more positive perception of the VWC experience was associated with master gardeners' increased satisfaction, suggesting the importance of VWC quality. However, volunteers trained via VWC did not report more or less of an intention to remain with the organization.

ActivityNet-QA: A Dataset for Understanding Complex Web Videos via Question Answering

Recent developments in modeling language and vision have been successfully applied to image question answering. It is both crucial and natural to extend this research direction to the video domain for video question answering (VideoQA). Compared to the image domain where large scale and fully annotated benchmark datasets exists, VideoQA datasets are limited to small scale and are automatically generated, etc. These limitations restrict their applicability in practice. Here we introduce ActivityNet-QA, a fully annotated and large scale VideoQA dataset. The dataset consists of 58,000 QA pairs on 5,800 complex web videos derived from the popular ActivityNet dataset. We present a statistical analysis of our ActivityNet-QA dataset and conduct extensive experiments on it by comparing existing VideoQA baselines. Moreover, we explore various video representation strategies to improve VideoQA performance, especially for long videos.

Web video classification with visual and contextual semantics

SummaryOn the social Web, the amount of video content either originated from wireless devices or previously received from media servers has increased enormously in the recent years. The astounding growth of Web videos has stimulated researchers to propose new strategies to organize them into their respective categories. Because of complex ontology and large variation in content and quality of Web videos, it is difficult to get sufficient, precisely labeled training data, which causes hindrance in automatic video classification. In this paper, we propose a novel content‐ and context‐based Web video classification framework by rendering external support through category discriminative terms (CDTs) and semantic relatedness measure (SRM). Mainly, a three‐step framework is proposed. Firstly, content‐based video classification is proposed, where twofold use of high‐level concept detectors is leveraged to classify Web videos. Initially, category classifiers induced from VIREO‐374 detectors are trained to classify Web videos, and then concept detectors with high confidence for each video are mapped to CDT through SRM‐assisted semantic content fusion function to further boost the category classifiers, which intuitively provide a more robust measure for Web video classification. Secondly, a context‐based video classification is proposed, where twofold use of contextual information is also harnessed. Initially, cosine similarity and then semantic similarity are measured between text features of each video and CDT through vector space model (VSM)‐ and SRM‐assisted semantic context fusion function, respectively. Finally, classification results from content and context are fused to compensate for the shortcomings of each other, which enhance the video classification performance. Experiments on large‐scale video dataset validate the effectiveness of the proposed solution.

Breaking Winner-Takes-All: Iterative-Winners-Out Networks for Weakly Supervised Temporal Action Localization.

We address the challenging problem of weakly supervised temporal action localization from unconstrained web videos, where only the video-level action labels are available during training. Inspired by the adversarial erasing strategy in weakly supervised semantic segmentation, we propose a novel iterative-winners-out network. Specifically, we make two technical contributions: we propose an iterative training strategy, namely, winners-out, to select the most discriminative action instances in each training iteration and remove them in the next training iteration. This iterative process alleviates the "winner-takes-all" phenomenon that existing approaches tend to choose the video segments that strongly correspond to the video label but neglects other less discriminative video segments. With this strategy, our network is able to localize not only the most discriminative instances but also the less discriminative ones. To better select the target action instances in winners-out, we devise a class-discriminative localization technique. By employing the attention mechanism and the information learned from data, our technique is able to identify the most discriminative action instances effectively. The two key components are integrated into an end-to-end network to localize actions without using the frame-level annotations. Extensive experimental results demonstrate that our method outperforms the state-of-the-art weakly supervised approaches on ActivityNet1.3 and improves mAP from 16.9% to 20.5% on THUMOS14. Notably, even with weak video-level supervision, our method attains comparable accuracy to those employing frame-level supervisions.

ESTUDIO DE LA CALIDAD DEL VÍDEO ONLINE EN LA COMUNICACIÓN DE LA CIENCIA

Communicators and scientific journalists who work in the communication of popular science face certain peculiarities typical of this specialized area: communicating complex concepts and making them attractive to the general public. To achieve this goal, and taking into account that the internet and online video have become great allies of science divulgation, the journalist needs to know and master the peculiarities of online scientific video. The objective of this research is to determine what features make online video an effective tool to communicate science. The methodology used combines qualitative and quantitative techniques. On the one hand, it carries out a content analysis on a sample of 900 online videos on science, specifically on three current scientific topics: vaccines, nanotechnology and climate change. And, in parallel, interviews have also been conducted with a panel of nine experts: scientific journalists, producers, academics and web video experts. The conclusions suggest that online videos about science, of very various types, are usually brief and easy to understand, with a mainly informative objective. Despite the recommendations of the experts, they do not usually have elements of entertainment.

The impact of a rich media platform to table tennis learners’ performance and participation motivation

ABSTRACT Based on the HTML5 video element, WebVTT (Web Video Text Tracks) and MySQL, we developed a Rich Media Platform for Table Tennis learning (RMPTTL). This platform can support online motion analysis. Meanwhile, social agents (peers, coaches and parents) can leave comments to influence players’ participation motivation. A mixed experimental design was followed, whereby researchers conducted pre-tests and post-tests to assess the effect of using RMPTTL on players’ performance and participation motivation. Participants were thirty 7–12 year old schoolchildren from a table tennis training school in Wuhan, China. They were randomly assigned to two groups, one of which used the RMPTTL (Group R) and the other did not (Group C). The intervention lasted 8 weeks. A series of one-way analyzes of covariance (ANCOVAs) were conducted. Analysis of the data illustrated that: (1) differences in mean scores of advanced skill (alternative forehand and backhand drives) and participation motivation between the two teaching groups were significantly different; (2) there was no significant difference in mean scores of basic table tennis skills (forehand drive or backhand drive) between the two groups. The findings conclusively support the effectiveness of using RMPTTL as an adjunct to standard coaching techniques.

YouTube as a branding tool for Agencia EFE. Pillars to succeed in web video platforms

YouTube as a branding tool for Agencia EFE. Pillars to succeed in web video platforms

Cognitive psychological analysis based on multilayer semantics of web video and feature extraction of psychological images

In order to improve the accuracy of psychological analysis by extracting network video and image features, it is difficult to analyze video semantics and blur the boundaries of psychological images. In order to solve the above problems, this paper proposes a cognitive psychological analysis algorithm based on multi-layer semantics and feature extraction of psychological images in network video. Firstly, this algorithm proposes a hierarchical semantic description of network video. It can be divided into two stages: network video layering and semantic template based on layering. Then, the algorithm divides the mental image into multiple segmentation. Each segmentation focuses on the impact of different network video and different types of image features on data convergence. At the same time, interface blurring processing mechanism and image data semantic hierarchical classification method are designed. Finally, the simulation results show that the proposed algorithm is credible, feasible and effective.

Social web video clustering based on multi-view clustering via nonnegative matrix factorization

Social web videos are rich data sources containing valuable information, which have a great potential to improve the performance of social web video clustering. Social web video data usually present a characteristic of multiple views. Multi-view clustering provides a useful way to generate clusters from multi-view data. Previous studies have applied different single-view data to do social web video clustering and classification; however, multi-view data has not been a factor considered in these methods. Therefore, in this paper, we propose a framework based on a novel online multi-view clustering algorithm (called SOMVCS) to cluster social web videos with large-scale possibly incomplete views into meaningful clusters. SOMVCS learns the latent feature matrices from all the views and then drives them towards a common consensus matrix based on nonnegative matrix factorization (NMF). Particularly, we incorporate graph regularization to preserve local structure information in the model. The experimental results show that online multi-view clustering via NMF is a preferable method for social web video clustering. Moreover, we find that using multi-view data with feature types from different feature families to do social web video clustering outperforms that using data with only the feature type from a single family.

Offred Versus June

Margaret Atwood's The Handmaid's Tale has recently made its way back into popular culture andmedia. This is a consequence of the streaming service Hulu launching a web television series basedon the novel. The protagonist, Ofred, plays a crucial role in both tellings of the story, but thatrole shifts depending on the medium. Within the novel, Ofred lacks the characteristics of a hero,demonstrating complacency in her tortured position. Meanwhile, the Ofred of the Hulu series is arebel, an empowered woman who refuses oppression. These intentional portrayals of Ofred speakto the inevitable distinctions between visual and written storytelling.

Classification Performance of Violence Content by Deep Neural Network with Monarch Butterfly Optimization

Violence is self-sufficient, it is perplexing due to visibility of content dissimilarities among the positive instances that been displayed on media. Besides, the ever-increasing demand on internet, with various types of videos and genres, causes difficulty for a proper search of these videos to ensure the contents is humongous. It involves in aiding users to choose movies or web videos suitable for audience, in terms of classifying violence content. Nevertheless, this is a cumbersome job since the definition of violence is broad and subjective. Detecting such nuances from videos becomes technical without a human’s supervision that can lead to conceptual problem. Generally, violence classification is performed based on text, audio, and visual features; to be precise, it is more relevant to use of audio and visual base. However, from this perspective, deep neural network is the current build-up in machine learning approach to solve classification problems. In this research, audio and visual features are learned by the deep neural network for more specific violence content classification. This study has explored the implementation of deep neural network with monarch butterfly optimization (DNNMBO) to effectively perform the classification of the violence content in web videos. Hence, the experiments are conducted using YouTube videos from VSD2014 dataset that are publicly available by Technicolor group. The results are compared with similar modified approaches such as DNNPSO and the original DNN. The outcome has shown 94% of violence classification rate by DNNMBO.

OakStreaming: A Peer-to-PeerVideo Streaming Library

Multimedia platforms dealing with movie streaming and video-based short messages have increased the global Internet video traffic substantially in the last couple of years. Over the same period, multimedia on the Web has been standardized in terms of codecs and browserbased JavaScript APIs. However, today the technological challenges concerning the distribution of large video files are mainly tackled by scaling up capacities in cloud data centers, or relying on content delivery networks. Both approaches favor financially strong, large companies, while independent video providers with highly demanded videos are disadvantaged. Peer-to-peer streaming provides an alternative by shifting the data streams to the clients. In this article, we conceptualize different methods to move video delivery from centralized cloud infrastructures to end user devices.We discuss their strengths & weaknesses and present design considerations. To exemplify a particular approach, we showcase the implementation and evaluation of OakStreaming, our system that streams videos peer-to-peer via WebTorrent in HTML5. Particularly, we offer Web video providers a library that has various parameters, for instance to limit the bandwidth available for peer-topeer uploads. The resulting library is available as open source software on GitHub.

Enhancing theWeb WithAdvanced Engineering

Enhancing theWeb WithAdvanced Engineering

Multi-video summarization with query-dependent weighted archetypal analysis

Given the tremendous growth of web videos, video summarization is becoming increasingly important to improve user’s browsing experience. Since most existing methods focus on generating an informative summarization from a single video and often fail to generate satisfying results for multiple videos, we propose an unsupervised framework for summarizing a set of topic-related videos. We develop a Multi-Video Summarization via Multi-modal Weighted Archetypal Analysis (MVS-MWAA) method to extract a concise summarization that is both representative and informative. To ensure the summarization query-dependent, we design a multi-modal graph to guide the generation of the weight in WAA, which we call query-dependent weight. Specifically, the multi-modal graph fuses information of video frames, tags, and query-dependent web images. Furthermore, we present a Ranking from Bottom to Top (RBT) approach to make it understandable. Extensive experimental results demonstrate that our approach clearly outperforms the state-of-the-art methods.

Generalized zero-shot learning for action recognition with web-scale video data

Action recognition in surveillance video makes our life safer by detecting the criminal events or predicting violent emergencies. However, efficient action recognition is not free of difficulty. First, there are so many action classes in daily life that we cannot pre-define all possible action classes beforehand. Moreover, it is very hard to collect real-word videos for certain particular actions such as steal and street fight due to legal restrictions and privacy protection. These challenges make existing data-driven recognition methods insufficient to attain desired performance. Zero-shot learning is potential to be applied to solve these issues since it can perform classification without positive example. Nevertheless, current zero-shot learning algorithms have been studied under the unreasonable setting where seen classes are absent during the testing phase. Motivated by this, we study the task of action recognition in surveillance video under a more realistic generalized zero-shot setting, where testing data contains both seen and unseen classes. To our best knowledge, this is one of the first works to study video action recognition under the generalized zero-shot setting. We firstly perform extensive empirical studies on several existing zero-shot leaning approaches under this new setting on a web-scale video data. Our experimental results demonstrate that, under the generalize setting, typical zero-shot learning methods are no longer effective for the dataset we applied. Then, we propose to deploy generalized zero-shot learning which transfers the knowledge of Web video to detect the anomalous actions in surveillance videos. To verify the effectiveness of methods, we further construct a new surveillance video dataset consisting of nine action classes related to the public safety situation.

Reduction in a high-risk sex script among young urban women in the Love, Sex, & Choices web video HIV prevention intervention study.

For women, the high-risk sex script of condomless sex to maintain relationships with men is concerning, particularly in regard to heterosexual HIV acquisition. This secondary analysis of data from a clinical trial evaluated the effect of a 12-episode, web-based video series intervention, entitled Love, Sex, and Choices, versus a text message control on lowering the high-risk sex script. The sample comprised 238 predominately urban Black women, ages 18-29, having HIV sex risk behavior. Data were collected at baseline and 6 months post-intervention. The Sex Script Video Response score was used to measure a high-risk sex script. HIV sex risk behavior was measured by Vaginal Equivalent Episodes with high-risk partners. At baseline, lower education, younger age at first sexual intercourse, and alcohol/non-injection drug use before sex were positively associated with a high-risk sex script. At baseline, as the high-risk sex script increased by one log unit, sex risk behavior increased by 0.47 log units (p < 0.001), equivalent to a 60% increase in the geometric mean of sex risk behavior. ANCOVA analysis indicated that from baseline to 6-months post intervention, lowering the high-risk sex script by one unit was associated with an additional reduction of 0.26 units in sex risk behavior. Compared to the text messages, the LSC video series was associated with a 27% greater reduction in the geometric mean of the sex script (p = 0.03). Further investigation into the effect of lowering sex scripts to reduce sex risk behavior is recommended.

A review on robust video copy detection

The unprecedented escalation and proliferation of digital multimedia and Internet technology have triggered the enormous copyright infringement issues and tampering of digital content. Detection or localization of copy–paste forgery of digital content and distinguishing between original and manipulated video have become a weighty challenge at the present era of multimedia technology. Several distortions such as rotation, scaling and gamma correction are applied into an original video by an adversary to manipulate the original video for copyright infringement. Due to the emergence of ubiquitous digital videos on the Internet and to surpass the challenges, various copy detection schemes have been introduced by several researchers. Many real-time applications such as detection of duplicate Web videos and monitoring of real-time TV commercial media content over multi-broadcast channels require the robust copy detection approach for high security purpose. The other applications include the rapid advancement of video navigation and editing technology such as finding the opening sequence of a TV show and combining or editing similar versions of the same video for copyright infringement. This paper depicts a comprehensive overview of robust visual hashing to identify similar video contents for digital piracy detection, which overcomes the demerits of conventional cryptographic hash functions and watermarking. The paramount goal of this scheme is to generate the perceptual hash code of fixed size of length from video segments which are robust against distinct distortions or attacks such as scaling, rotation, compression, frame rate change, frame dropping, contrast enhancement, etc., made by an adversary. Besides, in this paper, distinct state-of-the-art schemes used for copy detection have been studied thoroughly and classified based on the methodology they have implemented.

Austen Approved: Pemberley Digital and the Transmedia Commodification of Jane Austen

ABSTRACTThe web video production company Pemberley Digital is most notable for their YouTube adaptation of Pride and Prejudice, the internet sensation The Lizzie Bennet Diaries. Pemberley Digital specialises in the creation of transmedia adaptations of texts, with content stretching across a variety of platforms. However, their 2013–2014 adaptation of Emma, Emma Approved, is, by Pemberley Digital's own admission, the company's boldest experiment with the commodification potential of Jane Austen's novels. Emma Approved played out over five platforms: video, blog posts, photographs, social media and music. After watching the videos, viewers are encouraged to click on links to purchase the clothes worn, or technological devices used by, the characters, incorporating product placement and advertorial into the text of Emma Approved. Pemberley Digital's unique transmedia adaptation is a distinctly twenty-first century manifestation of Jane Austen's afterlife: one that privileges a multi-dimensional, interactive “bite-sized” experience of texts, and one that seeks to transform the cultural capital associated with Austen into profit. The texts are rendered malleable to a transmedia storytelling mode across multiple digital platforms. The Pemberley Digital productions of Pride and Prejudice and Emma speak to the twenty- and twenty-first century's impulse to turn Jane Austen into a marketable, profitable commodity.