Lyndon Kennedy

Large-scale concept ontology for multimedia

As increasingly powerful techniques emerge for machine tagging multimedia content, it becomes ever more important to standardize the underlying vocabularies. Doing so provides interoperability and lets the multimedia community focus ongoing research on a well-defined set of semantics. This paper describes a collaborative effort of multimedia researchers, library scientists, and end users to develop a large standardized taxonomy for describing broadcast news video. The large-scale concept ontology for multimedia (LSCOM) is the first of its kind designed to simultaneously optimize utility to facilitate end-user access, cover a large semantic space, make automated extraction feasible, and increase observability in diverse broadcast news video data sets

Tweet the debates: understanding community annotation of uncollected sources

We investigate the practice of sharing short messages (microblogging) around live media events. Our focus is on Twitter and its usage during the 2008 Presidential Debates. We find that analysis of Twitter usage patterns around this media event can yield significant insights into the semantic structure and content of the media object. Specifically, we find that the level of Twitter activity serves as a predictor of changes in topics in the media event. Further we find that conversational cues can identify the key players in the media object and that the content of the Twitter posts can somewhat reflect the topics of discussion in the media object, but are mostly evaluative, in that they express the posters reaction to the media. The key contribution of this work is an analysis of the practice of microblogging live events and the core metrics that can leveraged to evaluate and analyze this activity. Finally, we offer suggestions on how our model of segmentation and node identification could apply towards any live, real-time arbitrary event.

Video search reranking through random walk over document-level context graph

Multimedia search over distributed sources often result in recurrent images or videos which are manifested beyond the textual modality. To exploit such contextual patterns and keep the simplicity of the keyword-based search, we propose novel reranking methods to leverage the recurrent patterns to improve the initial text search results. The approach, context reranking, is formulated as a random walk problem along the context graph, where video stories are nodes and the edges between them are weighted by multimodal contextual similarities. The random walk is biased with the preference towards stories with higher initial text search scores - a principled way to consider both initial text search results and their implicit contextual relationships. When evaluated on TRECVID 2005 video benchmark, the proposed approach can improve retrieval on the average up to 32% relative to the baseline text search method in terms of story-level Mean Average Precision. In the people-related queries, which usually have recurrent coverage across news sources, we can have up to 40% relative improvement. Most of all, the proposed method does not require any additional input from users (e.g., example images), or complex search models for special queries (e.g., named person search).

Video search reranking via information bottleneck principle

We propose a novel and generic video/image reranking algorithm, IB reranking, which reorders results from text-only searches by discovering the salient visual patterns of relevant and irrelevant shots from the approximate relevance provided by text results. The IB reranking method, based on a rigorous Information Bottleneck (IB) principle, finds the optimal clustering of images that preserves the maximal mutual information between the search relevance and the high-dimensional low-level visual features of the images in the text search results. Evaluating the approach on the TRECVID 2003-2005 data sets shows significant improvement upon the text search baseline, with relative increases in average performance of up to 23%. The method requires no image search examples from the user, but is competitive with other state-of-the-art example-based approaches. The method is also highly generic and performs comparably with sophisticated models which are highly tuned for specific classes of queries, such as named-persons. Our experimental analysis has also confirmed the proposed reranking method works well when there exist sufficient recurrent visual patterns in the search results, as often the case in multi-source news videos.

To search or to label?: predicting the performance of search-based automatic image classifiers

In this work we explore the trade-offs in acquiring training data for image classification models through automated web search as opposed to human annotation. Automated web search comes at no cost in human labor, but sometimes leads to decreased classification performance, while human annotations come at great expense in human labor but result in better performance. The primary contribution of this work is a system for predicting which visual concepts will show the greatest increase in performance from investing human effort in obtaining annotations. We propose to build this system as an estimation of the absolute gain in average precision (AP) experienced from using human annotations instead of web search. To estimate the AP gain, we rely on statistical classifiers built on top of a number of quality prediction features. We employ a feature selection algorithm to compare the quality of each of the predictors and find that cross-domain image similarity and cross-domain model generalization metrics are strong predictors, while concept frequency and within-domain model quality are weak predictors. In a test application, we find that the prediction scheme can result in a savings in annotation effort of up to 75\%, while only incurring marginal damage (10% relative decrease in mean average precision) to the overall performance of the concept models.

Laughter Detection in Meetings

We build a system to automatically detect laughter events in meetings, where laughter events are defined as points in the meeting where a number of the participants (more than just one) are laughing simultaneously. We implement our system using a support vector machine classifier trained on mel-frequency cepstral coefficients (MFCCs), delta MFCCs, modulation spectrum, and spatial cues from the time delay between two desktop microphones. We run our experiments on the ‘Bmr’ subset of the ICSI Meeting Recorder corpus using just two table-top microphones and obtain detection results with a correct accept rate of 87% and a false alarm rate of 13%.

Less talk, more rock: automated organization of community-contributed collections of concert videos

We describe a system for synchronization and organization of user-contributed content from live music events. We start with a set of short video clips taken at a single event by multiple contributors, who were using a varied set of capture devices. Using audio fingerprints, we synchronize these clips such that overlapping clips can be displayed simultaneously. Furthermore, we use the timing and link structure generated by the synchronization algorithm to improve the findability and representation of the event content, including identifying key moments of interest and descriptive text for important captured segments of the show. We also identify the preferred audio track when multiple clips overlap. We thus create a much improved representation of the event that builds on the automatic content match. Our work demonstrates important principles in the use of content analysis techniques for social media content on the Web, and applies those principles in the domain of live music capture.

Kodak's consumer video benchmark data set: concept definition and annotation

Semantic indexing of images and videos in the consumer domain has become a very important issue for both research and actual application. In this work we developed Kodaks consumer video benchmark data set, which includes (1) a significant number of videos from actual users, (2) a rich lexicon that accommodates consumers. needs, and (3) the annotation of a subset of concepts over the entire video data set. To the best of our knowledge, this is the first systematic work in the consumer domain aimed at the definition of a large lexicon, construction of a large benchmark data set, and annotation of videos in a rigorous fashion. Such effort will have significant impact by providing a sound foundation for developing and evaluating large-scale learning-based semantic indexing/annotation techniques in the consumer domain.

Getting the look: clothing recognition and segmentation for automatic product suggestions in everyday photos

We present a scalable approach to automatically suggest relevant clothing products, given a single image without metadata. We formulate the problem as cross-scenario retrieval: the query is a real-world image, while the products from online shopping catalogs are usually presented in a clean environment. We divide our approach into two main stages: a) Starting from articulated pose estimation, we segment the person area and cluster promising image regions in order to detect the clothing classes present in the query image. b) We use image retrieval techniques to retrieve visually similar products from each of the detected classes. We achieve clothing detection performance comparable to the state-of-the-art on a very recent annotated dataset, while being more than 50 times faster. Finally, we present a large scale clothing suggestion scenario, where the product database contains over one million products.

A reranking approach for context-based concept fusion in video indexing and retrieval

We propose to incorporate hundreds of pre-trained concept detectors to provide contextual information for improving the performance of multimodal video search. The approach takes initial search results from established video search methods (which typically are conservative in usage of concept detectors) and mines these results to discover and leverage co-occurrence patterns with detection results for hundreds of other concepts, thereby refining and reranking the initial video search result. We test the method on TRECVID 2005 and 2006 automatic video search tasks and find improvements in mean average precision (MAP) of 15%-30%. We also find that the method is adept at discovering contextual relationships that are unique to news stories occurring in the search set, which would be difficult or impossible to discover even if external training data were available.