Abstract: Described is a technology by which an image is classified (e.g., grouped and/or labeled), based on multi-label multi-instance data learning-based classification according to semantic labels and regions. An image is processed in an integrated framework into multi-label multi-instance data, including region and image labels. The framework determines local association data based on each region of an image. Other multi-label multi-instance data is based on relationships between region labels of the image, relationships between image labels of the image, and relationships between the region and image labels. These data are combined to classify the image. Training is also described.

Abstract: The concurrent multiple instance learning technique described encodes the inter-dependency between instances (e.g. regions in an image) in order to predict a label for a future instance, and, if desired the label for an image determined from the label of these instances. The technique, in one embodiment, uses a concurrent tensor to model the semantic linkage between instances in a set of images. Based on the concurrent tensor, rank-1 supersymmetric non-negative tensor factorization (SNTF) can be applied to estimate the probability of each instance being relevant to a target category. In one embodiment, the technique formulates the label prediction processes in a regularization framework, which avoids overfitting, and significantly improves a learning machine's generalization capability, similar to that in SVMs. The technique, in one embodiment, uses Reproducing Kernel Hilbert Space (RKHS) to extend predicted labels to the whole feature space based on the generalized representer theorem.

Abstract: Visual concepts contained within a video clip are classified based upon a set of target concepts. The clip is segmented into shots and a multi-layer multi-instance (MLMI) structured metadata representation of each shot is constructed. A set of pre-generated trained models of the target concepts is validated using a set of training shots. An MLMI kernel is recursively generated which models the MLMI structured metadata representation of each shot by comparing prescribed pairs of shots. The MLMI kernel is subsequently utilized to generate a learned objective decision function which learns a classifier for determining if a particular shot (that is not in the set of training shots) contains instances of the target concepts. A regularization framework can also be utilized in conjunction with the MLMI kernel to generate modified learned objective decision functions. The regularization framework introduces explicit constraints which serve to maximize the precision of the classifier.

Abstract: A method for making online adverisement makes an impressionative presentation of an advertisement to a viewer. The impressionative presentation is an impressionized version of an original online source medium such as a photo. The method associates advertisements with the source medium based, at least in part, on calculated ad relevance, and determines one or more viewer iteractive points on the original source medium. The method then presents to the viewer an ad-augmented medium including an impressionized version of the source medium, which has the ability to change the form of impression to a viewer in response to an interactive act conducted by the viewer. The ad-augmented medium may include the associated advertisement content or direct the viewer's attention thereto.

Abstract: Systems and methods for determining insertion points in a first video stream are described. The insertions points being configured for inserting at least one second video into the first video. In accordance with one embodiment, a method for determining the insertion points includes parsing the first video into a plurality of shots. The plurality of shots includes one or more shot boundaries. The method then determines one or more insertion points by balancing a discontinuity metric and an attractiveness metric of each shot boundary.

Abstract: This document describes techniques capable of associating relevant entities, such as advertisements, with insertion points within a media file. These techniques calculate a global relevancy between entities and the media file. These techniques may also calculate a local relevancy between the entities and one or more insertion points within the media file. Both global and local relevancies may employ textual and non-textual information. With use of the calculated global and local relevancies, the techniques associate one or more entities with each of the one or more insertion points in the media file. These techniques thus enable, for each insertion point, associating a most relevant entity for a particular insertion point with the insertion point. Therefore, when a user consumes the media file the user may also consume a most relevant entity at and for each insertion point in the media file.

Abstract: A method, a computer-readable storage media, and a user interface describe techniques for creating a video collage synthesized from video content, selecting representative images from the video content, extracting and resizing regions of interest (ROI) from the representative images from the video content, and arranging the regions of interest on a canvas without seams while preserving a temporal structure of the video content. The described method, computer-readable storage, and user interface enhance the experience of the user in browsing a video collage that is compact.

Abstract: Automatic video recommendation is described. The recommendation does not require an existing user profile. The source videos are directly compared to a user selected video to determine relevance, which is then used as a basis for video recommendation. The comparison is performed with respect to a weighted feature set including at least one content-based feature, such as a visual feature, an aural feature and a content-derived textural feature. Multimodal implementation including multimodal features (e.g., visual, aural and textural) extracted from the videos is used for more reliable relevance ranking. One embodiment uses an indirect textural feature generated by automatic text categorization based on a set of predefined category hierarchy. Another embodiment uses self-learning based on user click-through history to improve relevance ranking.

Abstract: Systems and methods are described for detecting capture-intention in order to analyze video content. In one implementation, a system decomposes video structure into sub-shots, extracts intention-oriented features from the sub-shots, delineates intention units via the extracted features, and classifies the intention units into intention categories via the extracted features. A video library can be organized via the categorized intention units.