Abstract: A segmentation annotation technique for media items is disclosed herein. Given a weakly labeled media item, spatiotemporal masks may be generated for each of the concepts with which it is labeled. Segments may be ranked by the likelihood that they correspond to a given concept. The ranked concept segments may be utilized to train a classifier that, in turn, may be used to classify untagged or new media items.

Abstract: A motion manifold system analyzes a set of videos, identifying image patches within those videos corresponding to regions of interest and identifying patch trajectories by tracking the movement of the regions over time in the videos. Based on the patch identification and tracking, the system produces a motion manifold data structure that captures the way in which the same semantic region can have different visual representations over time. The motion manifold can then be applied to determine the semantic similarity between different patches, or between higher-level constructs such as images or video segments, including detecting semantic similarity between patches or other constructs that are visually dissimilar.

Abstract: The embodiments of the invention relate to method for image-based retrieval of an image or text, comprising capturing a digital image; matching the digital image against digitally-stored documents; and retrieving one or more of the digitally-stored documents that match the digital image. Other embodiments relate to a device for image-based retrieval and rendering, comprising an image capturing device for capturing digital image; a device for processing the digital image to match it against digitally-stored documents; a device for retrieving a digitally-stored document that matches the digital image; and a device for rendering the selected retrieved image.

Abstract: This disclosure generally relates to systems and methods that facilitate employing exemplar Histogram of Oriented Gradients Linear Discriminant Analysis (HOG-LDA) models along with Localizer Hidden Markov Models (HMM) to train a classification model to classify actions in videos by learning poses and transitions between the poses associated with the actions in a view of a continuous state represented by bounding boxes corresponding to where the action is located in frames of the video.

Abstract: A volume identification system identifies a set of unlabeled spatio-temporal volumes within each of a set of videos, each volume representing a distinct object or action. The volume identification system further determines, for each of the videos, a set of volume-level features characterizing the volume as a whole. In one embodiment, the features are based on a codebook and describe the temporal and spatial relationships of different codebook entries of the volume. The volume identification system uses the volume-level features, in conjunction with existing labels assigned to the videos as a whole, to label with high confidence some subset of the identified volumes, e.g., by employing consistency learning or training and application of weak volume classifiers. The labeled volumes may be used for a number of applications, such as training strong volume classifiers, improving video search (including locating individual volumes), and creating composite videos based on identified volumes.

Abstract: A volume identification system identifies a set of unlabeled spatio-temporal volumes within each of a set of videos, each volume representing a distinct object or action. The volume identification system further determines, for each of the videos, a set of volume-level features characterizing the volume as a whole. In one embodiment, the features are based on a codebook and describe the temporal and spatial relationships of different codebook entries of the volume. The volume identification system uses the volume-level features, in conjunction with existing labels assigned to the videos as a whole, to label with high confidence some subset of the identified volumes, e.g., by employing consistency learning or training and application of weak volume classifiers. The labeled volumes may be used for a number of applications, such as training strong volume classifiers, improving video search (including locating individual volumes), and creating composite videos based on identified volumes.

Abstract: A volume identification system identifies a set of unlabeled spatio-temporal volumes within each of a set of videos, each volume representing a distinct object or action. The volume identification system further determines, for each of the videos, a set of volume-level features characterizing the volume as a whole. In one embodiment, the features are based on a codebook and describe the temporal and spatial relationships of different codebook entries of the volume. The volume identification system uses the volume-level features, in conjunction with existing labels assigned to the videos as a whole, to label with high confidence some subset of the identified volumes, e.g., by employing consistency learning or training and application of weak volume classifiers. The labeled volumes may be used for a number of applications, such as training strong volume classifiers, improving video search (including locating individual volumes), and creating composite videos based on identified volumes.

Abstract: A volume identification system identifies a set of unlabeled spatio-temporal volumes within each of a set of videos, each volume representing a distinct object or action. The volume identification system further determines, for each of the videos, a set of volume-level features characterizing the volume as a whole. In one embodiment, the features are based on a codebook and describe the temporal and spatial relationships of different codebook entries of the volume. The volume identification system uses the volume-level features, in conjunction with existing labels assigned to the videos as a whole, to label with high confidence some subset of the identified volumes, e.g., by employing consistency learning or training and application of weak volume classifiers. The labeled volumes may be used for a number of applications, such as training strong volume classifiers, improving video search (including locating individual volumes), and creating composite videos based on identified volumes.

Abstract: A volume identification system identifies a set of unlabeled spatio-temporal volumes within each of a set of videos, each volume representing a distinct object or action. The volume identification system further determines, for each of the videos, a set of volume-level features characterizing the volume as a whole. In one embodiment, the features are based on a codebook and describe the temporal and spatial relationships of different codebook entries of the volume. The volume identification system uses the volume-level features, in conjunction with existing labels assigned to the videos as a whole, to label with high confidence some subset of the identified volumes, e.g., by employing consistency learning or training and application of weak volume classifiers. The labeled volumes may be used for a number of applications, such as training strong volume classifiers, improving video search (including locating individual volumes), and creating composite videos based on identified volumes.

Abstract: A volume identification system identifies a set of unlabeled spatio-temporal volumes within each of a set of videos, each volume representing a distinct object or action. The volume identification system further determines, for each of the videos, a set of volume-level features characterizing the volume as a whole. In one embodiment, the features are based on a codebook and describe the temporal and spatial relationships of different codebook entries of the volume. The volume identification system uses the volume-level features, in conjunction with existing labels assigned to the videos as a whole, to label with high confidence some subset of the identified volumes, e.g., by employing consistency learning or training and application of weak volume classifiers. The labeled volumes may be used for a number of applications, such as training strong volume classifiers, improving video search (including locating individual volumes), and creating composite videos based on identified volumes.

Abstract: A motion manifold system analyzes a set of videos, identifying image patches within those videos corresponding to regions of interest and identifying patch trajectories by tracking the movement of the regions over time in the videos. Based on the patch identification and tracking, the system produces a motion manifold data structure that captures the way in which the same semantic region can have different visual representations over time. The motion manifold can then be applied to determine the semantic similarity between different patches, or between higher-level constructs such as images or video segments, including detecting semantic similarity between patches or other constructs that are visually dissimilar.

Abstract: The embodiments of the invention relate to method for image-based retrieval of an image or text, comprising capturing a digital image; matching the digital image against digitally-stored documents; and retrieving one or more of the digitally-stored documents that match the digital image. Other embodiments relate to a device for image-based retrieval and rendering, comprising an image capturing device for capturing digital image; a device for processing the digital image to match it against digitally-stored documents; a device for retrieving a digitally-stored document that matches the digital image; and a device for rendering the selected retrieved image.

Abstract: The embodiments of the invention relate to method for image-based retrieval of an image or text, comprising capturing a digital image; matching the digital image against digitally-stored documents; and retrieving one or more of the digitally-stored documents that match the digital image. Other embodiments relate to a device for image-based retrieval and rendering, comprising an image capturing device for capturing digital image; a device for processing the digital image to match it against digitally-stored documents; a device for retrieving a digitally-stored document that matches the digital image; and a device for rendering the selected retrieved image.

Abstract: A system provides a means to detect and track a patient silhouette which may be used to instruct him/her in positioning a medical sensing device on his/her chest with guidance from a computer or from a remotely located physician. The medical sensing device may be, for example a stethoscope or other device.

Abstract: The embodiments of the invention relate to method for image-based retrieval of an image or text, comprising capturing a digital image; matching the digital image against digitally-stored documents; and retrieving one or more of the digitally-stored documents that match the digital image. Other embodiments relate to a device for image-based retrieval and rendering, comprising an image capturing device for capturing digital image; a device for processing the digital image to match it against digitally-stored documents; a device for retrieving a digitally-stored document that matches the digital image; and a device for rendering the selected retrieved image.

Abstract: Massive amounts of multimedia data are stored in databases supporting web pages and servers, including text, graphics, video and audio. Searching and finding matching multimedia images can be time and computationally intensive. A method for storing and retrieving image data includes computing a descriptor, such an a Fourier-Mellin Transform (FMT), corresponding to a multidimensional space indicative of each of the stored images and organizing each of the descriptors according to a set similarity metric. The set similarity metric is based on Locality-Sensitive Hashing (LSH), and orders descriptors near to other descriptors in the database. The set similarity metric employs set theory which allows distance between descriptors to be computed consistent with LSH. A target image for which a match is sought is then received, and a descriptor indicative of the target image is computed. The database is referenced, or mapped, to determine close matches in the database.

Abstract: A system and method corrects luminance non-uniformity caused by images being obliquely projected onto a screen. A camera is used to record the geometry of the obliquely displayed image. Utilizing this recorded geometry, a homography is then derived that maps pixels between the projector's coordinate system and the screen's coordinate system. Utilizing the homography, the projector pixel that attends to the largest projected area on the screen is identified. Next, the ratio of each pixel's projected area to the largest projected area is computed. These ratios are then organized into an attenuation array that is used to produce “corrected” luminance information from input image data. The projector is then driven with the “corrected” luminance information.

Abstract: Media slides are often employed in conference sessions, meetings, lectures, and other interactive forums. The proliferation of laptops and handheld computers allows a speaker to present directly from the laptop by connecting to the projector at the conference site. Physically connecting and disconnecting each presenter's laptop to the projection apparatus, however, can be a clumsy and disruptive process, particularly since the presenters may be seated at various locations around the room. A wireless interface between a presentation server and a laptop in a multi-user multi-projector presentation system allows a media sequence from each media source to be displayed on a common display via the presentation server and the wireless interface. Presenters need not run or swap cables or other physical connections to switch media sources to the common display. The interface requires no software modification to the media source laptops and maintains independence between media sources and the server for security.

Abstract: A system and method corrects luminance non-uniformity caused by images being obliquely projected onto a screen. A camera is used to record the geometry of the obliquely displayed image. Utilizing this recorded geometry, a homography is then derived that maps pixels between the projector's coordinate system and the screen's coordinate system. Utilizing the homography, the projector pixel that attends to the largest projected area on the screen is identified. Next, the ratio of each pixel's projected area to the largest projected area is computed. These ratios are then organized into an attenuation array that is used to produce “corrected” luminance information from input image data. The projector is then driven with the “corrected” luminance information.

Abstract: The present invention provides automatic correction of any distortions produced when computer projection displays are misaligned with respect to the projection surface (such as keystoning). Although sophisticated LCD projectors now offer partial solutions to this problem, they require specialized hardware and time-consuming manual adjustment. The two key concepts in the present invention are: (1) using an uncalibrated camera to observe the projected image; and (2) the image to be displayed is pre-warped so that the distortions induced by the misaligned projection system will exactly undo the distortion. The result is that an arbitrarily mounted projector (in an unknown orientation) still displays a perfectly aligned and rectilinear image.