摘要:
Methods for interactive selecting video queries consisting of training images from a video for a video similarity search and for displaying the results of the similarity search are disclosed. The user selects a time interval in the video as a query definition of training images for training an image class statistical model. Time intervals can be as short as one frame or consist of disjoint segments or shots. A statistical model of the image class defined by the training images is calculated on-the-fly from feature vectors extracted from transforms of the training images. For each frame in the video, a feature vector is extracted from the transform of the frame, and a similarity measure is calculated using the feature vector and the image class statistical model. The similarity measure is derived from the likelihood of a Gaussian model producing the frame. The similarity is then presented graphically, which allows the time structure of the video to be visualized and browsed. Similarity can be rapidly calculated for other video files as well, which enables content-based retrieval by example. A content-aware video browser featuring interactive similarity measurement is presented. A method for selecting training segments involves mouse click-and-drag operations over a time bar representing the duration of the video; similarity results are displayed as shades in the time bar. Another method involves selecting periodic frames of the video as endpoints for the training segment.
摘要:
Method for interactive selecting video consisting of training images from a video for a video similarity search and for displaying the results of the similarity search are disclosed. The user selects a time interval in the video as a query definition of training images for training an image class statistical model. Time intervals can be as short as one frame or consist of disjoint segments or shots. A statistical model of the image class defined by the training images is calculated on-the-fly from feature vectors extracted from transforms of the training images. For each frame in the video, a feature vector is extracted from the transform of the frame, and a similarity measure is calculated using the feature vector and the image class statistical model. The similarity measure is derived from the likelihood of a Gaussian model producing the frame. The similarity is then presented graphically, which allows the time structure of the video to be visualized and browsed. Similarity can be rapidly calculated for other video files as well, which enables content-based retrieval by example. A content-aware video browser featuring interactive similarity measurement is presented. A method for selecting training segments involves mouse click-and-drag operations over a time bar representing the duration of the video; similarity results are displayed as shades in the time bar. Another method involves selecting periodic frames of the video as endpoints for the training segment.
摘要:
A system is provided where a number of representative video snapshots are displayed on a timeline indicating a position to jump to in a video using control buttons, such as fast forward, reverse, or a timeline scroll bar. The video snapshots are obtained by opening a low resolution connection to a video server to receive forward looking video segments. The forward looking video segments are summarized and the representative video snapshots are selected and displayed to provide a reference frame for the user to select a timeline position, or snapshots directly linked to positions in the higher resolution video. Backward looking video segments may be similarly summarized from the low resolution connection, or alternatively from video previously played. As the video progresses, new forward looking video segments are received and new representative video snapshots are summarized and displayed in close proximity with the video being played.
摘要:
Techniques for generating timelines and event logs from one or more fixed-position cameras based on the identification of activity in the video are presented. Various embodiments of the invention include an assessment of the importance of the activity, the creation of a timeline identifying events of interest, and interaction techniques for seeing more details of an event or alternate views of the video. In one embodiment, motion detection is used to determine activity in one or more synchronized video streams. In another embodiment, events are determined based on periods of activity and assigned importance assessments based on the activity, important locations in the video streams, and events from other sensors. In different embodiments, the interface consists of a timeline, event log, and map.
摘要:
Techniques for generating action keyframes for a fixed-position camera based on the identification of activity in the video, an assessment of the importance of the activity, object recognition in the video, and interaction techniques for seeing more details of the video are presented. In different embodiments of the invention, the importance of activity is determined based on the amount of activity, important locations in the video streams, detected features such as faces, and events from other sensors.
摘要:
A hypervideo summary comprised of multiple levels of related content and appropriate navigational links can be automatically generated from a media file such as a linear video. A number of algorithms and selection criteria can be used to modify how such a summary is generated. Viewers of an automatically-generated hypervideo summary can interactively select the amount of detail displayed for each portion of the summary. This selection can be done by following explicit navigational links, or by changing between media channels that are mapped to the various levels of related content. This description is not intended to be a complete description of, or limit the scope of, the invention. Other features, aspects, and objects of the invention can be obtained from a review of the specification, the figures, and the claims.
摘要:
Systems and methods recommend documents or directories in an enterprise context by analyzing use proximity in the organizational hierarchy to find similar users. Evidence may be used from different sources to gage the degree of interest a user may have in a document. Such pieces of evidence may include viewing page thumbnails, viewing the document online, or printing, saving, or bookmarking the document. To make managing these different sources of evidence from users related by different degrees tractable, a model may be created where evidence values decay by different amounts over time and are combined with different weights. Additionally, these systems and methods also can make use of the directory structure of the document space to recommend directories as well as individual documents.
摘要:
An adaptive, interactive visual workspace for viewing groups of files based on their relationships. Relationships of files are visualized using iterative refinement of categories through a direct-manipulation graph-based layout. The visual workspace starts with a fully connected graph linking thumbnail images of related files that is then partitioned into neighborhoods in response to a user creating file stacks corresponding to different categories. Normalized spring lengths improve the overall quality of the layout. Different modes for membership in neighborhoods avoid confusing motion of files and help a user to manually organize the workspace. Additionally, retrieved files can be added without having to significantly move the previous files. Different visualization techniques indicate which files are related to each other. An elliptical-shaped window displaying rings of elliptic thumbnail images presents a group of files in an easily visible and space-efficient way. Different zoom rates are used for file location, and surrogate sizes allow users to increase the separation between files while still increasing the surrogate sizes.
摘要:
Techniques for generating a storyboard are disclosed. In one embodiment of the invention the storyboard is comprised of videos from one or more cameras based on the identification of activity in the video. Various embodiments of the invention include an assessment of the importance of the activity, the creation of a storyboard presentation based on importance and interaction techniques for seeing more details or alternate views of the video. In one embodiment, motion detection is used to determine activity in one or more synchronized video streams. Periods of activity are recognized and assigned importance assessments based on the activity, important locations in the video streams, and events from other sensors. In different embodiments, the interface consists of a storyboard and a map.
摘要:
Techniques for classifying video frames using statistical models of transform coefficients are disclosed. After optionally being decimated in time and space, image frames are transformed using a discrete cosine transform or Hadamard transform. The methods disclosed model image composition and operate on grayscale images. The resulting transform matrices are reduced using truncation, principal component analysis, or linear discriminant analysis to produce feature vectors. Feature vectors of training images for image classes are used to compute image class statistical models. Once image class statistical models are derived, individual frames are classified by the maximum likelihood resulting from the image class statistical models. Thus, the probabilities that a feature vector derived from a frame would be produced from each of the image class statistical models are computed. The frame is classified into the image class corresponding to the image class statistical model which produced the highest probability for the feature vector derived from the frame. Optionally, frame sequence information is taken into account by applying a hidden Markov model to represent image class transitions from the previous frame to the current frame. After computing all class probabilities for all frames in the video or sequence of frames using the image class statistical models and the image class transition probabilities, the final class is selected as having the maximum likelihood. Previous frames are selected in reverse order based upon their likelihood given determined current states.