Abstract: A method learns a structure of a video, in an unsupervised setting, to detect events in the video consistent with the structure. Sets of features are selected from the video. Based on the selected features, a hierarchical statistical model is updated, and an information gain of the hierarchical statistical model is evaluated. Redundant features are then filtered, and the hierarchical statistical model is updated, based on the filtered features. A Bayesian information criteria is applied to each model and feature set pair, which can then be rank ordered according to the criteria to detect the events in the video.

Abstract: A system for watermarking an image file selects coefficients using a selection procedure that is kept secret, and assigns the selected coefficients to coefficient pairs. The difference between the coefficients of the pairs is biased by a value that varies, preferably in a pseudo-random manner, and the biased differences are used to generate signature bits that characterize the authentic image at different locations. To detect an unauthorized alteration after the image file has been watermarked, coefficient pairs are selected using the same secret procedure that was originally used to generate the signature bits. The difference between the coefficients of the pairs is then biased and checked against the signature bits. Using a varying bias value permits a tolerance band for reducing false alarms to be used without the risk that would otherwise exist that evidence of an attack on the original image might be hidden in the tolerance band.

Abstract: A system for watermarking an image file selects coefficients using a selection procedure that is kept secret, and assigns the selected coefficients to coefficient pairs. The difference between the coefficients of the pairs is then used to generate multi-bit raw signature values that characterize the authentic image at different locations. To detect an unauthorized alteration after the image file has been watermarked, coefficient pairs are selected using the same secret procedure that was originally used to generate the raw signature values. The difference between the coefficients of the pairs is then checked against the raw signature values derived from the original image file. The raw signature values derived from the authentic image file may be placed in the header of the file or in a separate file. Alternatively, they may be embedded in host coefficients that are selected in accordance with a procedure that is kept secret.

Abstract: Digital watermarks are embedded in image data (102)in order to enable authentication of the image data and/or replacement of rejected portions of the image data. Authentication codes are derived by comparing selected discrete cosine transform (DCT) (104) coefficients within DCT data (106) derived from the original, spatial-domain image data. The authentication codes thus generated are embedded in DCT coefficients (612) other than the ones which were used to derive the authentication codes. The resulting, watermarked data can be sent or made available to one or more recipients who can compress or otherwise use the watermarked data. Image data derived from the watermarked data—e.

Abstract: A method analyzes a high-level syntax and structure of a continuous compressed video according to a plurality of states. First, a set of hidden Markov models for each of the states is trained with a training video segmented into known states. Then, a set of domain specific features are extracted from a fixed-length sliding window of the continuous compressed video, and a set of maximum likelihoods is determined for each set of domain specific features using the sets of trained hidden Markov models. Finally, dynamic programming is applied to each set of maximum likelihoods to determine a specific state for each fixed-length sliding window of frames of the compressed video.

Abstract: An invention for generating standard description records from multimedia information. The invention utilizes fundamental entity-relation models for the Generic AV DS that classify the entities, the entity attributes, and the relationships in relevant types to describe visual data. It also involves classification of entity attributes into syntactic and semantic attributes. Syntactic attributes can be categorized into different levels: type/technique, global distribution, local structure, and global composition. Semantic attributes can be likewise discretely categorized: generic object, generic scene, specific object, specific scene, abstract object, and abstract scene. The invention further classifies entity relationships into syntactic/semantic categories. Syntactic relationship categories include spatial, temporal, and visual categories. Semantic relationship categories include lexical and predicative categories.

Abstract: Caption boxes which are embedded in video content can be located and the text within the caption boxes decoded. Real time processing is enhanced by locating caption box regions in the compressed video domain (210) and performing pixel based processing operations within the region of the video frame in which a caption box is located. The captions boxes are further refined by identifying word regions (240) within the caption boxes and then applying character and word recognition processing (250) to the identified word regions. Domain based models are used to improve text recognition results. The extracted caption box text can be used to detect events of interest in the video content and a semantic model applied to extract a segment of video of the event of interest.

Abstract: A system and method analyzes a compressed video including a sequence of frames. The amount of a dominant feature in each frame of the compressed video is measured. A label is associated with each frame according the measured amount of the dominant feature. Views in the video are identified according to the labels, and the video is segmented into actions according to the views. The video can then be analyzed according to the action to determine significant events in the video.

Abstract: A method extracts high-level features from a video including a sequence of frames. Low-level features are extracted from each frame of the video. Each frame of the video is labeled according to the extracted low-level features to generate sequences of labels. Each sequence of labels is associated with one of the extracted low-level feature. The sequences of labels are analyzed using learning machine learning techniques to extract high-level features of the video.

Abstract: Techniques for signing multimedia content and verifying received multimedia content that are both robust and accurate are provided. Invariant features are extracted from multimedia content, and certain attributes are computed. Multimedia content is quantized, and extracted invariant features and quantized original multimedia content are encrypted to form a digital signature. The multimedia content and digital signature may be verified even after the introduction of distortions by using content registration. In a preferred embodiment, a refined authentication technique is used to obtain a continuous distance measure, to verify the authenticity of multimedia content based on a pre-defined threshold.

Abstract: Object-oriented methods and systems for permitting a user to locate one or more video objects from one or more video clips over an interactive network are disclosed. The system includes one or more server computers (110) comprising storage (111) for video clips and databases of video object attributes, a communications network (120), and a client computer (130). The client computer contains a query interface to specify video object attribute information, including motion trajectory information (134), a browser interface to browse through stored video object attributes within the server computers, and an interactive video player.

Abstract: Techniques for detecting moving video objects in a compressed digital bitstream (111) and for tools for editing compressed video are disclosed. Video objects (117) are detected and indexed by analyzing a compressed bitstream to locate scene cuts (112), estimating operating parameters for a camera which initially viewed the video (114), and detecting one or more moving video objects represented in the compressed bitstream by applying global motion compensation which account for the estimated operating parameters. Tools are provided for permitting dissolve, masking, freeze frame, slow and variable speed playback, and strobe motion special effects to compressed video. The tools may be implemented in a system for editing (130) compressed video information over a distributed network.

Abstract: Digital watermarks are embedded in image data (102)in order to enable authentication of the image data and/or replacement of rejected portions of the image data. Authentication codes are derived by comparing selected discrete cosine transform (DCT) (104) coefficients within DCT data (106) derived from the original, spatial-domain image data. The authentication codes thus generated are embedded in DCT coefficients (612) other than the ones which were used to derive the authentication codes. The resulting, watermarked data can be sent or made available to one or more recipients who can compress or otherwise use the watermarked data. Image data derived from the watermarked data—e.

Abstract: A method analyzes a high-level syntax and structure of a continuous compressed video according to a plurality of states. First, a set of hidden Markov models for each of the states is trained with a training video segmented into known states. Then, a set of domain specific features are extracted from a fixed-length sliding window of the continuous compressed video, and a set of maximum likelihoods is determined for each set of domain specific features using the sets of trained hidden Markov models. Finally, dynamic programming is applied to each set of maximum likelihoods to determine a specific state for each fixed-length sliding window of frames of the compressed video.

Abstract: A system for authentication of a digital image includes a signature generator for creating a robust digital signature for an original image based on instrument features of the image. An authentication processor extracts a set of invariant features for the original image from the digital signature, generates a corresponding set of invariant features for the present image to be authenticated and compares the two sets of invariant features to determine whether the image has been subjected to malicious manipulation. The invariant features include the polarity and magnitude of the difference between discrete cosine transform coefficients at corresponding coefficient locations in selected image block pairs. The intensity of the original image is also authenticated by comparing a mean value of coefficient of the original image to the mean value of the coefficient of the present image.

Abstract: Methods and a system are disclosed for processing an echocardiogram video of a patient's heart. The echocardiogram comprises at least a first sequence of consecutive video frames corresponding to a first view of the patient's heart concatenated with a second sequence of consecutive video frames corresponding to a second view of the patient's heart. The end-diastole phase of the patient's heart is monitored in each frame by detecting the electrocardiograph wave, and a key frame is selected upon the occurrence of the R-wave peak in the electrocardiograph wave in each of the first sequence of consecutive video frames and in the second sequence of consecutive video frames. The shape and color content of the echocardiogram image window is monitored in certain video frames, and a transition is detected when there is a change in the first feature between adjacent frames. A summary is generated which comprises by the video frames corresponding to the end-diastole phase.

Abstract: A system and method analyzes a compressed video including a sequence of frames. The amount of a dominant feature in each frame of the compressed video is measured. A label is associated with each frame according the measured amount of the dominant feature. Views in the video are identified according to the labels, and the video is segmented into actions according to the views. The video can then be analyzed according to the action to determine significant events in the video.

Abstract: Methods and a system are disclosed for processing an echocardiogram video of a patient's heart. The echocardiogram comprises at least a first sequence of consecutive video frames corresponding to a first view of the patient's heart concatenated with a second sequence of consecutive video frames corresponding to a second view of the patient's heart. The end-diastole phase of the patient's heart is monitored in each frame by detecting the electrocardiograph wave, and a key frame is selected upon the occurrence of the R-wave peak in the electrocardiograph wave in each of the first sequence of consecutive video frames and in the second sequence of consecutive video frames. The shape and color content of the echocardiogram image window is monitored in certain video frames, and a transition is detected when there is a change in the first feature between adjacent frames. A summary is generated which comprises by the video frames corresponding to the end-diastole phase.

Abstract: A method and system for maintaining the quality of video transported over wireless channels uses a transcoder to modify and maintain the optical resilience of an encoded bitstream. The transcoder increases the spatial resilience by reducing the number of blocks per slice, and increases the temporal resilience by increasing the proportion of I-blocks that are transmitted in each frame. Also, the transcoder maintains the same input bit rate by dropping less significant coefficients as it increases resilience. The transcoder of the present invention maintains the resilience at an optimal level to accommodate the prevailing channel conditions as measured by the BER of the wireless channel. Rate distortion theory is applied to determine the optimal allocation of bit rate among spatial resilience, temporal resilience and source rate, where it is has been found that the optimal allocation of the present invention (which occurs in near-real time) provides nearly the same result as doing an exhaustive search.

Abstract: Inverse motion compensating devices are used to transform motion compensated transform domain video signals to corresponding transform domain video signals. A compositing mechanism receives the transform domain video signals from the inverse motion compensating devices and performs transform domain video compositing operations directly on the transform domain video signals. The compositing operations result in a single composited signal incorporating each of the transform domain video signals. The single composited signal may be encoded for motion compensation to generate a single motion compensated transform domain video signal suitable for transmission. After transmission, inverse transform domain operations and inverse motion compensation operations may be performed to reconstruct the composited signal. The composited signal may then be projected on a video display device.