Abstract: A new component is provided within a switch that ‘learns’ from the messages between the switch and the controller regarding how to behave like a controller under different network conditions and how to optimize the switch flow tables by implementing techniques like aggregation to save memory space, wherein the new component can act as a ‘proxy controller’ that resides within the switch. This new component can be activated when a specific trigger happens, wherein the trigger is programmable into the switch. After the trigger is active, the proxy/local controller starts undertaking some or all of the roles of the controller until the trigger is deactivated. Because different types of triggers can be programmed, the new component can have one or more of the control functions. The teaching of the new component (which can be a hardware chip or software) is performed outside the switch using machine-learning techniques (e.g. deep learning).

Abstract: A novel (software defined network) SDN control plane is introduced having new system capabilities to activate and deactivate controllers in real-time upon automatic measurement of network control traffic and service requirements, and proper controller interactions with network switches as control plane topology changes. Also introduced is a control flow table, which defines the assignment of certain control flows (by originator, location, service type, etc.) to different controllers within the SDN.

Abstract: A novel (software defined network) SDN control plane is introduced having new system capabilities to activate and deactivate controllers in real-time upon automatic measurement of network control traffic and service requirements, and proper controller interactions with network switches as control plane topology changes. Also introduced is a control flow table, which defines the assignment of certain control flows (by originator, location, service type, etc.) to different controllers within the SDN.

Abstract: Systems and methods are described to dynamically set up and tear down service-enabled flow-paths across multiple SDN domains. An SDN controller of a domain periodically shares with other SDN controllers (of other domains) availability of service-enabled paths (aka, a summarized topology) of its respective network as well as associated service parameters of each path segment using a global nomenclature (aka, a dictionary) with other SDN controllers so as to enable each controller to understand and interpret the shared data in the same manner. Hence, each SDN controller can autonomously construct a complete network graph of available service-enabled paths across many domains. Another feature is the use of summarized (internal) topology of each domain as opposed to just using the peering-link related service information to compute desirable path alternatives.

Abstract: A lossless, reversible data embedding technique uses a generalization of least-significant-bit modification. The portions of the signal that are susceptible to embedding distortion are compressed and sent as part of the embedded payload. A prediction-based conditional entropy coder uses unaltered portions of the host signal to improve lossless data capacity.

Abstract: The system automatically extracts cinematic features, such as shot types and replay segments, and object-based features, such as the features to detect referee and penalty box objects. The system uses only cinematic features to generate real-time summaries of soccer games, and uses both cinematic and object-based features to generate near real-time, but more detailed, summaries of soccer games. The techniques include dominant color region detection, which automatically learns the color of the play area and automatically adjusts with environmental conditions, shot boundary detection, shot classification, goal event detection, referee detection and penalty box detection.

Abstract: A lossless, reversible data embedding technique uses a generalization of least-significant-bit modification. The portions of the signal that are susceptible to embedding distortion are compressed and sent as part of the embedded payload. A prediction-based conditional entropy coder uses unaltered portions of the host signal to improve lossless data capacity.

Abstract: An object-oriented method for describing the content of a video sequence comprises the steps of (a) establishing a temporal object-based segment for an object of interest; (b) describing the temporal object-based segment by describing one or more semantic motions of the object within its temporal object-based segment; and (c) describing the temporal object-based segment by describing one or more semantic interactions of the object with one or more other objects within its temporal object-based segment. The semantic motions of the object may be further described in terms of the properties of elementary coherent motions within the semantic motion. Additionally, the semantic interactions of the object may be further described in terms of the properties of the elementary spatio-temporal relationships among the interacting objects.

Abstract: A method and apparatus for coding video data permits coding of video information with optional, enhanced functionalities. Video data is coded as base layer data and enhancement layer data. The base layer data includes convention motion compensated transform encoded texture and motion vector data. Optional enhancement layer data contains mesh node vector data. Mesh node vector data of the enhancement layer may be predicted based on motion vectors of the base layer. Thus, simple decoders may decode the base layer data and obtain a basic representation of the coded video data. However, more powerful decoders may decode both the base layer and enhanced layer to obtain decoded video permitting functionalities. An embodiment of the present invention provides a back channel that permits a decoder to affect how mesh node coding is performed in the encoder. The decoder may command the encoder to reduce or eliminate encoding of mesh node motion vectors.

Abstract: A method and apparatus for coding video data permits coding of video information with optional, enhanced functionalities. Video data is coded as base layer data and enhancement layer data. The base layer data includes convention motion compensated transform encoded texture and motion vector data. Optional enhancement layer data contains mesh node vector data. Mesh node vector data of the enhancement layer may be predicted based on motion vectors of the base layer. Thus, simple decoders may decode the base layer data and obtain a basic representation of the coded video data. However, more powerful decoders may decode both the base layer and enhanced layer to obtain decoded video permitting functionalities. An embodiment of the present invention provides a back channel that permits a decoder to affect how mesh node coding is performed in the encoder. The decoder may command the encoder to reduce or eliminate encoding of mesh node motion vectors.

Abstract: A system and method of encoding and decoding a dynamic mesh includes encoding and decoding a mesh geometry of a set of node points and encoding and decoding a mesh node motion vector for each node point.

Abstract: A method for motion tracking and constructing a mosaic of video objects is disclosed. Also disclosed is a method for synthetic object transfiguration from a mosaic. A 2-D triangular mesh is employed to represent a video object, which permits to describe the motion of the object by the displacements of the node points of the mesh, and to describe any intensity variations by the contrast and brightness parameters estimated for each node point. Using the temporal history of the node point locations, the nodes of the 2-D mesh are continued to be tracked even when they become invisible because of self-occlusion or occlusion by another object. By adding new nodes or updating the 2-D triangular mesh, any uncovered parts of the video object are incrementally added to the mosaic. A finite number of views representing the constructed mosaic are selected and used for synthetic transfiguration of a replacement object in place of the original one.

Abstract: In a temporal sequence of digitized image frames of video signals, the spatial and temporal image gradients and the pixel-to-pixel motion vectors (dense motion vectors) are obtained between two consecutive image frames. A shape-adaptive triangular patch mesh model overlay is provided on the first image frame such that the location of node points of each patch is determined by the spatial image gradients of the first frame and the pixel-to-pixel motion vectors. A priority ranking of patches is established before determining the node point motion vectors. The node point motion vectors, representing the motion of each of the node points of the triangular mesh patches, are estimated by a linear least-squares solution to an affine transformation of the mesh overlay on the first frame into the second frame. Failure regions are identified and are revised in accordance with a data bit budget. All data are coded prior to transmission at a constant data bit rate by a sending unit to a receiving unit over a data link.