Abstract: A method and apparatus generate a forward error correcting code for use with a plurality of source packets to be transmitted over a network. In one exemplary embodiment, the method comprises identifying a plurality of candidate packet masks and selecting that one of the masks that has the lowest expected residual packet loss value (“RPL”). RPL is calculated using an effective recovery rate, which in turn is a function of at least one network performance parameter (such as packet loss) and one or more channel code parameters such as the number of source packets and/or FEC packets. The error correcting code can be generated using the selected packet mask and at least one of the source packets.

Abstract: Encoding and decoding is accomplished herein using a multi-reference picture buffer. Reference frames are assigned to reference picture buffers when used for encoding and are identified with picture IDs unique relative to other picture IDs currently associated with the reference picture buffers. The maximum picture number used as the picture ID can be based on a number of reference picture buffers available to encode and decode the frames plus a value based on an expected error rate. The picture IDs can be assigned based on a least recently used policy. When a reference frame is no longer needed for encoding and decoding, a picture ID number associated with the reference frame can be released for re-assignment to a new reference frame.

Abstract: Temporal classified filtering encodes image data by applying filters assigned to classes of pixels in a target frame to predict values for the pixels. The pixels are classified based on their associated motion vectors and the motion vectors are used to position the filters on the reference frame. Prediction error values are also calculated. The filters, motion vectors, and prediction errors represent the pixels in the encoded image data. The reference frame may be a past or future frame of the image data, and multiple reference frames of various combinations of past and future frames may be used. The filters for multiple reference frames are three-dimensional comprising a two-dimensional filter for each reference frame. The filters may be pre-determined or generated as the frames are encoded. The image data is recreated by applying the filters to the reference frames and correcting the resulting predictions with the prediction error values.

Abstract: An architecture and framework for speech/noise classification of an audio signal using multiple features with multiple input channels (e.g., microphones) are provided. The architecture may be implemented with noise suppression in a multi-channel environment where noise suppression is based on an estimation of the noise spectrum. The noise spectrum is estimated using a model that classifies each time/frame and frequency component of a signal as speech or noise by applying a speech/noise probability function. The speech/noise probability function estimates a speech/noise probability for each frequency and time bin. A speech/noise classification estimate is obtained by fusing (e.g., combining) data across different input channels using a layered network model.

Abstract: An initial value is assigned to a center point for each cluster in a plurality of clusters. Each point in a point space is assigned to a closest cluster based on the distance between the each point and the center of nearest cluster. A first-assignment value is determined for each center point using the clusters the points are assigned to. A first-assignment dynamic validity index of a current cluster configuration is evaluated. Each point in the point space is reassigned to the closest cluster based on the first-assignment value of each center. A second-assignment value is determined for the center of each cluster according to the reassigning. A second-assignment dynamic validity index is evaluated using the second-assignment values. The current cluster configuration is selected if the difference between the dynamic validity indices is less than a threshold.

Abstract: A method and apparatus generates a forward error correcting code for use with a plurality of source packets to be transmitted over a network. In one exemplary embodiment, the method comprises identifying a plurality of candidate packet masks and selecting that one of the masks that has the lowest expected residual packet loss value (“RPL”). RPL is calculated using an effective recovery rate, which is in turn is a function of at least one network performance parameter (such as packet loss) and one or more channel code parameters such as the number of source packets and/or FEC packets. The error correcting code can be generated using the selected packet mask and at least one of the source packets.

Abstract: A method and system for providing adaptive media optimization are described. Aspects of the invention modify video encoding and network transmission settings to optimize the user viewing experience. The system and method sample video content to determine various content features of the video. The system and method use the identified content features in conjunction with network statistics to modify encoding settings and network transmission options to ensure a minimum of interruption in the transmitted video. Previously generated lookup tables ensure efficient mapping of video content and network conditions to encoding and transmission settings.

Abstract: A method and system for providing content aware media adaptation are described. Aspects of the invention adaptively down-sample a source video to optimize the encoding process of the source video. The system and method extract content characteristics from the source video by sampling the source video, and then classify the video into one or more content classes based on the extracted characteristics. The content class of the video is used to determine one or more down-sampling settings for the source video. In some aspects, the down-sampling settings are derived by sampling a plurality of videos and determining optimal transitional rates for the plurality of videos. The sampled videos may be used to generate a decision boundary to classify whether a particular video is a good candidate for spatial down-sampling.

Abstract: An initial decimation filter is applied to an original frame to generate a decimated frame. An optimized prediction filter is extracted from both the decimated frame and the original frame, while the initial decimation filter is held fixed. A predicted from is generated from the optimized prediction filter and the decimated frame, and an optimize decimation filter is extracted from the decimated frame and the predicted frame, while the optimized prediction filter remains fixed.

Abstract: A plurality of initial decimation filters are concatenated into an initial combined decimation filter. The initial decimation filters are assigned to a plurality of classes which associate the initial decimation filters with regions of an image to be predicted. An input signal is generated from the initial combined decimation filter and is weighted with a plurality of prediction filters. A correlation matrix and an observation vector are generated for the initial combined decimation filter from the input signal. An optimized combined decimation filter is extracted from the correlation matrix and observation vector. The optimized filter comprises a plurality of optimized decimation filters. The optimized combined decimation filter is de-concatenated into the plurality of optimized decimation filters.

Abstract: An architecture and framework for speech/noise classification of an audio signal using multiple features with multiple input channels (e.g., microphones) are provided. The architecture may be implemented with noise suppression in a multi-channel environment where noise suppression is based on an estimation of the noise spectrum. The noise spectrum is estimated using a model that classifies each time/frame and frequency component of a signal as speech or noise by applying a speech/noise probability function. The speech/noise probability function estimates a speech/noise probability for each frequency and time bin. A speech/noise classification estimate is obtained by fusing (e.g., combining) data across different input channels using a layered network model.

Abstract: An architecture and framework for speech/noise classification of an audio signal using multiple features with multiple input channels (e.g., microphones) are provided. The architecture may be implemented with noise suppression in a multi-channel environment where noise suppression is based on an estimation of the noise spectrum. The noise spectrum is estimated using a model that classifies each time/frame and frequency component of a signal as speech or noise by applying a speech/noise probability function. The speech/noise probability function estimates a speech/noise probability for each frequency and time bin. A speech/noise classification estimate is obtained by fusing (e.g., combining) data across different input channels using a layered network model.

Abstract: A method of irregular motion compensation includes using contours of objects in a reference image to tile the reference image into a plurality of irregular shapes, and mapping each irregular shape to a location in a target image by assigning a motion vector to each irregular shape.

Abstract: Multi-field taps are defined based on a decimated field. The multi-field taps are used to generate correlation matrices, the elements of which are used to generate covariance matrices. A principal component space is obtained by projecting the correlation matrix elements on to eigenvectors. The principal component space is partitioned into classes and a least square filter set is generated for each class.

Abstract: A technique for performing motion estimation using an adaptive motion search range includes calculating motion vectors for blocks of pixels in a target image, using the calculated motion vectors to generate a search range associated with a target block of pixels in the target image, and using the generated search range to estimate motion of the target block of pixels.

Abstract: A method of generating an adaptive temporal filter is performed by constructing a motion vector area cell around each of a plurality of motion vectors in a target image, selecting a pixel in the target image, constructing a pixel area cell around the selected pixel, determining an overlap area between the motion vector area cells and the pixel area cell, generating filter weights from the overlap area, and using the filter weights to filter the selected pixel.

Abstract: An adaptive filter is applied to samples in an initial pattern of samples. The samples in the initial pattern correspond to a subset of the image pixels which are to be predicted using the samples. A level value of each sample in the pattern is varied. The level value corresponds to the intensity/signal value of the sample, and may have a strong effect on the prediction power of the samples. A first prediction pattern (i.e., the prediction signal corresponding to the sampling pattern) is generated from the samples. A sample movement process is performed on a current sample in the first prediction pattern to change a current position of the current sample in the first prediction pattern. A local prediction of pixels influenced by the sample is updated if the sample's position is changed. A sample removal process is performed on the current sample to remove the sample from the first pattern if a criteria is satisfied. A local prediction of pixels is updated if the current sample is removed.

Abstract: A sample cell corresponding to a sample is computed to include pixels of an image according to sample cell criteria and shape. The sample cell shape and criteria define pixels to include in the sample cell. A target pixel cell corresponding to a target pixel includes pixels of the image according to a target pixel cell criteria defining which pixels are included in the target pixel cell. A sample filter coefficient is computed from pixels in the sample and target pixel cells. A filter tap corresponding to the target pixel is defined based on the sample filter coefficient. A tap extension corresponding to the target pixel is generated. The tap extension defines additional pixels of the image to be included in an extended target pixel cell according to a tap extension criteria. The extended target pixel cell includes pixel in the target pixel cell and the additional pixels. A tap extension filter coefficient based on pixels located in the extended target pixel cell and the sample cell is computed.

Abstract: A method for performing a sample level variation on an irregular sample is described herein. An initial sample level value for the sample is determined. An area of influence of the sample including pixels for which the sample is a factor in the prediction thereof is computed. Prediction errors of pixels in the area of influence are computed, along with a prediction error of the sample based on the prediction errors of the pixels. If the prediction error of the sample exceeds a prediction error threshold, a sample level change for the sample is computed based on only the prediction errors of the pixels. If the sample level change is less than a level change threshold, a new sample level value is assigned to the sample based on the sample level change.

Abstract: A method for determining values of motion vectors includes receiving an irregular pattern of motion vectors for a target image, estimating an initial value for each of the motion vectors, using the motion vectors to generate a tap structure for an adaptive temporal prediction filter, and using the tap structure to re-estimate the value of each motion vector.