Abstract: One embodiment relates to a computer-implemented method for automated object recognition and classification in scenes using segment-based object extraction. The method includes automated procedures for receiving video images, creating segmentation maps from said images, grouping segments so as to form extracted objects, extracting features from said extracted objects, classifying said extracted objects using said features. Other features, aspects and embodiments are also disclosed.

Abstract: An automated method for classifying an object in a sequence of video frames. The object is tracked in multiple frames of the sequence of video frame, and feature descriptors are determined for the object for each of the multiple frames. Multiple classification scores are computed by matching said feature descriptors for the object for each of the multiple frames with feature descriptors for a candidate class in a classification database. Said multiple classification scores are aggregated to generate an estimated probability that the object is a member of the candidate class. Other embodiments, aspects and features are also disclosed.

Abstract: One embodiment disclosed relates to the use of object motion estimation to interlace a progressive video sequence. One of a plurality of consecutive frames is segmented and motion vectors for each segment are determined though object motion estimation. Interpolated motion vectors are used to construct at least one intermediate frame, and interlaced fields are extracted from the new sequence of frames that includes intermediate frames. An interlaced sequence with smooth, incremental motion is thus constructed from a progressive video sequence.

Abstract: One embodiment of the invention pertains to a method of determining a measure of image complexity. An image is subdivided the image into a plurality of small image regions. Multiple statistical tests are performed to determine the similarity of a pair of adjacent image regions. If said pair passes the multiple statistical tests, then the pair of adjacent image regions are grouped together into one new region. The resulting merged regions may be weighted according to geometry and/or color variance, and the weights may be summed to produce an image complexity measure.

Abstract: One embodiment of the invention relates to a method of merging segments to form supersegments in an image. The image consists of a plurality of segments that are constituent portions of the image. At least one candidate segment(s) and at least one neighboring segment(s) for each candidate segment are identified. An error statistic for each pair, consisting of a candidate segment and a corresponding neighboring segment, is computed. A neighboring segment is determined that results in a smallest error statistic for a given candidate segment. A determination is also made as to whether the smallest error statistic is sufficiently small to merit merging of the corresponding pair of segments. The corresponding pair of segments is merged to create one supersegment. The supersegment is a new segment including all pixels formerly contained in one of the two segments that were merged.

Abstract: A method and apparatus for client-side detection of network congestion in a best-effort packet network comprising streaming media traffic is disclosed. Said method and apparatus provide for quality streaming media services in a congested network with constrained bandwidth over the last-mile link. A client media buffer detects at least one level of congestion and signals a server to enact at least one error mechanism. Preferred error mechanisms include packet retransmissions, stream prioritization, stream acceleration, changes in media compression rate, and changes in media resolution. Said method and apparatus allow distributed management of network congestion for networks comprising multiple clients and carrying significant streaming media traffic.

Abstract: A process and apparatus for allocating bits between the stages of a multi-stage digital image compression scheme with quantizer feedback is disclosed. The process selects a quantization schedule (from a finite number of possibilities) for all stages that minimizes an estimate of total distortion while keeping total bit demand below a constraint; uses said schedule to actually quantize one stage of the compression scheme; selects a new quantization schedule for the remaining stages using estimates for distortion and bit demand that are updated to reflect the quantization of the previous stage; actually quantizes the next stage according to the new quantization schedule; and continues recursively until it has chosen an actual quantization scale for every stage of the compression scheme.

Abstract: A process and apparatus for identifying abrupt cuts or scene changes in any ordered sequence of images. In one specific embodiment, two or more consecutive images from a sequence are introduced to a segmenter as digital frames. The segmenter independently divides each of these frames into pixel regions or segments according to some common characteristic so that every pixel belongs to exactly one segment. A segment analysis unit then performs some statistical analysis on the segment data for each of the frames and generates composite statistics for each frame. A frame comparison unit then examines these composite statistics to determine whether these frames belong to a consistent scene of images. If the composite statistics for these frames differ sufficiently, the comparison unit declares the latter frame in the sequence to belong to a new scene. This information may then be transmitted back to the data source for the purpose of marking the scene change or for any other purpose.

Abstract: In a method and apparatus for predicting and coding motion vectors in a video compression scheme, an ordered list of segments for a reference frame is used to create a hierarchy of segments with a plurality of levels. Motion vectors for segments in the top level are entropy coded, and these vectors are used to predict vectors for segments at the next level. Residual vectors are entropy coded to correct these predictions, and the process of prediction from above and coding residuals continues recursively down through the hierarchy of segment levels. Information about the previous motion of segments may be exploited in the prediction process. In a complementary method and apparatus for decoding motion vectors, the same segment hierarchy is used to predict motion vectors by the same method used during encoding, and these predicted vectors are added to residual motion vectors to reconstruct the actual motion vectors.

Abstract: A system and method for delivering user targeted information from a web server to a personal digital assistant (PDA). A personal profile associated with a user is entered, wherein the personal profile includes demographic information associated with the user. Synchronization is initiated, wherein initiating synchronization includes generating a query as a function of the user's demographic information and transmitting the query to a web server. One or more Uniform Resource Locators (URLs) is selected as a function of the query. The computer receives the selected Uniform Resource Locators (URL) and opens web pages corresponding to the received URLs. The computer receives HTML code corresponding to the web pages, parses the received HTML code and converts the parsed HTML code into a format appropriate for the PDA.

Abstract: One embodiment disclosed relates to the use of object motion estimation to interlace a progressive video sequence. One of a plurality of consecutive frames is segmented and motion vectors for each segment are determined though object motion estimation. Interpolated motion vectors are used to construct at least one intermediate frame, and interlaced fields are extracted from the new sequence of frames that includes intermediate frames. An interlaced sequence with smooth, incremental motion is thus constructed from a progressive video sequence.

Abstract: An image segmenter uses one or more techniques to accurately segment an image, including the use of a progressive flood fill to fill incompletely bounded segments, the use of a plurality of scaled transformations and guiding segmentation at one scale with segmentation results from another scale, detecting edges using a composite image that is a composite of multiple color planes, generating edge chains using multiple classes of edge pixels, generating edge chains using the plurality of scaled transformations, and/or filtering spurious edges at one scale based on edges detected at another scale.

Abstract: A method and apparatus for predicting and coding motion vectors in a video compression scheme is disclosed. An ordered list of segments for a reference frame is used to create a hierarchy of segments with a plurality of levels. Motion vectors for segments in the top level are entropy coded, and these vectors are used to predict vectors for segments at the next level. Residual vectors are entropy coded to correct these predictions, and the process of prediction from above and coding residuals continues recursively down through the hierarchy of segment levels. Information about the previous motion of segments may be exploited in the prediction process. A complementary method and apparatus for decoding motion vectors is also disclosed. The same segment hierarchy is used to predict motion vectors by the same method used during encoding, and these predicted vectors are added to residual motion vectors to reconstruct the actual motion vectors.

Abstract: A process and apparatus for allocating bits between the stages of a multi-stage digital image compression scheme with quantizer feedback is disclosed. The process selects a quantization schedule (from a finite number of possibilities) for all stages that minimizes an estimate of total distortion while keeping total bit demand below a constraint; uses said schedule to actually quantize one stage of the compression scheme; selects a new quantization schedule for the remaining stages using estimates for distortion and bit demand that are updated to reflect the quantization of the previous stage; actually quantizes the next stage according to the new quantization schedule; and continues recursively until it has chosen an actual quantization scale for every stage of the compression scheme.

Abstract: A process and apparatus for identifying abrupt cuts or scene changes in any ordered sequence of images. In one specific embodiment, two or more consecutive images from a sequence are introduced to a segmenter as digital frames. The segmenter independently divides each of these frames into pixel regions or segments according to some common characteristic so that every pixel belongs to exactly one segment. A segment analysis unit then performs some statistical analysis on the segment data for each of the frames and generates composite statistics for each frame. A frame comparison unit then examines these composite statistics to determine whether these frames belong to a consistent scene of images. If the composite statistics for these frames differ sufficiently, the comparison unit declares the latter frame in the sequence to belong to a new scene. This information may then be transmitted back to the data source for the purpose of marking the scene change or for any other purpose.

Abstract: A method and apparatus for client-side detection of network congestion in a best-effort packet network comprising streaming media traffic is disclosed. Said method and apparatus provide for quality streaming media services in a congested network with constrained bandwidth over the last-mile link. A client media buffer detects at least one level of congestion and signals a server to enact at least one error mechanism. Preferred error mechanisms include packet retransmissions, stream prioritization, stream acceleration, changes in media compression rate, and changes in media resolution. Said method and apparatus allow distributed management of network congestion for networks comprising multiple clients and carrying significant streaming media traffic.