Abstract: A method of coarse representation of a visual object's shape for search/query/filtering applications uses a binding box that fully encompasses the object of interest within the image to extract a feature vector. Once the feature vector is available, matching based on specific queries may be performed using a search engine to compare the query number to an appropriate element of the feature vector, performing sorting to pick the best matches.

Abstract: A method of extracting regions of homogeneous color from a digital picture divides the digital picture into blocks and generates a feature vector for each block as a set of moments of the data for the block. The distance between the feature vector of each block and the feature vectors of the nearest neighboring blocks are determined using either a weighted Euclidean distance metric or a probability mass function-based distance metric. The maximum distance is the gradient value for the block, and the set of gradient values over all the blocks form a color gradient field. The gradient field is digitized and smoothed, and then segmented into regions of similar color characteristics using a watershed algorithm.

Abstract: A semi-automatic method of tracking color objects in a video image sequence starts by separating the objects on the basis of color and identifying an object of interest to track. A Kalman predictive algotithm in used to predict the position of the centroid of the object of interest through successive frames. From the predicted position the actual centroid is measured and the position and velocity are smoothed using a Kalman filter. Error recovery is provided in the event the centroid falls outside the field of view or falls into an area of a different color, or in the event the tracking algorithm breaks down.

Abstract: A method of generating normalized bitmap representation for the shape of a visual object for use in search/query/filtering applications segments an image into visual objects. The samples belonging to a visual object of interest are identified. The identified samples that form the largest connected blob are reduced to an un-normalized bitmap. The un-normalized bitmap is then normalized using the mean and covariance of the valid samples to generate the normalized bitmap representation having a standard height and having an orientation such that a principal direction is along a vertical direction. The normalized bitmap representation may be used with a query to search a database of images where the visual objects all have associated normalized bitmap representations. The query bitmap is normalized and matched to each normalized bitmap representation. The visual objects having the lowest mismatch values of their normalized bitmap representation with the query bitmap are identified as the objects of the search.

Abstract: A method of indexing and searching a video database having a plurality of video shots uses 3-D camera motion parameters. For each video shot the 3-D camera motion parameters are estimated, rates of tracking, booming, dollying, panning, tilting, rolling and zooming are computed, and the results are indexed in a metadata index file in the video database according to the types of camera. The video database is searched by selecting one of the types of camera motion and submitting a query. The query is processed to identify those video shots in the video database that satisfy the query in order of priority. The highest priority video shots are displayed for the user.

Abstract: A method of indexing and searching a video database having a plurality of video shots uses 3-D camera motion parameters. For each video shot the 3-D camera motion parameters are estimated, rates of tracking, booming, dollying, panning, tilting, rolling and zooming are computed, and the results are indexed in a metadata index file in the video database according to the types of camera. The video database is searched by selecting one of the types of camera motion and submitting a query. The query is processed to identify those video shots in the video database that satisfy the query in order of priority. The highest priority video shots are displayed for the user.

Abstract: A histogram-based segmentation of an image, frame or picture of a video signal into objects via color moments is initiated by defining a relatively large area within the object. The defined area is characterized by its color information in the form of a limited set of color moments representing a color histogram for the area. Based upon the set of color moments, color moments generated for small candidate blocks within the image, an automatically generated weighting vector, distance measures for the blocks from a central block in the object and a tolerance the area is grown to encompass the object to the extent of its boundaries. The initial set of color moments are then updated for the entire object. Those candidate blocks within the object serve to segment the object from the image.

Abstract: A method of generating a 2-D extended image from a video sequence representing a natural 3-D scene first determines motion parameters for a camera that recorded the scene with respect to a bakcground object from the video sequence using a structure-from-motion algorithm. The motion parameters include a rotation matrix, a translation vector and a depth map representing the depth of each point in the background object from the camera. Next from the motion parameters and depth map the 2-D extended image is generated for the background object as a composition of the images from the video sequence using a plane perspective projection technique. The background object may be layered as a function of depth and flatness criteria to form a set of layered 2-D extended images for the background object from the video sequence.

Abstract: A histogram-based segmentation of images in a video signal via color moments is initialized by a user defining regions in objects of interest from one or more images, key frames or pictures of the video signal. For each rectangle a normalized average color moment and associated co-variance matrix are determined which define a color class for that rectangle. From the normalized average color moment and associated co-variance garbage parameters are generated. Segmentation is then performed on a block basis on each image of the video sequence, a normalized color moment being generated for each block. Using a log likelihood test the closest color class for the block is determined. Based upon the closest color class and the garbage parameters for that color class a final determination is made in a two stage test as to whether the block belongs to the closest class or to a “garbage” class.

Abstract: A method of modifying a group of pictures (GOP) structure in an MPEG video signal from a low-delay mode bitstream having I and P pictures to a non-low-delay bitstream having I, P and B pictures uses the motion vectors from the low-delay mode bitstream to derive the motion vectors for the non-low-delay mode bitstream. Motion vectors for anchor pictures for the non-low-delay mode bitstream are converted from the motion vectors for the corresponding pictures in the low-delay mode bitstream. Motion vectors for the B pictures in the non-low-delay mode bitstream are converted from the motion vectors for the corresponding P pictures in the low-delay mode bitstream. The converted motion vectors for the non-low-delay mode bitstream are used in recoding an uncompressed video signal derived from the low-delay mode bitstream to produce the non-low-delay mode bitstream.

Abstract: A bit rate control mechanism for a digital image or video compression system estimates a complexity parameter for a current picture, or block of samples, of a video signal as a function of parameters for a prior picture of the video signal, which parameters include a bit rate. From the complexity parameter a quality factor for the current picture is determined and applied to a quantizer to compress the current picture. A complexity pre-processor may also be used to detect scene changes in the video signal prior to estimating the complexity parameter. If there is a scene change detected, then the rate control mechanism is reset prior to estimating the complexity parameter for the first picture in the new scene. Also a video buffer verifier is controlled so that the buffer occupancy at the end of a specified image sequence is at a target value so that looping and editing applications are facilitated.