Abstract: A watermark is embedded into video/image/multimedia data using spread spectrum methodology. The watermark is extracted from watermarked data without the use of an original or unwatermarked version of the data by using MPEG/JPEG coefficients. The image to be watermarked is divided into subimages. Each subimage is embedded with a watermark. When extracting the watermark, the result from each subimage is combined to determine the originally embedded watermark.

Abstract: A digital watermark is inserted into multimedia data containing two image fields by placing a positive watermark into a first field and a negative watermark into a second field. The positive watermark and negative watermark are opposite of one another. The two fields can be interlaced fields of a field-based video signal or alternate rows of a frame-based video signal. The watermark is extracted from field-based watermarked data by separating the watermarked data into two fields and subtracting one of the fields from the other field to generate a watermarked signal. The resultant watermarked signal is processed in a conventional manner to extract and detect the watermark.

Abstract: Digital watermarking of audio, image, video or multimedia data is achieved by inserting the watermark into the perceptually significant components of a decomposition of the data in a manner so as to be visually imperceptible. In a preferred method, a frequency spectral image of the data, preferably a Fourier transform of the data, is obtained. A watermark is inserted into perceptually significant components of the frequency spectral image. The resultant watermarked spectral image is subjected to an inverse transform to produce watermarked data. The watermark is extracted from watermarked data by first comparing the watermarked data with the original data to obtain an extracted watermark. Then, the original watermark, original data and the extracted watermark are compared to generate a watermark which is analyzed for authenticity of the watermark.

Abstract: Digital watermarking of data, including image, video and audio data, is performed by repeatedly inserting the watermark into subregions or subimages of the data. Similarly, the watermark is repeatedly extracted from the subregions of the data.

Abstract: A watermark is embedded into audio/video/image/multimedia data using spread spectrum methodology. The watermark is extracted from watermarked data without the use of an original or unwatermarked version of the data by using spatial or temporal local averaging of the frequency coefficients of the watermarked data.

Abstract: A pattern recognition method uses unsupervised metric learning starting from a mixture of normal densities which explains well observed data. An improved decision rule is provided for selecting the reference database element most likely to correspond to a query.

Abstract: A technique for determining the ego-motion between two frames of an image that includes estimating likely values of the translation and rotation and making a five-dimensional search over the space of likely translations and rotations and finding the global minimum at the current motion estimate by a sum of squared differences cost function.

Abstract: A computer-implemented process for determining a ranked set of solutions to a bipartite graph matching problem, such is involved in multi-target tracking, maximum flow matching, or data association in matching elements of a first set to elements of a second set to minimize the cost of the association. A feature of the process is a partitioning into subproblems such that the most probable solutions are in the smaller subproblems and the computation begins by solving the smallest subproblem in the search for the best solution.

Abstract: Dynamic histogram warping is performed on histograms extracted from an image pair of a scene. The warped histograms are remapped to the image pair and the resulting remapped image pair is subsequently subjected to image processing.

Abstract: A queryless, multimedia database search method incorporating a Bayesian inference engine that refines its answer with each user response. The set of user responses includes of a series of displays and user actions, and is defined by a relatively simple user interface.

Abstract: A technique for compensating for egomotion of the camera used to record a pair of two-dimensional views of a scene when the pair of images is to be used to provide a three dimensional representation of the scene. The technique involves comparing histograms of the intensity levels of pixels of corresponding epipolar lines in the pair of images for assumed amounts of egomotion to identify the amount that results in the smallest total of the sums of squared differences of the histograms.

Abstract: A stereoscopic computer vision system that uses a novel algorithm for obtaining the best match between corresponding features in the left and right images of desired objects in the image scene, finds the disparity between corresponding features in the left and right views and then uses the disparity to calculate the distance of the desired object from the two cameras used to provide the left and right images.

Abstract: A teleconferencing video system uses two cameras at each station symmetrically positioned on opposite sides of the optical axis between the speaker and the monitor used to produce an image of the listener. The two cameras are used to provide from the two images observed by the cameras a virtual image corresponding to the image that would be viewed by a camera located on the first-mentioned optical axis. This system permits eye contact to be maintained between a speaker and a listener at different stations whereby the feeling of intimacy is enhanced.

Abstract: Computer vision algorithms for detecting contours or localized regions of prescribed intensity distributions in an image scene utilizing multiple hypothesis probabilistic data association techniques to overcome the effect of noise or false readings.