Abstract: The invention provides an image processing device capable of reducing flickering on a screen by simplifying the structure of an image processing device. The image processing device can include a statistical value calculating unit, a correction parameter calculating unit, and an image correcting unit. The statistical value calculating unit can generate statistical value data. The correction parameter calculating unit generates a control signal in accordance with the degree of change in the scene thereby generating a correction parameter by using a low pass filter whose time constant changes based on the control signal and equalizing the statistical value data. A look-up table can then be generated based on the correction parameter.

Abstract: Within the frameworks of hierarchical neural feed-forward architectures for performing real-world 3D invariant object recognition a technique is proposed that shares components like weight-sharing (2), and pooling stages (3, 5) with earlier approaches, but focuses on new methods for determining optimal feature-detecting units in intermediate stages (4) of the hierarchical network. A new approach for training the hierarchical network is proposed which uses statistical means for (incrementally) learning new feature detection stages and significantly reduces the training effort for complex pattern recognition tasks, compared to the prior art. The incremental learning is based on detecting increasingly statistically independent features in higher stages of the processing hierarchy. Since this learning is unsupervised, no teacher signal is necessary and the recognition architecture can be pre-structured for a certain recognition scenario.

Abstract: A method classifies data into multiple classes so that the data in each class have a class-conditional probability distribution. The class-conditional probability distributions of measured data are projected into a likelihood space. The projected class-conditional probability distributions in the likelihood space are then classified according to a discriminant classifier in likelihood space.

Abstract: The present invention is embodied in a system and method for statistically comparing a first set of digital data to at least a second set of digital data and matching the first set of digital data to appropriately corresponding portions of the second set of digital data. The first or the second set of digital data can be transformed during statistical analysis to enhance statistical analysis of the digital data.

Abstract: A fusion estimator is determined as the location of the most significant mode of a density function, which takes into account the uncertainty of the estimates to be fused. A mode detection method relies on mode tracking across scales. The fusion estimator is consistent and conservative, while handling naturally outliers in the data and multiple source models. The new estimator is applied for multiple motion estimation. Numerous experiments validate the theory and provide very competitive results. Other applications include generic distributed fusion, robot localization, motion tracking, registration of medical data, fusion for automotive tasks.

Abstract: Multimodal utterances contain a number of different modes. These modes can include speech, gestures, and pen, haptic, and gaze inputs, and the like. This invention use recognition results from one or more of these modes to provide compensation to the recognition process of one or more other ones of these modes. In various exemplary embodiments, a multimodal recognition system inputs one or more recognition lattices from one or more of these modes, and generates one or more models to be used by one or more mode recognizers to recognize the one or more other modes. In one exemplary embodiment, a gesture recognizer inputs a gesture input and outputs a gesture recognition lattice to a multimodal parser. The multimodal parser generates a language model and outputs it to an automatic speech recognition system, which uses the received language model to recognize the speech input that corresponds to the recognized gesture input.

Abstract: The present invention includes a method of deciding whether a data set is acceptable for making a decision. A first probability partition array and a second probability partition array may be provided. A no-match zone may be established and used to calculate a false-acceptance-rate (“FAR”) and/or a false-rejection-rate (“FRR”) for the data set. The FAR and/or the FAR may be compared to desired rates. Based on the comparison, the data set may be either accepted or rejected. The invention may also be embodied as a computer readable memory device for executing the methods.

Abstract: The present invention provides an apparatus and a method for de-interlacing. The apparatus includes an edge detection module, a statistics module, and an interpolation circuit. The edge detection module performs an edge detection operation on a plurality of pixels of an image so as to generate edge information corresponding to the image. The statistics module performs a detection window based statistics operation on the edge information so as to generate statistics information corresponding to the image. The interpolation circuit interpolates the image according to the statistics information so as to generate an intra-field interpolation signal corresponding to the image.

Abstract: A method and system for monitoring the performance of a character recognition system is disclosed. According to a preferred embodiment, the method comprises utilizing an average confidence score for a plurality of characters for ongoing performance monitoring of the character recognition system, wherein a confidence score indicates a level of confidence that a character is accurately recognized.

Abstract: A multi-stage method is provided for automatically characterizing data sets containing data points which are each defined by measurements of three variables as either random or non-random. A three-dimensional Cartesian volume which is sized to contain all of a total number N of data points in the data set which is to be characterized. The Cartesian volume is partitioned into equal sized cubes, wherein each cube may or may not contain a data point. A predetermined route is defined that goes through every cube one time and scores each cube as a one or a zero thereby producing a stream of ones and zeros. The number of runs is counted and utilized to provide a Runs Test which predicts if the N data points in any data set are random or nonrandom. Additional tests are used in conjunction with the Runs Test to increase the accuracy of characterization of each data set as random or nonrandom.

Abstract: Image alias rejection when converting a high resolution rasterized waveform to a lower resolution rasterized waveform for display uses a statistical filter. The statistical filter provides a shaped probability density function either by combining the outputs of multiple random number generators, such as linear feedback shift registers, or by using a corresponding look-up table to produce a dither signal. The statistical filter may be applied to one or both of the dimensional values for each data point of the high resolution rasterized waveform by combining the dimensional values with the dither signal. The resulting filtered dimensional values may then be subsampled, such as by truncation, to produce values for a lower resolution rasterized waveform display.

Abstract: A face portion of an input image, an example of a predetermined structure, is applied to a mathematical model by the image recovery/addition section to recover a missing element of the face portion in the input image. The mathematical model is generated by a predetermined statistical method, such as the AAM scheme or the like, based on a plurality of sample images representing the face portion including the recovery target element. Thereafter, the face portion is reconstructed to include the missing element based on the parameter corresponding to the face portion obtained by applying the face portion to the model, and the face portion of the input image is replaced by the reconstructed face portion to produce a restored image by the image reconstruction section.

Abstract: Techniques are provided to classify patterns in isogenous pattern sources. Techniques are provided to determine a computationally inexpensive upperbound on the true score or joint probability of the field label and field features over all field labels. Candidate field labels associated with promising upperbound scores are dynamically queued. True scores are computed for a subset of the candidates fields resulting in reduced computations to determine a field label. Techniques are also provided to determine optimal variables for any system with shared constraints.

Abstract: A pattern recognition procedure forms a hierarchical statistical model using a hidden Markov model and a coupled hidden Markov model. The hierarchical statistical model supports a pa 20 layer having multiple supernodes and a child layer having multiple nodes associated with each supernode of the parent layer. After training, the hierarchical statistical model uses observation vectors extracted from a data set to find a substantially optimal state sequence segmentation.

Abstract: The invention performs handwriting recognition using mixtures of Bayesian networks. A mixture of Bayesian networks (MBN) consists of plural hypothesis-specific Bayesian networks (HSBNs) having possibly hidden and observed variables. A common external hidden variable is associated with the MBN, but is not included in any of the HSBNs. Each HSBN models the world under the hypothesis that the common external hidden variable is in a corresponding one of its states. The MBNs encode the probabilities of observing the sets of visual observations corresponding to a handwritten character. Each of the HSBNs encodes the probabilities of observing the sets of visual observations corresponding to a handwritten character and given a hidden common variable being in a particular state.

Abstract: A system or method (collectively “selection system”) is disclosed for selecting attributes for a classifier in a sensor system. The selection system selects attribute types using statistical distributions of the attribute values associated with those attribute types. Attribute types not within the selectively identified subset of attribute types can be filtered out before such data is sent to a classifier. The system can use a test data subsystem for storing and accessing actual sensor data. A distribution analysis subsystem can perform statistical analyses on the test data to identify underlying distributions, and to compare individual attribute types to such distributions. An attribute selection subsystem, wherein said attribute selection subsystem selectively identifies a subset of attribute types from said subset of attribute types.

Abstract: A method and apparatus are disclosed for generating and maintaining enhanced background models for use in background-foreground segmentation. Background models are modified to contain an indication of objects that are typically stationary. Thereafter, if an object moves and has been previously identified as an object that is typically stationary, the object will not unnecessarily be identified as part of the foreground during background-foreground segmentation. In an exemplary implementation, moving objects are classified into two sets. A first set includes objects that typically move independently and a second set includes objects that are typically stationary. Generally, once an object is assigned to the second (stationary object) set, the object will remain in the background, even if the object is moved (normally, movement of the object would cause the object to become part of the foreground).

Abstract: The present invention involves a new system and method for probabilistic exemplar-based tracking of patterns or objects. Tracking is accomplished by first extracting a set of exemplars from training data. The exemplars are then clustered using conventional statistical techniques. Such clustering techniques include k-medoids clustering which is based on a distance function for determining the distance or similarity between the exemplars. A dimensionality for each exemplar cluster is then estimated and used for generating a probabilistic likelihood function for each exemplar cluster. Any of a number of conventional tracking algorithms is then used in combination with the exemplars and the probabilistic likelihood functions for tracking patterns or objects in a sequence of images, or in a space, or frequency domain.

Abstract: The present invention recites a method and computer program product for classifying an input pattern into an associated output class. For each of a plurality of classification rungs, each rung representing at least one output class, the a priori probability that an input pattern will be associated with a class represented by that rung is determined. A sequential order of processing for the plurality of classification rungs is assigned, based upon the a priori probability associated with each classification rung. A confidence value, associated with a represented class, is computed for each classification rung, in the assigned order, until a class is selected or a termination event occurs. The associated confidence values are compared to a predetermined threshold, and a class associated with a confidence value that first exceeds the threshold is selected, if a confidence value exceeds the threshold.

Abstract: A pattern recognition method and apparatus decrease the amount of computation for pattern recognition and adapts flexibly to an increase and a change in learning samples. Learning is made beforehand on base vectors in a subspace of each category and a kernel function. Pattern data to be recognized is input, and projection of an input pattern to a nonlinear subspace of each category is decided. Based on the decided projection, a Euclidean distance or an evaluation value related to each category is calculated from the property of the kernel function, and is compared with a threshold value. If a category for which the evaluation value is below the threshold value exists, a category for which the evaluation value is the smallest is output as a recognition result. If there is no category for which the evaluation value is below the threshold value, a teaching signal is input for additional learning.

Abstract: A handheld device 100 with a graphical user interface for entering handwritten text 102. The handheld device includes word and character input areas 104, 106 within a designated input area 108. Icons 110, 112, 114, 116 and 118 are disposed at the right side of the handwriting user interface 102. A scroll bar 120 may be disposed at the right side of the interface display 102. An entry that begins in the word input area 104 is treated as a handwritten word. A handwritten entry that begins in the character input area 106 is treated as a single character and may be one character in character string. Handwritten character entries are each matched against all potential characters.

Abstract: Automatic detection and tracking of multiple individuals includes receiving a frame of video and/or audio content and identifying a candidate area for a new face region in the frame. One or more hierarchical verification levels are used to verify whether a human face is in the candidate area, and an indication made that the candidate area includes a face if the one or more hierarchical verification levels verify that a human face is in the candidate area. A plurality of audio and/or video cues are used to track each verified face in the video content from frame to frame.

Abstract: A method and apparatus for extracting information from symbolically compressed document images. A deciphering module generates first and second text strings by deciphering respective sequences of template identifiers in first and second symbolically compressed document images. A conditional n-gram module receives the first and second text strings from the deciphering module and extracts n-gram terms therefrom based on a predicate condition. A comparison module generates a measure of similarity between the first and second symbolically compressed document images based on the n-gram terms extracted by the conditional n-gram module.

Abstract: The picture region extraction method coarse-grains this picture data space, calculates a coarse-grained empirical probability distribution, initializes parameters, calculates a coarse-grained conditional probability distribution and a class membership probability, updates the parameters, and calculates an evaluation function, each process being repeated until there is no change in the evaluation function, at which point, a picture region is extracted based on the class membership probability.

Abstract: The present invention recites a method and computer program product for generating a set of training samples from a single ideal pattern for each output class of a pattern recognition classifier. A system equivalent pattern is generated for each of a plurality of classes from a corresponding ideal pattern. A noise model, simulating at least one type of noise expected in a real-world classifier input pattern, is then applied to each system equivalent pattern a set number times to produce, for each output class, a number of training samples. Each training sample simulates defects expected in real-world classifier input patterns.

Abstract: The present invention embodies a character recognition method for constructing a result string from a plurality of result sets. Each result set comprises at least one candidate character, and each candidate character has an associated confidence indication. The method can begin by selecting a plurality of character types. For each selected character type, a candidate string can be created by concatenating a candidate character of the selected character type from each result set. The associated confidence indication for each concatenated candidate character can be combined to form a corresponding combined confidence indication for each created candidate string. The created candidate string with the most favorable corresponding combined confidence indication can be selected as the result string.

Abstract: A data classifier performs a data classification process using prototypes classified into clusters. A prototype map is formed using mapping means and clustering means. The mapping means forms, through learning, a prototype map by adjusting coupling weights between a plurality of prototypes provided in a map space based on a plurality of input data. The clustering means calculates a predetermined measure between the prototypes and classifies the prototypes into a plurality of clusters based on the measure.

Abstract: The invention relates to a method for recording and displaying fluorescence images with high spatial resolution. Several recordations of an image of an object are made sequentially with an electronic camera and, between recordations, the image of the object and the electronic camera are shifted relative to each other. The recordations are subsequently unified to a compiled image of higher spatial resolution. Before or after the partial images are brought together, the signals, which belong to image points of the images and which are obtained at different times, are amplified with amplification factors which are dependent upon the time intervals between the image recordations and are essentially reciprocal to the time dependency of the fluorescence. Artefacts caused by decreasing fluorescence intensities (fading, bleaching) are thereby avoided. These fluorescence intensities decrease as a function of time.

Abstract: A picture quality diagnostics apparatus and method generates a human vision model response based on a human vision model for a test input video signal. Also objective measure maps for different impairment types are generated from the test input video signal. The objective measure maps are applied as masks to the human vision model response to produce objectively filtered subjective maps. The objectively filtered subjective maps are analyzed to give the respective proportions of the different objective impairment types contributing to perceptual impairment or difference for the test input video signal.

Abstract: A method for processing a digital color image includes the steps of: providing a subject matter detector for distinguishing between target and background subject matters; applying the subject matter detector to the image to produce a belief map indicating the degree of belief that pixels in the image belong to target subject matter; providing an image enhancement operation that is responsive to a control signal for controlling the degree of image enhancement; and applying image enhancement to the digital image by varying the control signal according to the belief map to produce an enhanced image.

Abstract: To inspect a status of an inspection object by using an inspection apparatus for extracting amount of characteristic to an inputted waveform signal and determining a status on the basis of the extracted amount of characteristic. Specifically, the inspection apparatus uses a normal knowledge that is generated on the basis of only the data of a normal status at an initial stage to determine whether or not the status of the inspection object complies with the normal status. The inspection apparatus generates an abnormal kind knowledge by abnormal kind on the basis of the data of an abnormal status that are collected in accordance with repeat of the determination, and then, determines the status by using the normal knowledge and the abnormal kind knowledge.

Abstract: A method for embedding digital watermark data in digital data contents includes the steps of obtaining a frequency coefficient of block data of digital data contents, obtaining a complexity of the block data, obtaining an amount of transformation of the frequency coefficient from the complexity and the digital watermark data, and embedding the digital watermark data by transforming the frequency coefficient. In addition, a method for reading digital watermark data includes the steps of calculating a probability of reading ‘1’ or ‘0’ in a read bit sequence by using a test method on the basis of binary distribution, determining the presence or absence of digital watermark data according to the probability, and reconstituting digital watermark data. Another method includes the steps of performing soft decision in code theory by assigning weights to the digital watermark sequence with a weighting function, and reconstituting digital watermark data.

Abstract: Finite-state systems and methods allow multiple input streams to be parsed and integrated by a single finite-state device. These systems and methods not only address multimodal recognition, but are also able to encode semantics and syntax into a single finite-state device. The finite-state device provides models for recognizing multimodal inputs, such as speech and gesture, and composes the meaning content from the various input streams into a single semantic representation. Compared to conventional multimodal recognition systems, finite-state systems and methods allow for compensation among the various input streams. Finite-state systems and methods allow one input stream to dynamically alter a recognition model used for another input stream, and can reduce the computational complexity of multidimensional multimodal parsing.

Abstract: Disclosed is a method for detecting subject matter regions in a digital color image having pixels of (red, green, blue) values, comprising the steps of: assigning to each pixel a belief value as belonging to the subject matter region based on color and texture features; forming spatially contiguous candidate subject matter regions by thresholding the belief values; analyzing the spatially contiguous regions based on one or more unique characteristics of the subject matter to determine the probability that a region belongs to the subject matter; and generating a map of detected subject matter regions and associated probability that the regions belongs to the subject matter.

Abstract: In a method of improving a characteristic of an image according to its material content, where the image is comprised of image pixels, a belief map corresponding spatially to the image pixels is generated. The belief map includes belief values indicating the likelihood that respective pixels are representative of a particular material, such as flesh. An improvement parameter is generated from the belief map, and the improvement parameter is applied uniformly to the image pixels to improve the characteristic, such as the sharpness, of the image.

Abstract: A work identification system comprises a work storage configured to store digital data representing at least one of a shape, area, and color of an only one work, a collation section configured to calculate a degree of deviation between digital data representing at least one of a shape, area, and color of a target work to be identified and the digital data stored in the work storage, and a test section configured to perform a test of hypothesis based on a predetermined hypothesis using the degree of deviation.

Abstract: A method for increasing efficiency of interaction by an operator with data on a computer display includes presenting the data to the operator on the computer display, and providing multiple instances of an on-screen control at different locations on the display for selection by the operator using a pointing device linked to the display. The control is actuated responsive to the selection by the operator of any of the instances of the control on the display.

Abstract: A method for classification of image regions by probabilistic merging of a class probability map and a cluster probability map includes the steps of a) extracting one or more features from an input image composed of image pixels; b) performing unsupervised learning based on the extracted features to obtain a cluster probability map of the image pixels; c) performing supervised learning based on the extracted features to obtain a class probability map of the image pixels; and d) combining the cluster probability map from unsupervised learning and the class probability map from supervised learning to generate a modified class probability map to determine the semantic class of the image regions. In one embodiment the extracted features include color and textual features.

Abstract: The present invention involves a new system and method for probabilistic exemplar-based tracking of patterns or objects. Tracking is accomplished by first extracting a set of exemplars from training data. The exemplars are then clustered using conventional statistical techniques. Such clustering techniques include k-medoids clustering which is based on a distance function for determining the distance or similarity between the exemplars. A dimensionality for each exemplar cluster is then estimated and used for generating a probabilistic likelihood function for each exemplar cluster. Any of a number of conventional tracking algorithms is then used in combination with the exemplars and the probabilistic likelihood functions for tracking patterns or objects in a sequence of images, or in a space, or frequency domain.

Abstract: A method is disclosed for classifying an input pattern into an associated class through use of a compound classifier. Data pertaining to preselected features present within the input pattern are extracted. A discriminant value for each of a plurality of classes is then determined via a first classification technique. This value reflects the relative likelihood that a class is the associated class. The class with the highest relative likelihood is selected. A confidence value is generated via a second classification technique. This confidence value is reflective of the a posteriori probability that the selected class is the associated class. The selected class is rejected if the determined confidence value is below a predetermined threshold value.

Abstract: Connected components of dark pixels are clustered from across the image. A “most likely” representative image for each cluster of images is determined, with likelihood determined by a probabilistic model of the image capturing process. An a priori (prior) probability distributions on bitmaps may be used to determine the most likely representative images. For example, a priori probability distributions based on so-called chain codes are implemented. The representative images are used to cluster connected components. Clustering may be repeated. The output page is assembled by replacing each member of a cluster of images by that cluster's representative image.

Abstract: The present invention is directed to a method of discriminating between textual content and graphical content. The method includes the steps of receiving a plurality of pixel values for a pixel line segment, calculating a plurality of spatial gradients based on the pixel values of adjacent pixels, determining a smoothness index by processing the plurality of spatial gradients, and identifying the pixel line segment as text or graphics by comparing the smoothness index to a threshold value.

Abstract: The present invention relates to a method for visually detecting and tracking an object through a space. The method chooses modules for a restricting a search function within the space to regions with a high probability of significant change, the search function operating on images supplied by a camera. The method also derives statistical models for errors, including quantifying an indexing step performed by an indexing module, and tuning system parameters. Further the method applies a likelihood model for candidate hypothesis evaluation and object parameters estimation for locating the object.

Abstract: A set of difference vectors is generated by calculating the difference between the feature vector of each pattern in a specific pattern set and the average feature vector of each correct category. When a feature vector of an unknown pattern is inputted, the expected value of the probability density function of a specific category is obtained using an error distribution corresponding to the difference vector set as the probability density function. Then, the discriminant function value for the category is defined based on the obtained expected value and the pattern can be recognized.

Abstract: The invention performs handwriting recognition using mixtures of Bayesian networks. A mixture of Bayesian networks (MBN) consists of plural hypothesis-specific Bayesian networks (HSBNs) having possibly hidden and observed variables. A common external hidden variable is associated with the MBN, but is not included in any of the HSBNs. Each HSBN models the world under the hypothesis that the common external hidden variable is in a corresponding one of its states. The MBNs encode the probabilities of observing the sets of visual observations corresponding to a handwritten character. Each of the HSBNs encodes the probabilities of observing the sets of visual observations corresponding to a handwritten character and given a hidden common variable being in a particular state.

Abstract: Classification of targets in a digital image is accomplished by evaluating windowed portions of the image. A weighted sum is formed for each overlapping windowed portion of an image using a feature set corresponding thereto, each window's feature set and weighted sum is normalized, and a context matrix is defined for each window. A normalized score is defined for each window. A threshold criteria is compared to a maximum score that is based on the context matrix and the normalized score associated with each window. Each window having its maximum score satisfy the threshold criteria is classified as a possible target window and assigned to a group based on location of the possible target window in the image and its maximum score. A group score is assigned to each group. Each group having its corresponding group score satisfying a group threshold criteria is classified as a target.

Abstract: Detection and classification of targets in a digital image is accomplished by evaluating windowed portions of the image. A feature set is generated for each of a plurality of overlapping windowed portions, a weighted sum is formed for each portion based upon its feature set, and a context matrix is defined for each window. A score is formed from each context matrix and is normalized for each window. A threshold criteria is compared to a maximum score for each window. Each window having its maximum score satisfy the threshold criteria is classified as a possible target window and is assigned to a group based on location of the possible target window and its maximum score. A group score is formed for each group and compared to a group threshold criteria. Each group having its corresponding group score satisfying the group threshold criteria is classified as a target.

Abstract: The present invention is embodied in a system and method for statistically comparing a first set of digital data to at least a second set of digital data and matching the first set of digital data to appropriately corresponding portions of the second set of digital data. The first or the second set of digital data can be transformed during statistical analysis to enhance statistical analysis of the digital data.

Abstract: A method and system of object classification uses measurements for a training set of objects to classify an unknown object as being a member in one of several classes of the training set. The classes are defined by features of the training set, the objects of which have known class memberships. The method comprises performing principal component analysis on the training set measurements to discard features that have negligible information regarding class membership, conducting projection pursuit on the remaining training set measurements to accentuate differences between the classes, estimating a distribution of each accentuated class in the training set, and evaluating measurements of the unknown object to determine the membership of the unknown object in one of the accentuated classes. The system implements the method in a computer program stored in computer memory.

Abstract: A method is provided for the detection and classification of targets in a digital image of a structure having known characteristics. In general, windowed portions of the image are evaluated in context with the entire image and in terms of their location in the image. More specifically, a scoring scheme is used to identify relevant windows with the relevance of each window being evaluated in terms of location in the image and the known characteristics of the structure being imaged. Relevant windows satisfying a threshold criteria are grouped based on their relative location in the image. A group scoring scheme is applied to each group to identify and classify targets.