Abstract: A classification scheme assigns samples of the watermarked media to classes based on classification criteria indicating a likely presence of a watermarked signal. Once classified, the scheme determines statistical characteristics of the media signal for the classes, and assigns a figure of merit to the samples in each class. A watermark decoder (e.g., reader and/or detector) use the figure of merit to adapt a watermark decoding function to the statistical features of the classes, such as weighting to samples to improve watermark decoding.

Abstract: A method for improving scene classification of a sequence of digital images is disclosed herein. Such a method may include providing a sequence of images captured in temporal succession; (b) classifying each of the images individually based on information contained in the image alone to generate a first image classification; and (c) imposing a pre-determined temporal context model on the sequence of images to generate a final image classification for each image in the sequence.

Abstract: An image capturing apparatus is provided that is capable of performing both object detection using image recognition and object detection using movement detection on successively captured images. In the image capturing apparatus, the reliability of the result of the object detection using image recognition is evaluated based on the previous detection results. If it is determined that the reliability is high, execution of the object detection using movement detection is determined. If it is determined that the reliability is low, non-execution of the object detection using movement detection is determined. With this configuration, the object region can be tracked appropriately.

Abstract: Methods of statistical learning for Bayesian inference in the context of efficient optimization schemes for image restoration are presented. Second and third order priors that may be learned while maintaining graph representability are identified. A framework to learn and impose prior knowledge on the distribution of pairs and triplets of labels via graph cuts is presented. The disclosed methods optimally restore binary textures from very noisy images with runtimes in the order of seconds while imposing hundreds of statistically learned constraints per node.

Abstract: A method and an apparatus automatically detect eyeglasses in an image. The method according to one embodiment accesses digital image data representing an image including a face; detects eyeglasses in the image by using nose ridge masking; and outputs a decision about presence or absence of eyeglasses in the image.

Abstract: A method and system for generating a Directed Acyclic Graph (DAG) from an initial multi-chain, subject to a constraint. The initial multi-chain is expressed as a string serving as a current input string to which the constraint is subsequently applied. A provided string P expresses the constraint. P is applied to the current input string to generate at least one output string, wherein each generated output string violates the constraint to a lesser extent than does the input string or does not violate the constraint. Each generated output string violating the constraint serves as a current input string to which the constraint is subsequently applied. P is recursively applied to each current input string that had been determined from applying P previously, until applying P does not generate any more output strings violating the constraint. A set of the generated output strings not violating the constraint represents the DAG.

Abstract: Systems, methods, and computer program products, implementing techniques for detecting objects using a soft cascade. The techniques include receiving a digital data segment and determining whether the digital data segment resembles an object of interest by passing the digital data segment through a cascade. The cascade includes an ordered sequence of stages and a rejection function after each stage that determines whether to reject the digital data segment at the current stage as not resembling the object of interest, or to allow the digital data segment to pass to the next stage of evaluation. The rejection function allows the digital data segment to fail the current stage and still pass to the next stage.

Abstract: Images are classified as photos (e.g., natural photographs) or graphics (e.g., cartoons, synthetically generated images), such that when searched (online) with a filter, an image database returns images corresponding to the filter criteria (e.g., either photos or graphics will be returned). A set of image statistics pertaining to various visual cues (e.g., color, texture, shape) are identified in classifying the images. These image statistics, combined with pre-tagged image metadata defining an image as either a graphic or a photo, may be used to train a boosting decision tree. The trained boosting decision tree may be used to classify additional images as graphics or photos based on image statistics determined for the additional images.

Abstract: A system includes at least one sensor device configured to transmit a first detection signal over a first spatial region and a second detection signal over a second spatial region. The second region has a first sub-region in common with the first region. The system further includes a processing device configured to assign a first occupancy value to a first cell in an evidence grid. The first cell represents the first sub-region, and the first occupancy value characterizes whether an object has been detected by the first detection signal as being present in the first sub-region. The processing device is further configured to calculate, based on the first and second detection signals, the probability that the first occupancy value accurately characterizes the presence of the object in the first sub-region, and generate a data representation of the first sub-region based on the probability calculation.

Abstract: A face feature is extracted by a face feature extracting unit, a confidence index is extracted by a confidence index extracting unit, and then, they are outputted as a face meta-data. At a time of matching, by using the confidence index of the face meta-data, a distribution estimating unit estimates the data of a parameter and the like with regard to a posterior distribution when the confidence index is obtained. A distance calculating unit calculates a similarity between feature values. Consequently, the precision of face verification can be improved, thereby carrying out a practical face matching.

Abstract: A method and system for video-based encroachment detection are provided, the method including receiving first and second images, modeling a background from the first image, subtracting the background from the second image to provide a detection map, calibrating the size of an object from the pixel level, integrating a projection of the object with the detection map using dynamic programming, and detecting the object in a region if the projection matches that region of the detection map; and the system including a processor, a background modeling unit coupled with the processor for modeling a background from the first image and subtracting the background from the second image to provide a detection map, and a dynamic programming unit coupled with the processor for calibrating the size of an object from the pixel level, integrating a projection of the object with the detection map, and detecting the object in a region if the projection matches that region of the detection map.

Abstract: The present invention provides a technique for automated selection of a parameterized operator sequence to achieve a pattern classification task. A collection of labeled data patterns is input and statistical descriptions of the inputted labeled data patterns are then derived. Classifier performance for each of a plurality of candidate operator/parameter sequences is determined. The optimal classifier performance among the candidate classifier performances is then identified. Performance metric information, including, for example, the selected operator sequence/parameter combination, will be outputted. The operator sequences selected can be chosen from a default set of operators, or may be a user-defined set. The operator sequences may include any morphological operators, such as, erosion, dilation, closing, opening, close-open, and open-close.

Abstract: An apparatus and method are disclosed for context dependent cropping of a source image. The method includes identifying a context for the source image, identifying a visual class corresponding to the identified context from a set of visual classes, applying a class model to the source image to identify a candidate region of the image based on its relevance to the visual class, and identifying a subpart of the source image for cropping, based on the location of the candidate region.

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 of identifying at least one particular person of interest in a collection of digital images taken over time, includes providing the collection of digital images with each digital image containing one or more persons; storing in a digital database an appearance model having a set of features associated with the particular person of interest and a prior probability of the person of interest appearing in images by a particular photographer; and comparing the appearance model with features extracted from the digital images in the collection of digital images and using the prior probability and the appearance model to determine if the particular person of interest is in one or more digital images in the collection of digital images.

Abstract: A method of automatically establishing the correct orientation of an image using facial information. This method is based on the exploitation of the inherent property of image recognition algorithms in general and face detection in particular, where the recognition is based on criteria that is highly orientation sensitive. By applying a detection algorithm to images in various orientations, or alternatively by rotating the classifiers, and comparing the number of successful faces that are detected in each orientation, one may conclude as to the most likely correct orientation. Such method can be implemented as an automated method or a semi automatic method to guide users in viewing, capturing or printing of images.

Abstract: In one aspect, the invention is based on a process that combines information present in a joint distribution of the predictor variables and the variable (or variables) to be predicted. This information may be captured in the form of a table or other like data structure that includes a set of vectors (referred to as a “TAB”). The process uses the information in the TAB in conjunction with one or more rules. In one embodiment, a set of different rules are applied to the TAB to determine which rule in the set produces the most accurate predictions. The RULE that produces the most accurate predictions is then used in conjunction with observed information to make predictions.

Abstract: The invention relates to a method for analyzing image processing procedures, in which unprocessed original image data is stored, stored original image data is retrieved, retrieved original image data is processed, the individual processing steps during the processing of the image data are in some instances stored together with the respective processing step parameter values, and the processed image data is stored such that it can be assigned to the processing steps stored during its processing and in some instances processing step parameter values. According to the invention the processed image data is analyzed statistically and the result of the statistical analysis is stored such that it can be assigned to the stored processing steps and in some instances processing step parameter values. The original image data is optionally included in the statistical analysis of the processed image data.

Abstract: A method for performing image recognition is disclosed. The method includes obtaining a collection of pixels and grouping at least some of the pixels into a set of cluster features based on gradient magnitude. For each cluster feature in the set, statistical variables are generated. The statistical variables represent a collective property of the pixels in the cluster feature. The statistical variables are utilized as a basis for comparing the collection of pixels to a different collection of pixels.

Abstract: Dynamic inference is leveraged to provide online sequence data labeling. This provides real-time alternatives to current methods of inference for sequence data. Instances estimate an amount of uncertainty in a prediction of labels of sequence data and then dynamically predict a label when an uncertainty in the prediction is deemed acceptable. The techniques utilized to determine when the label can be generated are tunable and can be personalized for a given user and/or a system. Employed decoding techniques can be dynamically adjusted to tradeoff system resources for accuracy. This allows for fine tuning of a system based on available system resources. Instances also allow for online inference because the inference does not require knowledge of a complete set of sequence data.

Abstract: The present invention comprises using error propagation for building feature spaces with variable uncertainty and using variable-bandwidth mean shift for the analysis of such spaces, to provide peak detection and space partitioning. The invention applies these techniques to construct and analyze Hough spaces for line and geometrical shape detection, as well as to detect objects that are represented by peaks in the Hough space. This invention can be further used for background modeling by taking into account the uncertainty of the transformed image color and uncertainty of the motion flow. Furthermore, the invention can be used to segment video data in invariant spaces, by propagating the uncertainty from the original space and using the variable-bandwidth mean shift to detect peaks. The invention can be used in a variety of applications such as medical, surveillance, monitoring, automotive, augmented reality, and inspection.

Abstract: A method and apparatus for comparing data is described. In one embodiment, an exemplary method includes receiving a first set of data pertaining a first object and a second set of data pertaining to a second object, and comparing the first object with the second object using an earth mover's distance method that is based on computation of a series of Hausdorff distances.

Abstract: A system and process for automatically learning a reliable color-based tracking system is presented. The tracking system is learned by using information produced by an initial object model in combination with an initial tracking function to probabilistically determine the configuration of one or more target objects in a temporal sequence of images, and a data acquisition function for gathering observations relating to color in each image. The observations gathered by the data acquisition function include information that is relevant to parameters desired for a final color-based object model. A learning function then uses probabilistic methods to determine conditional probabilistic relationships between the observations and probabilistic target configuration information to learn a color-based object model automatically tailored to specific target objects.

Abstract: A method and system efficiently and accurately detects humans in a test image and classifies their pose. In a training stage, a probabilistic model is derived in an unsupervised or semi-supervised manner such that at least some poses are not manually labeled. The model provides two sets of model parameters to describe the statistics of images containing humans and images of background scenes. In a testing stage, the probabilistic model is used to determine if a human is present in the image, and classify the human's pose based on the poses in the training images. A solution is efficiently provided to both human detection and pose classification by using the same probabilistic model to solve the problems.

Abstract: A method for classifying features in a digital medical image includes providing a plurality of feature points in an N-dimensional space, wherein each feature point is a member of one of two sets, determining a classifying plane that separates feature points in a first of the two sets from feature points in a second of the two sets, transforming the classifying plane wherein a normal vector to said transformed classifying plane has positive coefficients and a feature domain for one or more feature points of one set is a unit hypercube in a transformed space having n axes, obtaining an upper bound along each of the n-axes of the unit hypercube, inversely transforming said upper bound to obtain a new rule containing one or more feature points of said one set, and removing the feature points contained by said new rule from said one set.

Abstract: A method for analyzing in real-time in a paper machine or board machine the surface structure of a web (1) of paper or board which method comprises the direction of an imaging system towards a pre-determined area (3) of the web (1), the arrangement of an illumination system to illuminate the region from a pre-determined direction with obliquely incident light, and the arrangement of an image analysis system in association with the imaging system.

Abstract: A method is for segmentation of section image data of an examination object, in which a target structure is determined in the section image data and an anatomical norm model is selected whose geometry can be varied on the basis of model parameters. The model parameters are organized hierarchically on the basis of their influence on the anatomical overall geometry of the model. The norm model is matched to the target structure for individualization purposes in a number of iteration steps, with the number of model parameters which can be set being increased in accordance with the hierarchical organization as the number of iteration steps increases. Finally, all of those pixels within the section image data are selected which lie within a contour of the individualized model or of model part, or which differ from this by at most a specific difference value. A corresponding image processing system is also described.

Abstract: A discriminative machine learning system for labels text and non-text strokes in digital ink. The learning system considers stroke features and the context of the strokes, such as temporal information about one or more strokes, in a probabilistic framework. The learning system can also consider gap features within the probabilistic framework to label associated strokes.

Abstract: In an image processing apparatus which processes time-series images picked up by an imaging device in time series, a motion-vector calculating unit calculates a motion vector between plural images constituting the time-series images with respect to plural pixel regions set in the image. A newly-appearing-rate estimating unit estimates a newly-appearing rate which is a rate of a region newly appears in an image between plural images based on the motion vector. A display-time determination coefficient calculating unit calculates a display time of an image according to the newly-appearing rate.

Abstract: A range map of a visual scene generated by a stereo vision and associate image processing system, and is filtered to remove objects beyond a region of interest and for which a collision is not possible, and to remove an associated road surface. Objects clustered in range bins are separated by segmentation. A composite range map is generated using principle components analysis and processed with a connected components sieve filter. Objects are identified using one or more of a harmonic profile and other features using an object recognition processor using a combination of inclusive, exclusive and harmonic networks to generate a classification metric.

Abstract: In a method and apparatus for incremental calculation of a general linear model given intermittent correlation of the model functions, data composed of a number of data sets with a number of different random samples are processed. For each independent random sample contained in the data set, a dependency on an order quantity is compared with the dependency on the order quantity in model functions contained in a model matrix G using the general linear model in order to check the occurrence of specific characteristics in the dependency of the random sample on the order quantity. The calculation of the general linear model ensues incrementally from data set to data set. Before the calculations for the data of a data set, a check is made as to whether the model functions contained in the model matrix G exhibit orthogonal portions to a sufficient degree in a segment (represented by the data set) of the dependency on the order quantity for the calculations.

Abstract: A memory footprint of an Modified Quadratic Discriminant Function (MQDF) pattern recognition classifier is reduced without resulting in unacceptable classification accuracy degradation. Covariance matrices for multiple classes are clustered into a smaller number of matrices where different classes share the same set of eigenvectors. According to another approach, different numbers of principal components are stored for different classes based on criteria such as class usage frequency, larger variation in writing, and the like, resulting in fewer principal components to be stored in memory.

Abstract: A probabilistic boosting tree framework for computing two-class and multi-class discriminative models is disclosed. In the learning stage, the probabilistic boosting tree (PBT) automatically constructs a tree in which each node combines a number of weak classifiers (e.g., evidence, knowledge) into a strong classifier or conditional posterior probability. The PBT approaches the target posterior distribution by data augmentation (e.g., tree expansion) through a divide-and-conquer strategy. In the testing stage, the conditional probability is computed at each tree node based on the learned classifier which guides the probability propagation in its sub-trees. The top node of the tree therefore outputs the overall posterior probability by integrating the probabilities gathered from its sub-trees. In the training stage, a tree is recursively constructed in which each tree node is a strong classifier. The input training set is divided into two new sets, left and right ones, according to the learned classifier.

Abstract: A digital pictorial book system for searching for and explaining a name of an object captured by an image capturing module includes an image capturing module for capturing an image; a main object selecting module for selecting a main object out of the image; a feature extracting module for extracting a feature of the main object; a candidate name searching module for searching a candidate name, which is a candidate for the name of the main object, by using the feature extracted by said feature extracting module in each of a plurality of image databases which store names of objects corresponded to a plurality kinds of different features of the objects; a probability acquiring module for acquiring an index value of a probability of the main object's having the candidate name for each combination of the image database and the candidate name; and a name determining module for determining a most certain name of the main object on the basis of the index value of the probability of the candidate name in each of the

Abstract: The present invention comprises using error propagation for building feature spaces with variable uncertainty and using variable-bandwidth mean shift for the analysis of such spaces, to provide peak detection and space partitioning. The invention applies these techniques to construct and analyze Hough spaces for line and geometrical shape detection, as well as to detect objects that are represented by peaks in the Hough space. This invention can be further used for background modeling by taking into account the uncertainty of the transformed image color and uncertainty of the motion flow. Furthermore, the invention can be used to segment video data in invariant spaces, by propagating the uncertainty from the original space and using the variable-bandwidth mean shift to detect peaks. The invention can be used in a variety of applications such as medical, surveillance, monitoring, automotive, augmented reality, and inspection.

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: The present invention provides a technique for automated selection of a parameterized operator sequence to achieve a pattern classification task. A collection of labeled data patterns is input and statistical descriptions of the inputted labeled data patterns are then derived. Classifier performance for each of a plurality of candidate operator/parameter sequences is determined. The optimal classifier performance among the candidate classifier performances is then identified. Performance metric information, including, for example, the selected operator sequence/parameter combination, will be outputted. The operator sequences selected can be chosen from a default set of operators, or may be a user-defined set. The operator sequences may include any morphological operators, such as, erosion, dilation, closing, opening, close-open, and open-close.

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: Searching a database involves creating an access structure including a first tree data structure having a root node and at least one child node. Each child node is associated with match data corresponding to a data value of a field of a database record. Leaf child nodes of the first tree data structure include a link to another tree data structure in the access structure. Leaf child nodes of a further tree data structure include a link to a database record. The tree structures are traversed and scores are computed for the paths traversed that reflect the level of matching between the match pattern data of the nodes in a path and a search request to identify a database record that best matches the request.

Abstract: The present invention provides a statistical modeling approach to automatic linguistic indexing of photographic images. The invention uses categorized images to train a dictionary of hundreds of statistical models each representing a concept. Images of any given concept are regarded as instances of a stochastic process that characterizes the concept. To measure the extent of association between an image and a textual description associated with a predefined concept, the likelihood of the occurrence of the image based on the characterizing stochastic process is computed. A high likelihood indicates a strong association between the textual description and the image. The invention utilizes two-dimensional multi-resolution hidden Markov models that demonstrate accuracy and high potential in linguistic indexing of photographic images.

Abstract: A system and method for image performing classification through generative models of features occurring in an image. Category-conditional probability distributions of features occurring in a plurality of training images are maintained. Each distribution is assigned to a category. The features occurring in an unclassified image are identified. Category-conditional likelihoods for the identified features are determined using the category-conditional probability distributions for each category. The unclassified image is assigned to one such category based upon the category-conditional likelihoods.

Abstract: Systems and methods for object or pattern detection that use a nonlinear support vector (SV) machine are described. In the illustrated and described embodiment, objects or patterns comprising faces are detected. The decision surface is approximated in terms of a reduced set of expansion vectors. In order to determine the presence of a face, the kernelized inner product of the expansion vectors with the input pattern are sequentially evaluated and summed, such that if at any point the pattern can be rejected as not comprising a face, no more expansion vectors are used. The sequential application of the expansion vectors produces a substantial saving in computational time.

Abstract: A human tracking system for tracking a plurality of humans in motion, in a video of the humans in motion, includes a human detection subsystem, and a combined tracker. The human detection subsystem is configured to generate a detection output by detecting the plurality of humans in a part-based representation, in each one of a sequence of static frames in the video. The human detection subsystem is further configured to account for partial occlusion of one or more of the humans in the image. The combined tracker is configured to receive and combine the detection responses generated by the human detection subsystem, and to track the humans in response to the received detection responses and image appearance properties.

Abstract: The invention relates to a method and apparatus for determining a classification boundary between an object, such as a vehicle, and a background, using an object classifier. In an embodiment of the invention, a trained classifier is configured to classify and recognize each a plurality of object images and a plurality of background images. Next, a confidence probability density distribution function is calculated for the vehicle and the background using the determined confidence values for the vehicle images and background images. Once the probability density distribution functions for the vehicle and the background are calculated, the classification boundary between the vehicle and the background is determined using the probability density distribution functions for the vehicle or the background, or both, and a predefined model that is appropriate for the application.

Abstract: In an embodiment, the present invention relates to a method for semantic analysis of digital multimedia. In an embodiment of the invention, low level features are extracted representative of one or more concepts. A discriminative classifier is trained using these low level features. A collective annotation model is built based on the discriminative classifiers. In various embodiments of the invention, the frame work is totally generic and can be applied with any number of low-level features or discriminative classifiers. Further, the analysis makes no domain specific assumptions, and can be applied to activity analysis or other scenarios without modification. The framework admits the inclusion of a broad class of potential functions, hence enabling multi-modal analysis and the fusion of heterogeneous information sources.

Abstract: A method of classifying cells into subpopulations using cell classifying data is described. The method comprises receiving and analyzing image data to identify object areas in the image data to determine, for at least one selected first cell, one or more measurements. A first parameter set is derived from the measurements for the first cell, the first parameter set comprising at least one of said one or more measurements. The first set of cells are classified into subpopulations, and identified to produce first identifying data. Cell classifying data for use in classifying a second set of cells into subpopulations is derived from the first parameter set and the first identifying data. A second set of cells is classified into subpopulations on the basis of one or more measurements taken for cells in the second set of cells, by use of the cell classifying data. The parameter sets of cells may be represented as vectors in an n-dimensional space.

Abstract: A method and apparatus is disclosed for pattern analysis by arranging given data so that highdimensional data can be more effectively analyzed. The method allows arrangements of given data so that patterns can be discovered within the data. By utilizing maps that characterizes the data and the type or the set it belongs to, the method produces many data items from relatively few input data items, thereby making it possible to apply statistical and other conventional data analysis methods. In the method, a set of maps from the data or part of the data is determined. Then, new maps are generated by combining existing maps or applying certain transformations on the maps. Next, the results of applying the maps to the data are examined for patterns. Optionally, certain strong patterns are chosen, idealized, and propagated backwards to find a data reflecting that pattern.

Abstract: Category context models (64) and a universal context model (62) are generated including sums of soft co-occurrences of pairs of visual words in geometric proximity to each other in training images (50) assigned to each category and assigned to all categories, respectively. Context information (76) about an image to be classified (70) are generated including sums of soft co-occurrences of pairs of visual words in geometric proximity to each other in the image to be classified. For each category (82), a comparison is made of (i) closeness of the context information about the image to be classified with the corresponding category context model and (ii) closeness of the context information about the image to be classified with the universal context model. An image category (92) is assigned to the image to be classified being based on the comparisons.

Abstract: Expression profiling using DNA microarrays is an important new method for analyzing cellular physiology. In “spotted” microarrays, fluorescently labeled cDNA from experimental and control cells is hybridized to arrayed target DNA and the arrays imaged at two or more wavelengths. Statistical analysis is performed on microarray images and show that non-additive background, high intensity fluctuations across spots, and fabrication artifacts interfere with the accurate determination of intensity information. The probability density distributions generated by pixel-by-pixel analysis of images can be used to measure the precision with which spot intensities are determined. Simple weighting schemes based on these probability distributions are effective in improving significantly the quality of microarray data as it accumulates in a multi-experiment database. Error estimates from image-based metrics should be one component in an explicitly probabilistic scheme for the analysis of DNA microarray data.

Abstract: Secure detection of a finger in contact with a sensor. A fingerprint collation program is proposed, in which data outputted from a fingerprint reading sensor is classified into predetermined patterns according to data of a maximum threshold value or more, data of a minimum threshold value or less and an amplitude constituted by a difference obtained based on the maximum and the minimum values of the outputted data. Such fingerprint collation program determines the object that is in contact with the sensor based on the classified patterns. For example, determines whether the object is a finger or not. The present invention may be applied to a fingerprint collation apparatus.