Abstract: A system of determining three-dimensional information from the information that is contained into discrete two-dimensional images. The two images may be obtained from two uncalibrated cameras. The information from the two uncalibrated cameras is first rectified, and then used to form a disparity surface indicating the third dimensional information. Various techniques are used to increase the speed of conversion.

Abstract: Described herein is a technology for semantically classifying areas of an image (and/or the images themselves) as one of a number of multiple discriminating categories. More particularly, the technology employs one or more hierarchical, probabilistic techniques for performing such classification. Such technology is particularly useful in fields of image classification and image retrieval. The architecture of such technology employs multiple hierarchical layers. The architecture is based on modeling class likelihoods at each of such layers separately and then combining these to form an overall estimate of the posterior, conditioned on the data. The task of combining the estimated class likelihoods at each layer is made more efficient by assuming statistical independence between layers. This abstract itself is not intended to limit the scope of this patent. The scope of the present invention is pointed out in the appending claims.

Abstract: The invention relates to a method and a device for reading the addresses of mailings with an OCR device and a video coding device. On an ambiguous reading result from the OCR device an extract is automatically generated with specified extraction rules from address alternatives as a partial result of the OCR process and the number of concurrent or non-concurrent extracts determined. When the number of the address alternatives does not exceed a maximum value, an extraction coding with the specified extraction rules is carried out with predominantly or only different extracts and a selection coding, for the address alternatives transmitted to the video coding device with the image, is carried out with predominantly or only the same extracts.

Abstract: In a mobile robot and a method for measuring a moving distance thereof, by including an image capture unit for photographing the bottom surface according to motion of a mobile robot at a certain intervals and capturing images; a displacement measurer for measuring displacement about the captured image; and a microcomputer for outputting an actual moving distance by direction and motion of the mobile robot on the basis of the measured displacement value, it is possible to measure an accurate moving distance of the mobile robot with only one image sensor installed at the center of a body of the mobile robot, and accordingly it is possible to simplify a mechanical structure and facilitate maintenance and repairing.

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: 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 multilevel Chain-And-Tree (CAT) model provides a framework to facilitate highly effective image analysis and measurement. Methods are described to automatically or semi-automatically create a processing sequence for image based decision applications using the CAT model framework. The invention supports new image based decision applications with little or no involvement of image processing experts. A new representation for processing sequence is consistent with CAT data representation, which allows easy understanding, updating, and debugging.

Abstract: A multilevel Chain-And-Tree model provides a framework for an image based decision system. The decision system enables separation of effects of defects within one component from other components within a common subject. The framework provides for linking of structure constraints of components of a common subject and for checking and resolving their consistency. The framework allows discrimination between subtle image changes and natural variations of the subject. The framework for standard data representation facilitates production process control.

Abstract: A digital watermark information detection method for detecting information inserted as a digital watermark from contents has a step of converting the contents into a predetermined specified format and a step of trying to detect information from the converted contents. When a plurality of specified formats are defined in advance and detection of the information from the converted contents fails, the contents is converted into another specified format and the information detection is retried.

Abstract: An efficient and non-iterative post processing method and system is proposed for mosquito noise reduction in DCT block-based decoded images. The post-processing is based on a simple classification that segments a picture in multiple regions such as Edge, Near Edge, Flat, Near Flat and Texture regions. The proposed technique comprises also an efficient and shape adaptive local power estimation for equivalent additive noise and provides simple noise power weighting for each above cited region. An MMSE or MMSE-like noise reduction with robust and effective shape adaptive windowing is utilized for smoothing mosquito and/or random noise for the whole image, particularly for Edge regions. Finally, the proposed technique comprises also, for chrominance components, efficient shape adaptive local noise power estimation and correction.

Abstract: A systematic way of linking structure constraints of components of a common object and checking and resolving their inconsistency is used to improve detection results in image-based decision systems. A multilevel Chain-And-Tree (CAT) model is used to direct processing using both forward and backward scans through the related components. Since components occur as parts of an object, the context (relational structure) in which the component appears can be used to reduce noise and variation affects. In the method, object knowledge is translated into constraints between components. The constraints are used to enhance feature detection, defect detection, and measurement accuracy and consistency.

Abstract: An image processing method of extracting the points of a path following a threadlike structure in an image (IP) formed by a grid of Potential points. A first processing step (4) implements a first path-tracking operation using a front marching technique denoted Filiation Front Marching Technique (FFM) for supplying at least one First Track of the threadlike structure, formed by succeeding points denoted Fathers and Children, by marching a Front of points forwards from a fixed Start point (A) to a fixed End point (B). A second processing step (5) implements a second path-tracking operation using the Filiation Front Marching Technique for supplying a Best Path from one First Track by back propagating the Front starting at the End Point and going through already determined Children and Fathers until the Start Point is reached.

Abstract: The invention relates to a method for segmentation of a digital image being represented by picture element data for the purposes of visualization, segmentation and quantification, such as determining the volume of a segment. The method comprises the steps of performing a watershed transformation of the digital image by applying a watershed transformation method on the picture element data to provide one or more basins and post-processing by processing the picture element data belonging to at least one of the basins, where post-processing is one or a combination of the following steps: (i) gray-scale based segmentation, (ii) visualization using a transfer function, or (iii) volumetry using histogram analysis.

Abstract: An image processing apparatus deals with a plurality of objects each having a pair of image data representative of an image and image attribute information representative of an attribute of the image, wherein the plurality of objects are linked in form of a hierarchy structure including a parentage. The image processing apparatus comprises object producing means for producing new objects upon succession of image attribute information of a parent object, object deleting means for deleting an object and the associated descendant objects as well, and attribute information altering means for altering an attribute of an object and an attribute of the associated descendant objects as well.

Abstract: An image processing device of the present invention is provided with four kinds of sub masks in total including two kinds in a main scanning direction and two kinds in a sub scanning direction, in a main mask constituted by a plurality of pixels including a target pixel. In the image display device, when determining a target pixel of an inputted image data, a difference in a total density of the two kinds of sub masks in a main scanning direction is added to a normalized difference in total density of the two kinds of sub masks in a sub scanning direction, and a resultant value is compared with a threshold value so as to determine if the target pixel is an edge area or not.

Abstract: A method (300) and apparatus (400) is described for segmenting an image (102). Starting with each pixel of the image (102) being a separate region, segments are formed by merging the regions. As merging proceeds, a merging cost of the regions being merged generally increases. This increase however is not purely monotonic as the overall rise in the merging cost is punctuated by departures from monotonicity. A complete pass is made through the segmentation, in which all regions are merged until only one remains. By analysing the points immediately after significant departures from monotonicity, a final segmentation stopping value (?stop) is chosen as being the last return to monotonicity from such a significant departure. Segmentation is repeated until the merging cost reaches the final segmentation stopping value (?stop).

Abstract: The present invention relates to an information processing method for recognizing a pattern of an image, character, speech or the like. According to the method, a feature space in which a point set given as learning patterns is present is divided to form a classification tree on the basis of the learning patterns. The method comprises the linear combination feature amount generation step of generating a new feature amount by a linear combination of the feature amounts of the learning patterns; the hierarchization pre-preprocessing step of hierarchizing, in advance, the new feature amount generated in the linear combination feature amount generation step; and the classification tree generation step of generating a classification tree on the basis of the learning patterns hierarchized in the hierarchization pre-processing step.

Abstract: A method for a recognition system, for example for recognizing a location name according to a preset, spelled voice entry, in order to shorten the reaction time for a new start of a system in which data must first be transferred from a reading memory into a working memory via a connection at a low data rate. According to the method, the data are grouped in data blocks, corresponding to successive structural levels, with the data content being arranged in the manner of a branched tree structure, and with the data blocks in the hierarchical sequence of the associated structural branch levels being consecutively transferred from the reading to the working memory. The recognition has already begun, on the basis of the still-incomplete data set, after a plurality of levels is present, and is continued in real time with the transfer of further data blocks. The average processing time can be reduced further through an advantageous structuring and grouping of the data.

Abstract: A digital image processing method for locating human eyes in a digital image, includes the steps of detecting a skin colored region in the image; detecting human iris color pixels in the skin colored region; forming initial estimates of eye positions using the locations of the detected iris color pixels in the skin colored region; estimating the size of each eye based on the distance between the estimated initial eye positions; forming a first search window for one eye, the center of the window being the estimated initial position for the one eye and the size of the window being proportional to the estimated size of the one eye; and employing a template to locate an eye in the first search window.

Abstract: Methods and systems used to associate color calibration profiles with scanned images based on identifying the marking process used for an image on a substrate using spatial characteristics and/or color of the image. Image types which are classified and identified include continuous tone images and halftone images. Among halftone images separately identified are inkjet images, xerographic images and lithographic images. Locally adaptive image threshold techniques may be used to determine the spatial characteristics of the image.

Abstract: Methods and systems used to associate color calibration profiles with scanned images based on identifying the marking process used for an image on a substrate using spatial characteristics and/or color of the image. Image types which are classified and identified include continuous tone images and halftone images. Among halftone images separately identified are inkjet images, xerographic images and lithographic images. Locally adaptive image threshold techniques may be used to determine the spatial characteristics of the image.

Abstract: Methods and systems used to associate color calibration profiles with scanned images based on identifying the marking process used for an image on a substrate using spatial characteristics and/or color of the image. Image types which are classified and identified include continuous tone images and halftone images. Among halftone images separately identified are inkjet images, xerographic images and lithographic images. Locally adaptive image threshold techniques may be used to determine the spatial characteristics of the image.

Abstract: Methods and systems used to automatically identify the marking process used for an image on a substrate based on spatial characteristics and/or color of the image. Image types which are classified and identified include continuous tone images and halftone images. Among halftone images separately identified are inkjet images, xerographic images and lithographic images. Locally adaptive image threshold techniques may be used to determine the spatial characteristics of the image.

Abstract: Methods and systems used to associate color calibration profiles with scanned images based on identifying the marking process used for an image on a substrate using spatial characteristics and/or color of the image. Image types which are classified and identified include continuous tone images and halftone images. Among halftone images separately identified are inkjet images, xerographic images and lithographic images. Locally adaptive image threshold techniques may be used to determine the spatial characteristics of the image.

Abstract: A classification model optimum for realization of a defect classification request by a user is not known by the user. Then, the user sets a classification model which is not necessarily suitable and makes classification, resulting in degradation in classification performance. Therefore, the present invention automatically generates plural potential classification models and combines class likelihoods calculated from the plural classification models to classify. To combine, an index about the adequacy of each model, in other words, an index indicating a reliable level of likelihood calculated from the each potential classification model, is also calculated. Considering the calculated result, the class likelihoods calculated from the plural classification models are combined to execute classification.

Abstract: A method for candidate frame selection involves sampling the source frames of the source video at a predetermined fixed periodic interval. The largest frame differences represent candidate boundaries, and the frames before and after the N/2 largest candidate boundaries are selected as candidate frames. A method for selecting keyframes involves clustering all candidate frames into a hierarchical binary tree using a hierarchical agglomerative clustering algorithm. Optionally, the pairwise distances for the members of the two clusters is modified according to class membership of the members, which is preferably determined statistically from image class statistical models. A method for selecting M clusters from which keyframes are extracted involves splitting the M-1 largest clusters of a hierarchical binary tree of clusters. Optionally, clusters not having at least one uninterrupted sequence of frames of at least a minimum duration are filtered out.

Abstract: A method for indexing and retrieving manufacturing-specific digital images based on image content comprises three steps. First, at least one feature vector can be extracted from a manufacturing-specific digital image stored in an image database. In particular, each extracted feature vector corresponds to a particular characteristic of the manufacturing-specific digital image, for instance, a digital image modality and overall characteristic, a substrate/background characteristic, and an anomaly/defect characteristic. Notably, the extracting step includes generating a defect mask using a detection process. Second, using an unsupervised clustering method, each extracted feature vector can be indexed in a hierarchical search tree. Third, a manufacturing-specific digital image associated with a feature vector stored in the hierarchicial search tree can be retrieved, wherein the manufacturing-specific digital image has image content comparably related to the image content of the query image.

Abstract: Many tasks in computer vision involve assigning a label, such as disparity for depth of field data to every pixel. Energy minimization may be used to accomplish this labeling. The present invention provides an efficient way of minimizing energy functions in assigning labels to pixels. Two methods of using graph cuts to compute a local minimum are described. The first move is an &agr;&bgr; swap. For a pair of labels, &agr;, &bgr;, this move swaps the labels between an arbitrary set of pixels labeled a and another arbitrary set of pixels labeled &bgr;. The first method generates a labeling such that there is no swap move that decreases the energy. The second move is the &agr;-expansion. For a label &agr;, this move assigns an arbitrary set of pixels with the label &agr;.

Abstract: A method for fractal-darwinian object generation consists of the steps of: preparing a fractal object library including predetermined objects and associated rules of property and context, forming objects and comparing the formed objects with the objects in the fractal object library. By using the property rules, a local classification likelihood is allocated to each formed object. Thereupon, by using the context rules for each object, a respective fractal classification likelihood is formed. For optimisation of the fractal classification likelihood, alteration rules are applied to the objects. The above method is carried out iteratively, whereby a process of gradual optimisation takes place.

Abstract: A method and system for pre-processing a radiation image containing a clinical region of interest and a non-clinical region. The method includes forming a border proximate the clinical region and performing image filtering to filter the clinical region and the border. The non-clinical region may be collimated, and at least a portion of the collimated region may be replaced with values based on pixel values within the region of interest. The system includes obtaining a radiation image having clinical and non-clinical regions, apparatus identifying the non-clinical region, a multi-resolution pre-processing subsystem removing the non clinical region to form an intermediate image and then adding a border to the intermediate image, and a multi-resolution imaging subsystem performing multi-resolution processing of the clinical region and border. Assigned border pixels values may be based on, equal to, or mirror associated or adjacent clinical region pixel values.

Abstract: A method is presented for encoding auxiliary information in digital images which enables the encoding of both compact auxiliary information specific to a given image and a secret message into a given image such that the hidden message contained therein is only revealed in such a way that neither the image itself nor the auxiliary information alone are sufficient to reveal the secret message. Specifically, a method for encoding an auxiliary message in a digital image, is presented having the first step of creating a set of Training Vectors from all N×N windowed observations of a digital image (D) and corresponding pixels of a message (I). Then, statistically estimating a function T from the set of Training Vectors which map the N×N windowed observations of the digital image (D) to the corresponding pixels of the message (I); and writing out the statistically estimated function T as a representation of the hidden message.

Abstract: A two-dimensional (2D) image model models the layout structure of a class of document images as an image grammar and includes production rules having explicit layout parameters as data items that indicate information about the spatial relationships among image constituents occurring in images included in the class. The parameters are explicitly represented in the grammar rules in a manner that permits them to be automatically trained by a training operation that makes use of sample document images from the class of modeled documents. After each sample image is aligned with the 2D grammar, document-specific measurements about the spatial relationships between image constituents are taken from the image. Optimal values for the layout parameters are then computed from the measurement data collected from all samples.

Abstract: A multilevel Chain-And-Tree (CAT) model provides a framework to facilitate highly effective image analysis and measurement. Methods are described to automatically or semi-automatically create a processing sequence for image based decision applications using the CAT model framework. The invention supports new image based decision applications with little or no involvement of image processing experts. A new representation for processing sequence is consistent with CAT data representation, which allows easy understanding, updating, and debugging.

Abstract: A text recognition system represents the decoded message of a document image as a path through an image network. A method for integrating a language model into the network selectively expands the network to accommodate the language model only for certain ones of the paths in the network, effectively managing the memory storage requirements and computational complexities of integrating the language model efficiently into the network. The language model generates probability distributions indicating the probability of a certain character occurring in a string, given one or more previous characters in the string. Selectively expanding the image network is achieved by initially using upper bounds on the language model probabilities on the branches of an unexpanded image network. A best path search operation is then performed to determine an estimated best path through the image network using these upper bound scores.

Abstract: The invention provides a technique for pattern recognition that employs a state machine that incorporates a sequence of table-look-up operations. A sequence of input parameters, which could be derived according to some application-specific algorithm, is used to generate a corresponding sequence of memory addresses that is used for the table-look-up operations. The memory tables are organized in a hierarchical, or tree-like, structure that corresponds to the input sequence. The table data is designed for the recognition of a specific set, or library, of input patterns. A given input sequence traces an input-specific path through the memory tables until one of the patterns in the library is recognized or until it is determined that the input sequence is inconsistent with any of the patterns in the library. For each pattern in the library, the table data is designed to accommodate the variations in the input values that are specific to the application (e.g., variations due to noise and/or tolerances).

Abstract: Given a system which detects simple events, one can define a complex event by constructing a list of sub-events. In order to recognize a complex event, the system keeps a record of the sub-events that have occurred thus far and the objects involved in these sub-events. Whenever the first sub-event in a complex event's sequence is recognized, an activation for that complex event is created. The activation contains an indication of the identity of the object involved in the event. The activation also includes an index initialized to one. If a newly detected event matches the next sub-event in any of the currently open complex events, the index for that complex event is incremented. If the index reaches the total number of sub-events in that complex event, the complete complex event is recognized. Thus any desired alarm is generated.

Abstract: A multilevel Chain-And-Tree model provides a framework for an image based decision system. The decision system enables separation of effects of defects within one component from other components within a common subject. The framework provides for linking of structure constraints of components of a common subject and for checking and resolving their consistency. The framework allows discrimination between subtle image changes and natural variations of the subject. The framework for standard data representation facilitates production process control.

Abstract: A method using CC and MLO views localized in a display in 3-D as a breast virtual model, incorporating, feature capturing using gradient, energy and entropy codes, calibration classification according to features using binary decision trees, nipple detection, and 3-D spatial coordinate transformation and display of breast clustered microcalcifications.

Abstract: Methods of organizing a series of sibling data entities in a digital computer are provided for preserving sibling ranking information associated with the sibling data entities and for attaching the sibling ranking information to a joint parent of the sibling data entities to facilitate on-demand generation of ranked parent candidates. A rollup function of the present invention builds a rollup matrix (126) that embodies information about the sibling entities and the sibling ranking information and provides a method for reading out the ranked parent candidates from the rollup matrix in order of their parent confidences (141). Parent confidences are based on the sibling ranking information, either alone or in combination with n-gram dictionary ranking or other ranking information.

Abstract: A multipoint inspection system for evaluating manufactured assemblies includes a robotic arm and a robot controller for controlling the robotic arm. A camera is mounted to an end of the robotic arm and includes a camera controller for capturing images. The robot controller is in communication with the camera controller and the robot controller causes the robotic arm to position the camera at a first inspection point. The camera controller then causes the camera to capture a first inspection point image of a manufactured assembly at the first inspection point. The robotic controller then causes the robotic arm to position the camera at a next inspection point where the camera controller then causes the camera to capture a next inspection point image of the manufactured assembly at the next inspection point.

Abstract: A system and method for object inspection that includes an image recognition program stored on a tangible medium for classifying and subclassifying regions of interest on an image. The image recognition program can be used in an image inspection system to determine defects on objects such as printed wiring assemblies. The image recognition program is executable to collect raw image data, segment out rectangular regions of interest that can be component sites defined by CAD data, preprocess each region of interest by scaling, gain and offset correction, and gamma correction, generating a set of image spaces for each region of interest using a set of spatial image transforms, generating features on the image spaces, scoring the features, comparing the feature scores to a knowledge base of feature scores to make a class determination for the features, generating a presence/absence decision confidence for the features, calculating a class determination and decision confidence for each region of interest.

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: Stroke information from pen-down to pen-up being one unit is grouped automatically so that editing thereafter can be performed efficiently. Grouping of stroke information is determined in accordance with a predetermined rule of grouping each time stroke information is input or with timing indicated by a user. Strokes input within a predetermined time interval and stroke information input at a position within a predetermined distance interval are assumed to be the same group, and in the editing thereafter, a collective pointing and processing on a plurality of grouped strokes becomes possible.

Abstract: The present invention is an apparatus and method for classifying high-dimensional sparse datasets. A raw data training set is flattened by converting it from categorical representation to a boolean representation. The flattened data is then used to build a class model on which new data not in the training set may be classified. In one embodiment, the class model takes the form of a decision tree, and large itemsets and cluster information are used as attributes for classification. In another embodiment, the class model is based on the nearest neighbors of the data to be classified. An advantage of the invention is that, by flattening the data, classification accuracy is increased by eliminating artificial ordering induced on the attributes. Another advantage is that the use of large itemsets and clustering increases classification accuracy.

Abstract: An image processing apparatus having an image judgement section for judging whether or not a specific pattern image is included in at least a part of the image data input from an image input section, and a central processing unit for controlling the output of an image based on the judgement result by the image judgement section, wherein the image judgement section comprises a plurality of specific pattern image control tables for controlling different specific pattern images, respectively, and every time image data is input from the image data input section, the order of the plurality of specific pattern image control tables is changed over to perform the image judgement.

Abstract: A measure of importance is calculated for segmented parts of a video. The segmented parts are determined by segmenting the video into component shots and then merging by iteration the component shots based on similarity or other factors. Segmentation may also be determined by clustering frames of the video, and creating segments from the same cluster ID. The measure of importance is calculated based on a normalized weight of each segment and on length and rarity of each shot/segmented part. The importance measure may be utilized to generate a video summary by selecting the most important segments and generating representative frames for the selected segments. A thresholding process is applied to the importance score to provide a predetermined number or an appropriate number generated on the fly of shots or segments to be represented by frames. The representative frames are then packed into the video summary.

Abstract: A method of machine learning utilizing binary classification trees combined with Bayesian classifiers in which a training component includes creating nodes, utilizing the expectation maximization algorithm to create statistical kernels and mixtures for leaf nodes and kappa statistics to identify nodes as leaf nodes or branch nodes, creating hyperplanes for branch nodes, splitting the training data for branch nodes into subsets to be provided to subtrees, and in which an operating component traverses a binary classifier tree with a feature vector to be classified by determining for each branch node whether the feature vector lies to the left or the right of the branch node hyperplane and finally classifies the feature vector upon arriving at a leaf node by computing log likelihoods for the feature vector for each mixture in the leaf node and determining the classification of the feature vector according to the highest log likelihood, in which the preferred embodiment described is classification of internal com

Abstract: Disclosed is a method for classifying samples wherein the information content of a sample is divided up into two information areas. The first area contains necessary information and the second area contains sufficient information. An initially imprecise preselection concerning a plurality but extremely limited number of classes is made through the necessary information area. After said preselection, identification occurs with the sufficient information area in order to pinpoint the initially imprecise preselection in relation to an effective target class. This enables classification quality to be improved by means of new classifiers or new principal formulations of classification formulations.

Abstract: Provided is a method and apparatus of processing an image using filters. The method and apparatus receives an input pixel and a pixel window associated with the input pixel from the image, classifies the input pixel using the pixel window into a range of classes identifying pixels suitable for various degrees of smoothing and sharpening operations, receives parameter independently set for sharpening and smoothing the image, and selects a filter for processing the input pixel based upon the pixel classification and the parameter settings for sharpness and smoothness.

Abstract: It is an object to print an image as it is inherently formed. Objects constructing the image are separated to character train objects in which there is no need to consider an overlap and the other objects by a character train separator. A character train converter converts the separated character train objects in which there is no need to consider the overlap into character code information and transmits to a printer. The other objects are converted into image information by a drawer and transmitted to the printer. The printer synthesizes a glyph formed by a glyph generator from the character code information onto the received image information and outputs.