Abstract: An image classification device includes a characteristic value set calculation unit 11 that calculates a characteristic value set of the whole image for each of multiple sets of image data in an image database 51, detects an edge of the set of the image data, and calculates a characteristic value set of the detected edge portions; a first clustering unit 12 that classifies the multiple sets of image data into multiple clusters on the basis of the characteristic value sets of the whole images; a second clustering unit 13 that further classifies the multiple clusters classified by the first clustering unit 12 into multiple clusters on the basis of the characteristic value sets of the edge portions; and a cluster integration unit 14 that determines which pixels constitutes a subject in each of the multiple sets of image data, based on the composition of the image, and integrates some of the multiple clusters classified by the second clustering unit 13 together based on the pixels constituting the subject.

Abstract: Disclosed is an image processing apparatus in which vector data from which noise has been removed and amount of data reduced can be generated in ideal fashion with regard to an illustration area selected from a document image. The document image is input to the apparatus and is segmented into a plurality of areas. A clip-art image is selected from these areas and a plurality of clusters are generated with regard to this image area. When a small area included in the plurality of clusters is determined as a noise, the noise is eliminated by combining the small area with a adjacent cluster. After noise is removed, the cluster group eventually obtained is converted to vector data.

Abstract: A method of detecting a road feature in an image signal derived from an infrared-sensitive camera. The method, in overview, comprises processing an image frame by assigning binary values to pixels in the frame in response to their representative temperature, and then to analyze spatially the binary mask to identify regions of pixels having mutually similar assigned binary values. The road feature is subsequently found from the analysis of the identified regions of mutually similar binary values and a visual indication of the road feature in relation to the image frame provided to the user.

Abstract: A method, system and computer product for visualizing affinities between objects. The method includes the steps of: forming a representation of a minimum spanning tree where the minimum spanning tree connects the plurality of objects based on a pairwise distance between the plurality of objects; forming a hierarchical cluster of the plurality of objects where the hierarchical cluster includes a level; agglomerating the plurality of objects based on the pairwise distance; displaying a view of the representation of the minimum spanning tree in a graphical user interface; receiving a user selection of a parameter containing a hierarchical level; and identifying, in the view, a target cluster that corresponds to the hierarchical level; where at least one of the steps is carried out using a computer device so that affinities between the plurality of objects are visualized.

Abstract: A handwriting processing apparatus and method effective for search of, e.g., a document file including handwriting is provided. When a handwriting characters are input to a coordinate input unit as a search key, a corresponding character in a dictionary is recognized for each of the handwritten characters, a search unit searches for a text code stored in a document file storage unit based on a text code of the corresponding character if the corresponding character is recognized and the search unit searches for handwriting trail data stored in a handwriting trail storage unit based on a handwriting trail of the handwriting character if the corresponding character is not recognized, thereby finding a desired document file.

Abstract: Disclosed is an apparatus and method for detecting a face from an image. The apparatus and method uses color components and enables a Support Vector Machine (SVM) having superior recognition performance to previously learn face and non-face images and determine whether an image is a face image based on a learned image database by reducing the size of a feature vector of a face as compared to conventional systems. Accordingly, the apparatus converts a face image into a mosaic image having a minimum size to reduce the dimension of the feature vector, in order to rapidly and correctly detect a face image.

Abstract: A method for associating with a digital image a class of a plurality of predefined classes having respective models, the method including the phases of dividing the digital image pixel by pixel into one or more regions belonging to a set of predefined regions that differ from each other on account of their type of content, the division being effected by establishing whether or not a pixel of the image belongs to a respective region on the basis of an operation of analyzing the parameters this pixel, the analysis operation being carried out by verifying that the parameters satisfy predefined conditions and/or logico-mathematical relationships of belonging to the respective region, acquiring from the digital image divided into regions information regarding the regions that are present in it, comparing this information with at least one model characterizing a respective class of said plurality, and associating with the digital image a class on the basis of the comparison phase.

Abstract: A system and method for an efficient semantic similarity search of images with a classification structure are provided. The system and method provide for building a semantic classification-search tree for the plurality of images, the classification tree including at least two categories of images, each category of images representing a subset of the plurality of images, receiving a query image, classifying the query image to select one of the at least two categories of images, and restricting the search for the image of interest using the query image to the selected one of the at least two categories of images.

Abstract: Video data matching includes both a single region of data and sequences of regions of data. In an embodiment, a video processing system selects, from a first matrix row corresponding to a test appearance model, one or more other appearance models as a function of similarity measures populating the first matrix row. After selection of the one or more other appearance models, the system then selects, from other matrix rows corresponding to the one or more other appearance models selected in the first step, one or more additional appearance models as a function of the similarity measures populating the other matrix rows. The system then ranks the appearance models selected from the first matrix row and the other matrix rows.

Abstract: A method and an apparatus process digital images. The method according to one embodiment accesses element data representing a plurality of elements belonging to a plurality of digital images; performs a similarity analysis between the elements from the plurality of elements to obtain inter-relational data results relating to the elements; and performs clustering of the plurality of elements, the step of performing clustering including incorporating in the inter-relational data results at least one hard constraint relating to elements from the plurality of elements, to obtain constrained inter-relational data results, performing a spectral analysis to obtain eigenvector results from the constrained inter-relational data results, and performing discretization of the eigenvector results using constrained clustering with a criterion to enforce the at least one hard constraint to obtain clusters.

Abstract: Vectors representing objects in n-dimensional space are approximated by local polar coordinates on partitioned cells of the data space in response to a query, e.g., a query data vector entered with a request to find “k” nearest neighbors to the query vector. A set of candidate near neighbors is generated using the approximations, with the local polar coordinates being independent of the dimensionality of the data space. Then, an answer set of near neighbors is returned in response to the query. Thus, the present invention acts as a filter to reduce the number of actual data vectors in the data set that must be considered in responding to the query.

Abstract: A local bi-gram model object recognition system and method for constructing a local bi-gram model and using the model to recognize objects in a query image. In a learning phase, the local bi-gram model is constructed that represents objects found in a set of training images. The local bi-gram model is a local spatial model that only models the relationship of neighboring features without any knowledge of their global context. Object recognition is performed by finding a set of matching primitives in the query image. A tree structure of matching primitives is generated and a search is performed to find a tree structure of matching primitives that obeys the local bi-gram model. The local bi-gram model can be found using unsupervised learning. The system and method also can be used to recognize objects unsupervised that are undergoing non-rigid transformations for both object instance recognition and category recognition.

Abstract: A method and system for detecting and classifying anomalies in a medical image. During anomaly detection, once the intensity of an image pixel crosses a detection threshold, the pixel is detected and linking inputs are provided to its nearest-neighbor pixels. The linking inputs increase the intensities of the neighbor pixels, which may result in the detection of these nearest-neighbor pixels if their linked intensities are above the threshold. Each detected anomaly is classified by determining a genetic response surface methodology (GRSM) model for the detected anomaly, determining a cancerous GRSM model from a database of cancerous anomalies, and comparing the cancerous GRSM model to the GRSM model for the detected anomaly to classify the detected anomaly as cancerous or non-cancerous.

Abstract: This invention is to enable retrieving of a content the searcher imagines in mind. The search method includes: obtaining a query brain image representing a brain activity state of a searcher when perceiving or imagining the content to be retrieved; identifying the content corresponding to the query brain image by using a structure associating a content with a brain image representing the brain activity state when perceiving the content; and outputting the content corresponding to the query brain image. Thus, by using the query brain image, without specifically indicating the content such as a drawing, the searcher can extract the pertinent content only by imaging the content.

Abstract: An image processing method comprises: clustering images of a set of images to generate a plurality of scenes each comprising an unordered cluster of images; ordering images within scenes respective to video coherence to generate ordered scenes comprising ordered sequences of images; and generating video content as images ordered in accordance with the ordered sequences of images comprising the scenes. In some embodiments, the video content is converted to video comprising video frames played at a predetermined frame rate.

Abstract: In a method of automatically creating a scalable relevance ordered representation of an image collection, the images in the image collection are classified into a plurality of clusters based upon a feature of the images. In addition, respective relevance levels of the images contained in each of the plurality of clusters are determined and the images in each of the plurality of clusters are ordered according to the relevance levels. Moreover, the images from the ordered plurality of clusters are arranged according to a predefined arrangement process to create the scalable relevance ordered representation of the image collection.

Abstract: A model selection method is provided for choosing the number of clusters, or more generally the parameters of a clustering algorithm. The algorithm is based on comparing the similarity between pairs of clustering runs on sub-samples or other perturbations of the data. High pairwise similarities show that the clustering represents a stable pattern in the data. The method is applicable to any clustering algorithm, and can also detect lack of structure. We show results on artificial and real data using a hierarchical clustering algorithm.

Abstract: A computer-implemented method for visualization of diffusion tensor images includes providing a diffusion tensor image input and providing a volume of interest within the diffusion tensor image input. The method includes determining a plurality of direction-based classifications of the volume of interest, wherein classes are defined by a set of parameters, An optimal solution is then selected within the classifications by using a criterion defined as a ratio of inertia indicators. To represent the chosen classification, a cone graph is determined for each of the directional classes to be displayed or stored, each cone pair being the geometrical interpretation of the class parameters. The method further includes determining a spherical scatterplot of the volume of interest augmented with a cone graph for visualization of at least one of the directional classes, and displaying and/or storing the scatterplot.

Abstract: A system and method for grouping cluster spines into a two-dimensional visual display space is provided. Clusters of concepts, each extracted from documents, are obtained. Spines are formed by placing the clusters sharing at least one of the concepts along a spine. The spines are ordered based on a length of each spine. One or more of the spines are selected as unique spines and placed into a visual display space. A similarity is determined between one of the spines not placed and each of the placed unique spines. The placed unique spine most similar is identified. An anchor cluster on the most similar unique spine that satisfies a threshold similarity with the unplaced spine is selected. A cluster on the unplaced spine that is most similar to the selected anchor cluster is identified and grafted to the selected anchor cluster as a group of cluster spines.

Abstract: Object comparison is disclosed, including: adapting N universal mixture model components to a first object to generate N corresponding first object mixture model components, where N is an integer greater than or equal to two; and generating a similarity measure based on component-by-component comparison of the N first object mixture model components with corresponding N second object mixture model components obtained by adaptation of the N universal mixture model components to a second object.

Abstract: An image of high-aspect-ratio objects is analyzed to locate linear features within the image. The gradient direction for each pixel is determined, and connected pixels having similar gradient directions are grouped into line support regions. A linear feature is determined for each line support region. The linear features are analyzed to identify those that are co-linear. The linear features are then analyzed to find groups that are parallel and that sufficiently overlap. Additional sets of linear features that intersect the identified group are combined into that group. The line support regions of the grouped linear features are analyzed to determine representative orientation and/or dimensional information, which is analyzed to determine statistical information about a plurality of the objects. This information is used to monitor or control one or more processes associated with the objects, to reject the objects or a structure or device comprising the objects, or the like.

Abstract: A clustering processing method for dividing a samples into a plurality of clusters based on a feature amount of each sample, the plurality of clusters each belonging to one of a plurality of layers composed of M layers (M=2, . . . , K), the clustering processing method comprises a sample allocating step of allocating a sample targeted for processing to a cluster belonging to a first layer, based on a result of comparing the feature amount of the target sample with a representative feature amount of each of clusters belonging to the first layer; a determination step of determining whether to allocate a cluster belonging to an M?1th layer to an Mth layer; and a cluster allocating step of allocating a cluster belonging to the M?1th layer to the Mth layer, if it is determined in the determination step to allocate a cluster belonging to the M?1th layer to the Mth layer.

Abstract: A method and an apparatus process images. The method according to one embodiment accesses digital image data representing an image including an object; clusters pixels of the image to obtain clusters; generates a graph with pixels of the image and cluster membership information of pixels in the clusters; and segments the graph using a max-flow segmentation to obtain pixels of the image associated with the object.

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: The performance of image processing is assured by executing region segmentation of an image at high speed. Pixels are classified into clusters conforming to a feature such as color. In a case where a suitable cluster for classifying a pixel is undefined, a new cluster is defined and the pixel is classified into this cluster. An upper-limit value on number of clusters is set prior to classification. If the number of clusters exceeds the upper-limit value, clusters already defined are merged. This limits an increase in the number of clusters.

Abstract: Every time clustering processing for a predetermined number of pixels is complete, a small cluster having the number of allocated pixels, which is equal to or smaller than a pixel count threshold, is discriminated. The small cluster, which is discriminated to have the number of allocated pixels equal to or smaller than the pixel count threshold, is merged to a cluster having the nearest representative feature vector. With this arrangement, the number of clusters which are to undergo distance calculations of feature vectors is reduced. According to this arrangement, region segmentation of an image can be executed faster by the clustering processing.

Abstract: Disclosed is a method for assigning the content of a digital image to a class of a classification system. Said method comprises the following steps: —a predetermined number of F numerical shape characteristics ?m are determined; —the value of each shape characteristic of the F numerical shape characteristics determined for the image is compared to the value filed in a table for the respective shape characteristic, values for the individual numerical shape characteristics being allocated to each class in the table; —the class in which the F numerical shape characteristics determined for said image correspond best to the values of the numerical shape characteristics indicated in the table for said class is output as the class into which the image that is to be recognized is classified.

Abstract: According to embodiments, an image display apparatus includes: a face detection processing section configured to detect face areas included in video content to generate face cut-out images including the face areas; a face clustering processing section configured to group a plurality of face cut-out images included in the video content by respective characters of the video content to classify clusters corresponding to the characters; an evaluation section configured to obtain evaluation values by evaluating the plurality of face cut-out images in relation to one or more evaluation items among a plurality of evaluation items corresponding to a plurality of features included in the face cut-out images; and a selection section configured to select face cut-out images in which evaluation values are within a predetermined range from among the plurality of face cut-out images in the clusters as representative face icon images used for display.

Abstract: Representative embodiments are directed to systems and methods for processing training data for a statistical classification application. In one embodiment, confidence values are calculated for training data elements to identify the probabilities of the training data elements belonging to identified classes. An interactive scatter plot is generated using the calculated confidence values. The scatter plot visually indicates the confidence values of points in the scatter plot. Accordingly, the user is able to identify potentially misclassified training data elements. The user may select training data elements from the scatter plot and reclassify training elements as appropriate. Upon reclassification, the confidence values may be recalculated and the scatter plot revised.

Abstract: In a computer-mediated method for providing representative images, image records are classified spatio-temporally into groups. In each group, image records are partitioned into clusters and the hierarchically highest cluster is ascertained. The partitioning is between a hierarchy of feature clusters and a remainder cluster, based on a predetermined plurality of saliency features. Feature clusters each have one or more of the saliency features. The remainder cluster lacks the saliency features. Feature clusters are each exclusive of the saliency features of any higher clusters in the hierarchy and non-exclusive of the saliency features of any lower feature clusters in the hierarchy. A representative image of each group is designated from respective image records based on: the respective saliency feature of the highest cluster when the highest cluster is a feature cluster and independent of the saliency features when the highest cluster is the remainder cluster.

Abstract: A feature amount obtained from an image group for registration is stored as dictionary data in a memory. When an input image group constituted of a plurality of images is supplied to be collated with the dictionary data, the images constituting the input image group are classified based on a specific classification standard, a feature amount obtained from each of the classified image groups is collated with the feature amount stored as the dictionary data in the memory, and a result of the collation of all the input image groups with the dictionary data is determined based on a result of integrating a plurality of collection results of the classified image groups with the dictionary data.

Abstract: A system and method for determining a classifier to discriminate between two classes—object or non-object. The classifier may be used by an object detection program to detect presence of a 3D object in a 2D image (e.g., a photograph or an X-ray image). The overall classifier is constructed of a sequence of classifiers (or “sub-classifiers”), where each such classifier is based on a ratio of two graphical probability models (e.g., Bayesian networks). A discrete-valued variable representation at each node in a Bayesian network by a two-stage process of tree-structured vector quantization is discussed. The overall classifier may be part of an object detector program that is trained to automatically detect many different types of 3D objects (e.g., human faces, airplanes, cars, etc.). Computationally efficient statistical methods to evaluate overall classifiers are disclosed. The Bayesian network-based classifier may also be used to determine if two observations (e.g., two images) belong to the same category.

Abstract: A computer research tool for indexing, searching and displaying data is disclosed. Specifically, a computer research tool for performing computerized research of data including textual objects in a database or a network and for providing a user interface that significantly enhances data presentation is described. Textual objects and other data in a database or network is indexed by creating a numerical representation of the data. The indexing technique called proximity indexing generates a quick-reference of the relations, patterns and similarity found among the data in the database. Proximity indexing indexes the data by using statistical techniques and empirically developed algorithms.

Abstract: An apparatus, method, and medium for detecting a face in an image using classifiers generated by a boost algorithm are provided. The apparatus may include an image dividing unit dividing a black and white image into sub-windows; a sub-window detection unit detecting a sub-window corresponding to a face image in the divided sub-windows using the front part of the cascade of the classifiers and storing the sub-window detected as the face image; and a post-processing unit verifying whether each of sub-windows included in a group among a plurality of groups into which stored sub-windows are grouped is the face image using the number of child detections in each group and the latter part of the cascade of classifiers, integrating the sub-windows in the group according to the result of the verification, and outputting the result of the integration as the face image included in the black and white image.

Abstract: A database has chronologically ordered images classified into event groups based upon a time difference threshold, and into subgroups based upon a similarity measure. In a method and system for combining new images into such a database, new images are ordered into clusters based upon assessed image features. A representative image is selected in each cluster. A database segment chronologically overlapping the new images is designated and a set of database images similar to each representative image are identified in the segment. Different subgroups including one or more retrieved images are associated with each cluster to provide matched subgroups. The new images are assigned to matched subgroups associated with respective clusters.

Abstract: A method and apparatus of animation planning for a dynamic graph is provided. A first method includes grouping changed nodes into atomic partitions according to the relationships between the changed nodes and existing nodes. The next step is clustering the atomic partitions according to clustering distances between the atomic partitions to generate a running list for planning animation of node changes in the dynamic graph. The running list includes elements selected from the group consisting of clusters of atomic partitions and atomic partitions. The present invention is able to plan a multi-stage animation demonstration of changes of a dynamic graph and has a improved readability and animation speed.

Abstract: Properties of pixels of a digital image are sampled within different subdivisions of an editing tool impression to produce different property distributions. The subdivisions may be differently-located within the tool impression. The property distributions from each region are classified to identify different edit classes within the property space, which are then used to apply an edit effect to the digital image within the tool impression. The edit classes may be represented by an edit profile in two or more dimensions (e.g., applying to one or more pixel properties).

Abstract: A system and method of recording digitally stored surveillance images comprises dividing a series of continuous digital picture image frames according to timing intervals; storing a reference image in a frame buffer of a motion analyzer; detecting a motion difference between a captured image and the reference image; attributing a motion code and time stamp on the captured image; and recording the captured image when the motion code and time stamp match a predetermined threshold.

Abstract: A person cluster selection part selects one or more person clusters as representative person clusters in order of largest number of facial images included therein. A unit selection part selects, from each representative person cluster, (i) a unit having the highest likelihood as a first representative unit, and (ii) units in order of lowest likelihood as a second representative unit onward. A representative facial image selection part selects, from each representative unit, (i) a facial image having the highest likelihood as a first representative facial image, and (ii) facial images in order of lowest likelihood as a second representative facial image onward.

Abstract: A pattern retrieving unit includes an acquiring unit that acquires a representative pattern image; a setting unit that sets a use condition of a type of shape feature of the representative pattern image; a receiving unit that receives a query pattern image used as a query; a determining unit that determines the type of shape feature of the representative pattern image to be the type of shape feature of the query pattern image, based on the use condition; a retrieving unit that retrieves a pattern image identical or similar to the query pattern image, using determined type of shape feature; and an output unit configured to output a result of retrieval by the retrieving unit.

Abstract: A system for automatically classifying an object of a target image is provided. A classification system provides a collection of classified images along with a classification of the dominant object of the image. The classification system attempts to classify the object of a target image based on similarity of the target image to the classified images. To classify a target image, the classification system identifies the classified images of the collection that are most similar to the target image based on similarity between salient points of the target image and the classified images. The classification system selects a classification associated with the classified images that are most similar to the target image as a classification for the object of the target image.

Abstract: Clustering biometric templates is performed by determining fiduciary templates and cluster seed templates, both from a gallery of biometric templates. Similarity vectors are formed by comparing members of the cluster seed templates to the fiduciary templates. The gallery is then partitioned into clusters based upon the similarity vectors, and the clusters are populated from the remainder of the gallery. Partitioning may be performed by a classifier that implements a supervised machine learning algorithm that is trained with the similarity vectors, such as a multi-decision tree classification system. Matching may be accommodated by accessing a probe template, determining a cluster neighborhood for the probe template, and searching the cluster neighborhood to determine whether the gallery includes a match corresponding to the probe template. The same similarity metric is used both to partition a gallery into clusters, and in matching a probe template to the so-clustered gallery.

Abstract: An apparatus and method for generating additional information about moving picture content, including: comparing image feature information about each image frame in moving picture content with image feature information about each image frame in web information, searching for an image frame in the moving picture content, the image frame matching the image frame in the web information, determining location information about the found image frame in the moving picture content, and generating additional information by use of the determined location information and the web information.

Abstract: Techniques for shape clustering and applications in processing various documents, including an output of an optical character recognition (OCR) process.

Abstract: A data analyzer/classifier comprises using a preprocessing step, and energy minimization step, and a postprocessing step to analyze/classify data.

Abstract: A method of browsing face regions in digital images in a photo displaying system includes detecting a plurality of face regions from a plurality of images, grouping the face regions into a plurality of clusters based on similarities of the face regions, determining a degree of connection between the clusters, modifying the degree of connection between the clusters according to a relationship of the face regions, and displaying the face regions according to the degree of connection between the clusters.

Abstract: The invention enables the digital management of large scale image databases, to efficiently classify and index image data independent of language. Complex processing requirements are required only on reduced and operably small subsets of the entire collection, thereby effectively scaling large document collections. Embodiments of the present invention provide image-based classification and retrieval of documents based on image recognition, e.g., signatures, logos, stamps, or word-spotting; in documents within real time for large datasets such as in the millions of documents.

Abstract: Clustering-definition-based encoding schemes printed as a computer-oriented digital layer onto a physical surface whereby the surface functions as an interface enabling, via a simple image-acquisition device scanning a minimal field required to establish the clustering definition, the provision, to a suitable data processing and human-oriented display means, of feedback on the orientation of the image-acquisition device relative to the digital layer overall.

Abstract: Properties of pixels of a digital image are sampled within different subdivisions of an editing tool impression to produce different pixel property distributions. The property distributions from each region may be automatically classified to identify different edit classes within the property space, which are then used to apply an edit effect to the digital image within the tool impression. The edit classes are represented by an edit profile, the generation of which may be completely automated based on selection of a tool impression, or partially automated using the selection of the tool impression and receipt of classification guidance input, such as one or more parameters received from user input or a configuration file. The edit classes may also be generated without reference to the pixel property distributions, such as via user input.

Abstract: Techniques for media segmentation are disclosed. A number of measures used to generate similarity values are compared. Two different approaches to calculate feature vectors based on pairwise frame similarity in combination with different kernel functions are disclosed. A supervised classification method is used with the feature vectors to assess segment boundaries.