Abstract: Face detection is performed on a plurality of images to identify a plurality of faces. A subset of the plurality of faces is activated, including by loading into memory the subset of faces and/or data corresponding to the subset of faces. At least one of the plurality of faces is left un-activated. A distance for each pair of activated faces is calculated, wherein the distance is associated with a measure of a similarity between a first face and a second face. The activated faces are clustered into a plurality of groups based at least in part on the distances. A representative face is selected from each group resulting from the clustering and the representative faces are displayed.

Abstract: Systems and methods are disclosed for identifying objects captured by a depth camera by condensing classified image data into centroids of probability that captured objects are correctly identified entities. Output exemplars are processed to detect spatially localized clusters of non-zero probability pixels. For each cluster, a centroid is generated, generally resulting in multiple centroids for each differentiated object. Each centroid may be assigned a confidence value, indicating the likelihood that it corresponds to a true object, based on the size and shape of the cluster, as well as the probabilities of its constituent pixels.

Abstract: A system and method for predicting a file size of an image subject to transformation by scaling and a change of at least one quality-controlling parameter are described. The system receives the file size of the image before transformation, information about at least one quality-controlling parameter of the image before transformation, information about at least one quality-controlling parameter and a scaling factor to be applied to the image during transformation. A relative size prediction is calculated using the received quality-controlling parameters information and the scaling factor. The file size of the image after the transformation is finally calculated as a function of the file size of the image before the transformation and the calculated relative size prediction. Images are partitioned into classes of images, and for each class of images, an array of relative file size predictions, having at least two dimensions, is computed.

Abstract: A system and method for detecting objects and background in digital images using range information includes receiving the digital image representing a scene; identifying range information associated with the digital image and including distances of pixels in the scene from a known reference location; generating a cluster map based at least upon an analysis of the range information and the digital image, the cluster map grouping pixels of the digital image by their distances from a viewpoint; identifying objects in the digital image based at least upon an analysis of the cluster map and the digital image; and storing an indication of the identified objects in a processor-accessible memory system.

Abstract: A plurality of images inputted in an image signal input portion are divided into a plurality of regions by an image dividing portion, and a feature value in each of the plurality of regions is calculated by a feature value calculation portion and divided into a plurality of subsets by a subset generation portion. On the other hand, a cluster classifying portion classifies a plurality of clusters generated in a feature space into any one of a plurality of classes on the basis of the feature value and occurrence frequency of the feature value. And a classification criterion calculation portion calculates a criterion of classification for classifying images included in one subset on the basis of a distribution state of the feature value in the feature space of each of the images included in the one subset.

Abstract: Methods and apparatus to methods and apparatus to identify images in print advertisements are disclosed. An example method comprises computing a first image feature vector for a first presented image, comparing the first image feature vector to a second image feature vector, and when the first image feature vector matches the second image feature vector, storing printed-media information associated with the first presented image in a database record associated with the second image feature vector.

Abstract: A system and method for generating groups of cluster spines for display are provided. One or more concepts for each cluster in a set of clusters are generated. Spines are formed from at least a portion of the clusters based on the concepts. At least one spine unique from all other spines is placed. One or more spine groups are generated by positioning at least one unplaced spine in relation to one of the placed unique spines. The spine groups are displayed.

Abstract: Exemplary embodiments are described in which is performed not only a shot detection (continuous recording with a camera) and an association of several key-frames to the shots, it then being possible for a subsequent scene recognition to be based on the grouping of shots into scenes. Rather, it is observed that a scene only relates to one event in a setting. Since both can change within a shot, not every scene boundary is at the same time also a shot boundary. In addition, not every shot is short enough, so that a reliable retrieval of different picture contents is not guaranteed. Therefore, exemplary embodiments are shown which are capable of defining sub-shots so that in principle, scene and shot boundaries are also sub-shot boundaries at the same time. Sub-shots furthermore include only video pictures with a small change in picture content.

Abstract: A clustering method for high-dimensionality data includes identifying a set of nearest neighbors of a point in a multidimensional space and determining the centroid of the set of nearest neighbors, where the centroid is a member of the set of nearest neighbors. The method is then repeated using the neighbors identified around the computed centroid. In one embodiment, the method may terminate when the computed centroid becomes stationary over successive iterations. The resulting centroid may be returned as a mode of the data set. Points of the data set having common modes may be assigned to the same cluster.

Abstract: A plurality of images inputted in an image signal input portion are divided into a plurality of regions by an image dividing portion, and a feature value in each of the plurality of regions is calculated by a feature value calculation portion and divided into a plurality of subsets by a subset generation portion. On the other hand, a cluster classifying portion classifies a plurality of clusters generated in a feature space into any one of a plurality of classes on the basis of the feature value and occurrence frequency of the feature value. And a classification criterion calculation portion calculates a criterion of classification for classifying images included in one subset on the basis of a distribution state of the feature value in the feature space of each of the images included in the one subset.

Abstract: Systems and methods for clustering photos that include both time stamps and location coordinates. A two step method that first detects boundaries using time and location information independently to form a set of candidate boundaries is implemented. Such boundaries partition the set of time-ordered photos into clusters. A subset of the candidate boundaries is selected by an efficient dynamic programming procedure to optimize a cost function. Several cost functions are used to design clusterings that are coherent in space, time, or both. One set of cost functions minimizes inter-photo distances directly. A second set maximizes an information measure to select clusterings for consistency in both time and space.

Abstract: A system and method for semantic event detection in digital image content records is provided in which an event-level “Bag-of-Features” (BOF) representation is used to model events, and generic semantic events are detected in a concept space instead of an original low-level visual feature space based on the BOF representation.

Abstract: An initial value is assigned to a center point for each cluster in a plurality of clusters. Each point in a point space is assigned to a closest cluster based on the distance between the each point and the center of nearest cluster. A first-assignment value is determined for each center point using the clusters the points are assigned to. A first-assignment dynamic validity index of a current cluster configuration is evaluated. Each point in the point space is reassigned to the closest cluster based on the first-assignment value of each center. A second-assignment value is determined for the center of each cluster according to the reassigning. A second-assignment dynamic validity index is evaluated using the second-assignment values. The current cluster configuration is selected if the difference between the dynamic validity indices is less than a threshold.

Abstract: Aspects of the disclosure provide a method for crowd segmentation that can globally optimize crowd segmentation of an input image based on local information of the input image. The method can include receiving an input image of a site, initializing a plurality of hypothesis based on the input image, dividing the input image into a plurality of patches, calculating an affinity measure of one or more patches to a hypothesis based on a partial response of the patches to a whole body classifier of the hypothesis that includes a combination of weak classifiers, and optimizing assignments of the plurality of patches to the plurality of hypothesis based on the affinity measures of the plurality of patches to the plurality of hypothesis.

Abstract: Systems and methods are provided to facilitate architectural modeling. In one aspect, repetitive patterns are automatically detected and analyzed to generate modeled structural images such as building facades. In another aspect, structural symmetry is analyzed to facilitate architectural modeling and enhanced image generation.

Abstract: An information processing apparatus includes a storing unit that stores information concerning model feature points and model feature quantities at the model feature points, a first acquiring unit that acquires an input moving image, a first feature-point extracting unit that extracts input feature points for recognizing an action from the input moving image, a first feature-quantity extracting unit that extracts input feature quantities at the input feature points, a feature-quantity comparing unit that compares the input feature quantities and the model feature quantities and generates candidate corresponding feature point pairs, a posture estimating unit that removes outliers from the candidate corresponding feature point pairs, estimates postures of models on the input moving image, and obtains a recognition corresponding feature point pair group corresponding to the postures, and a recognition-result generating unit that generates a recognition result on the basis of the recognition corresponding feature

Abstract: A computer implemented method includes accessing a digital image including a plurality of faces including a first face and a second face. The computer implemented method includes identifying a plurality of identification regions of the digital image including a first identification region associated with the first face and a second identification region associated with the second face. The computer implemented method also includes assigning the digital image to a first face cluster of a plurality of face clusters when a difference between data descriptive of the first identification region and data descriptive of a face cluster identification region of the first face cluster satisfies a threshold. The computer implemented method further includes assigning the digital image to a second face cluster of the plurality of face clusters based at least partially on a probability of the second face and the first face appearing together in an image.

Abstract: A multispectral iris recognition system includes a multispectral camera adapted to acquire spatially registered iris images simultaneously in at least three wavelengths and a database adapted to store the acquired iris images. A texture analysis section identifies an area within each acquired iris image having a maximum texture at each of the wavelengths. The identified areas are combined to generate an enhanced iris image. Additionally, a visible light iris image is acquired and stored along with a set of transformation mappings in a database. The acquired visible light iris image is modeled in a texture model, which clusters textures from the acquired visible light iris image. A mapping is selected from the database for each of the clusters. The selected mappings are applied to the acquired visible light iris image to generate a Near-Infrared equivalent.

Abstract: A method for organizing images from multiple image capture devices includes automatically determining a coarse offset between image capture times recorded in a first image capture device and image capture times recorded in a second image capture device. The coarse offset is determined by a computing a correlation between image counts of images captured by the first image capture device and images captured by the first image capture device. The method also includes adjusting the image capture times of images recorded in the second image capture device by the coarse offset to produce adjusted image capture times for images captured by the second image capture device.

Abstract: A method and apparatus for processing images. Clusters of first features identified in a first image are identified. Each cluster in the clusters comprises a first group of features from the first features. A transformation for registering each cluster in the clusters with a corresponding cluster comprising a second group of features from second features identified in a second image is identified using an initial correspondence between the first features in the first image and the second features in the second image. A set of clusters from the clusters of the first features is identified using the transformation identified for each cluster. A final transformation for registering the first image with the second image is identified using the set of clusters.

Abstract: In an image conversion method, a value which reflects the mutual relationship between the classes of pixel patterns each formed from a pixel classified as one of a plurality of classes and peripheral pixels is set as a converted value corresponding to each of the plurality of classes, a pixel of interest is sequentially selected from the input image, and a pixel pattern formed from the selected pixel of interest and a predetermined number of pixels around it is classified as one of the plurality of classes in accordance with a neighboring pattern obtained based on the relationship between the value of the pixel of interest and the values of peripheral pixels located at predetermined relative positions with respect to the pixel of interest. The value of the pixel of interest is converted into a converted value set for a class to which the pixel of interest has been classified.

Abstract: A moving object detection method is provided which can accurately perform segmentation on an image including an object such as a person that moves changing shape.

Abstract: Various methods for clustering images of faces are provided. One example method may include receiving a plurality of input clusters including a first cluster and at least a second cluster, where each input cluster including at least one feature that was extracted from an image. The example method may further include determining a first medoid for the first cluster and a second medoid for the second cluster, determining an inter-cluster distance between the first medoid and the second medoid, and merging the first cluster with the second cluster to generate a merged cluster in an instance in which the inter-cluster distance is less than a distance threshold. Similar and related example methods, example apparatuses, and example computer program products are also provided.

Abstract: Presented are systems and methods for identifying information about a particular entity including acquiring electronic documents having unstructured text, that are selected based on one or more search terms from a plurality of terms related to the particular entity. Tokenizing the acquired documents to form a data matrix and then calculating a plurality of eigenvectors, using the data matrix and the transpose of the data matrix. The variance is then acquired for determining the amount of intra-clustering between the documents and then the acquired documents are clustered using some of the eigenvectors and the variance.

Abstract: An apparatus for processing image information regarding image data pieces each having retrievable information including shooting time and a shooting interval includes a grouping unit configured to group the image data pieces, which are arranged in order of the shooting time, by sequentially carrying out grouping steps that each divide or merge the image data pieces into groups according to the shooting intervals, an evaluation unit configured to calculate a score for each of the grouping steps according to one or a plurality of predetermined evaluation items, and a determination unit configured to determine a specific one of the grouping steps according to the calculated scores.

Abstract: An object position area is calculated according to a position in the space of a detected truck. A template corresponding to the object position area recorded at the previous time is then called. The template is moved to a position on a reference image Ib where the similarity is highest. Overlap rate of the object position area and the template is calculated. A decision is made whether the object is identical to that detected in the past by using the overlap rate.

Abstract: OCR errors are identified and corrected through learning. An error probability estimator is trained using ground truths to learn error probability estimation. Multiple OCR engines process a text image, and convert it into texts. The error probability estimator compares the outcomes of the multiple OCR engines for mismatches, and determines an error probability for each of the mismatches. If the error probability of a mismatch exceeds an error probability threshold, a suspect is generated and grouped together with similar suspects in a cluster. A question for the cluster is generated and rendered to a human operator for answering. The answer from the human operator is then applied to all suspects in the cluster to correct OCR errors in the resulting text. The answer is also used to further train the error probability estimator.

Abstract: Disclosed is a method of determining one or more event subsets within a plurality of images. Each image is associated with time and location data specifying the time and location of capture of the image. The method determines a time variable for each adjacent pair of images in a capture time ordered list of the plurality of images. A distance variable for each adjacent pair of images in the ordered list of images is then determined. The method determines speed data of the image capture device at the time and location of capture of each image. The ordered list of images is then partitioned into one or more event subsets on the basis of a cost function, the cost function being determined in accordance with a normalisation of the time variable and distance variable, wherein the time variable and the distance variable are weighted relative to the speed data.

Abstract: Rank ordered-based object image clustering may facilitate robust clustering of digital images. The rank order-based clustering of object images may include defining asymmetric distances between each object image and one or more other object images in a set of multiple object images using generated ordered lists. The rank order-based clustering may further include obtaining a rank order distance for each pairing of object images by normalizing the asymmetric distances of corresponding object images. The multiple object images are further clustered into object image clusters based on the rank order distances and adaptive absolute distance.

Abstract: Provided are an apparatus, medium, and method segmenting video sequences based on a topic. The apparatus may include a start-shot determination unit detecting a plurality of key-frames by using character information from video sequences including a plurality of frames to determine the detected key-frames as start-shots for each topic, and a topic list creation unit creating a topic list by using the start-shots for each topic.

Abstract: The invention segments detector input according to the time and the level of activity in different geographic regions of a locality. In one embodiment of the invention the detector input is comprised of video stream from one or more cameras to identify activity in the video. In one embodiment of the invention the detector input is comprised of sensor outputs such as RFID, pressure plates, etc. Various embodiments of the invention include identifying boundaries based on the level of activity. In embodiments of the invention, the boundaries can be used to select time dimensions. In one embodiment, by recognizing time dimensions with distinctive activity patterns, systems can better present overviews of activity over time.

Abstract: Systems and methods for incorporating restricted access to feature vectors are described. A method is described for storing and accessing feature vectors relating to individuals in digital images. One or more individuals are identified within photos, and feature vectors are generated for individuals in one or more photos. User profiles are downloaded from a database over a network. The feature vectors and downloaded user profiles are grouped. The face tagging system collects tags for the generated grouped feature vectors for generating enhanced user profiles for the individuals, and the enhanced user profiles are stored in a storage resource.

Abstract: A method of automatically identifying bone components in a medical image data set of voxels, the method comprising: a) applying a first set of one or more tests to accept voxels as belonging to seeds, b) applying a second set of one or more tests to accept seeds as bone seeds, and c) expanding the bone seeds into bone components by progressively identifying candidate bone voxels, adjacent to the bone seeds or to other previously identified bone voxels, as bone voxels, responsive to predetermined criteria which distinguish bone voxels from voxels of other body tissue.

Abstract: The image processing apparatus and method, and the program and the recording medium according to the present invention can make the coefficient vector into high precision by noise elimination or correction utilizing the mutual correlation of the divided image areas in the intermediate eigenspace, and allows relaxation of the input condition and robustness. The high correlation in the divided image areas in the intermediate eigenspace can reduce the divided image areas to be processed, and actualize reduction in processing load and enhancement of the processing speed.

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: The present invention relates to a system for detecting and classifying events during motion actions, in particular “offside” event in the football game. The system allows determining such event in a real-time and semi-automatic context, by taking into account the variability of the environmental conditions and of the dynamics of the events which can be traced back to the offside. The present invention proposes itself with a not-invasive technique, compatible with the usual course of the match.

Abstract: A plurality of images inputted in an image signal input portion are divided into a plurality of regions by an image dividing portion, and a feature value in each of the plurality of regions is calculated by a feature value calculation portion and divided into a plurality of subsets by a subset generation portion. On the other hand, a cluster classifying portion classifies a plurality of clusters generated in a feature space into any one of a plurality of classes on the basis of the feature value and occurrence frequency of the feature value. And a classification criterion calculation portion calculates a criterion of classification for classifying images included in one subset on the basis of a distribution state of the feature value in the feature space of each of the images included in the one subset.

Abstract: Automatic red-eye object classification in digital photographic images. A method for classifying a candidate red-eye object in a digital photographic image includes several acts. First, a candidate red-eye object in a digital photographic image is selected. Next, RGB pixels of the candidate red-eye object are converted into YUV pixels. Then, the YUV pixels satisfying a constraint that is a function of the YUV pixels are summed. Next, the sum is determined to be greater than or equal to a scaled version of the total number of YUV pixels in the candidate red-eye object. Finally, the candidate red-eye object is transformed into a true red-eye object.

Abstract: Techniques are disclosed for performing robust feature matching for visual search. An apparatus comprising an interface and a feature matching unit may implement these techniques. The interface receives a query feature descriptor. The feature matching unit then computes a distance between a query feature descriptor and reference feature descriptors and determines a first group of the computed distances and a second group of the computed distances in accordance with a clustering algorithm, where this second group of computed distances comprises two or more of the computed distances. The feature matching unit then determines whether the query feature descriptor matches one of the reference feature descriptors associated with a smallest one of the computed distances based on the determined first group and second group of the computed distances.

Abstract: A method for segmenting an object region of interest from an image includes detecting each corner in a captured image; acquiring edges of the object from the captured image based on the detected corners; clustering the detected corners based on corners on the acquired edges; and segmenting the image region in the captured image formed by the clustered corners as a region of interest that likely includes the object. With this method and apparatus for the method, an object region of interest may be segmented from an image without utilizing features such as an object shadow and edges.

Abstract: Methods of generating image anchor templates from low variance regions of document images of a first class are provided. The methods select a document image from the document images of the first class and align the other document images of the first class to the selected document image. Low variance regions are then determined by comparing the aligned document images and the selected document image and used to generate image anchor templates.

Abstract: Techniques are disclosed for determining anomalous trajectories of objects tracked over a sequence of video frames. In one embodiment, a symbol trajectory may be derived from observing an object moving through a scene. The symbol trajectory represents semantic concepts extracted from the trajectory of the object. Whether the symbol trajectory is anomalous may be determined, based on previously observed symbol trajectories. A user may be alerted upon determining that the symbol trajectory is anomalous.

Abstract: [Problem] To reduce omission of searched results and improve search efficiency in indexing by means of telop and another text information attached to a video. [Solving Means] A video indexing system has an index generating unit which generate an index, an index accumulating unit which accumulates the index together with attached information such as an index position, an image feature generating unit which generates an image feature, an image feature accumulating unit which accumulates the image feature together with the attached information such as the position, and an image feature checking unit which compares the image feature with past image features, and, when matched, outputs the index at the time of outputting the corresponding past image feature as a current index.

Abstract: A computer-implemented method for automatically sharing media between users is provided. Collections of images are received from different users, where each collection is associated with a particular user and the users may be associated with each other. The collections are grouped into one or more albums based on the content of the images in the collection, where each album is associated with a particular user. The albums from the different users are grouped into one or more event groups based on the content of the albums. The event groups are then shared automatically, without user intervention, between the different users based on their associations with each other and their individual sharing preferences.

Abstract: One embodiment, among others, is a method for clustering a plurality of images, wherein the plurality of images comprises faces of a plurality of individuals. The method comprises arranging the plurality of images associated with a plurality of individuals into a plurality of subgroups for each individual based on time stamps associated with the plurality of images, wherein the plurality of images are arranged according to increments of a time interval. The method further comprises determining whether adjacent subgroups are correlated and forming groups comprising correlated subgroups. Based on correlations between adjacent groups, the groups are associated with a particular individual.

Abstract: An image-capturing apparatus for capturing an image by using a solid-state image-capturing device may include a face detector configured to detect a face of a human being on the basis of an image signal in a period until an image signal obtained by image capturing is recorded on a recording medium; an expression evaluation section configured to evaluate the expression of the detected face and to compute an expression evaluation value indicating the degree to which the detected face is close to a specific expression in relation to expressions other than the specific expression; and a notification section configured to notify notification information corresponding to the computed expression evaluation value to an image-captured person.

Abstract: A system, a computer readable storage medium including instructions, and method for generating genre models used to identify genres of a document. For each document image in a set of document images that are associated with one or more genres, the document image is segmented into a plurality of tiles, wherein the tiles in the plurality of tiles are sized so that document page features are identifiable, and features of the document image and the plurality of tiles are computed. At least one genre classifier is trained to classify document images as being associated with one or more genres based on the features of the document images in the set of document images, the features of the plurality of tiles of the set of documents images, and the one or more genres associated with each document image in the set of documents images.

Abstract: A control unit (41) included in an image classification apparatus of the present invention performs a step of clustering a plurality of training images for each of a plurality of combination patterns of a plurality of feature quantities that an image has, and a step of selecting, from among the plurality of combination patterns, a classification-use combination pattern to be used in image classification, based on a result of the clustering. The clustering is performed based on degrees of similarity between the training images that have been calculated with use of the feature quantities constituting the combination patterns.

Abstract: In a method and system of service management, a radiative sensor is positioned to observe an area of interest. At least one frame of data of the area of interest is electronically acquired from the radiative sensor. The acquired frame of data is electronically processing to determine the presence or absence of at least one object in the area of interest. Based on the presence or absence of the object in the area of interest, (1) an alert is electronically caused to be generated in response to also electronically detecting another object in another area of interest, and/or (2) a timer is electronically caused to initiate or terminate counting a period of time.

Abstract: A system and method for generating groups of cluster spines for display are provided. One or more concepts for each cluster in a set of clusters are generated. Spines are formed from at least a portion of the clusters based on the concepts. At least one spine unique from all other spines is placed. One or more spine groups are generated by positioning at least one unplaced spine in relation to one of the placed unique spines. The spine groups are displayed.