Abstract: A hybrid estimator system using visual and inertial sensors for real-time pose tracking on devices with limited processing power using at least one processor, a memory, a storage and communications through a protocol and one or more than one software module for a hybrid estimator, real-time algorithm selection to process different measurements, statistical learning for these characteristics to compute the expected device computing cost of any strategy for allocating measurements to algorithms, and algorithm selection based on the statistical learning module.

Abstract: A detection device includes an image acquisition section that acquires an image that has been captured by an imaging section, and includes an image of an object, a distance information acquisition section that acquires distance information about the distance from the imaging section to the object, a feature quantity calculation section that calculates a feature quantity from the image, a learning feature quantity storage section that stores a learning feature quantity calculated by a learning process corresponding to each of a plurality of distance ranges that are set corresponding to the distance from the imaging section to the object, and a detection section that determines a distance range that corresponds to the feature quantity based on the distance information, and detects the target area based on the learning feature quantity that corresponds to the determined distance range, and the feature quantity calculated by the feature quantity calculation section.

Abstract: A handwriting recognition system converts word images on documents, such as document images of historical records, into computer searchable text. Word images (snippets) on the document are located, and have multiple word features identified. For each word image, a word feature vector is created representing multiple word features. Based on the similarity of word features (e.g., the distance between feature vectors), similar words are grouped together in clusters, and a centroid that has features most representative of words in the cluster is selected. A digitized text word is selected for each cluster based on review of a centroid in the cluster, and is assigned to all words in that cluster and is used as computer searchable text for those word images where they appear in documents. An analyst may review clusters to permit refinement of the parameters used for grouping words in clusters, including the adjustment of weights and other factors used for determining the distance between feature vectors.

Abstract: Application of inter-class and intra-class filtering, based on aggregate point-to-point distances, to vector data for purposes of filtering the vector data for purposes of pattern recognition. In some embodiments: (i) the inter-class filtering is based on Euclidean distance, in all dimensions, between vector data points in vector space; and/or (ii) the intra-class filtering is based on a distance, in all dimensions, between vector data points in vector space.

Abstract: For experience-based dynamic sequencing of a set of process options, a first process option is executed at a first stage in a first sequence for a first set of members. The sequence includes several stages of executing a subset of the process options. An experience value is determined corresponding to executing the first process option. The experience value is normalized to calculate a normalized experience value corresponding to the first process option. Using the normalized experience value in a later stage from the plurality of stages, an evaluation is made whether a first trigger threshold is satisfied in a first activation function of a second process choice. When the first trigger threshold of the first activation function being satisfied, the second process choice is included in the first sequence at the later stage.

Abstract: A method, system, and computer program product for updating an auction monitor display with highlighting or emphasis in the form of graphic data change indicators. The method commences by capturing a first set of auction variables, the auction variables pertaining to an ecommerce auction involving a plurality of auction participants in the ecommerce auction. Then, for ongoing real-time display, the method introduces a delay of a variable duration of time, the variable duration of time based on the length of time before the end of the auction. At various points within each delay period, the method captures a second set of auction variables which are compared to the first set of auction variables to identify a candidate set of changed auction variables for emphasis. The determination of the manner of emphasis is based on the changed auction variable and characteristics of the type of change of the changed auction variable.

Abstract: An object investigation and classification system may include an object test system, a data storage system, and a data processing system. The object test system may receive a command to perform at least one action with a test object, perform the at least one action with the test object, and return test information indicative of at least one percept resulting from the at least one action. The data storage system may contain an experience database containing data indicative of multiple classifications and, for each classification, at least one action that was performed with at least one previously-observed reference object having this classification, and at least one percept value that is based in whole or in part on the test information resulting from the at least one action.

Abstract: A line segmentation method which starts with determining a first starting point coordinate and generating a list of potential character widths dependent on a maximum character width stored in a database and on characteristics of the portion of the line of text corresponding to the maximum character width. The method determines a second portion of the line of text corresponding to the first starting point coordinate and the first width on the list of potential character widths. On the second portion, a classification method is applied providing a likelihood of error for the first width and a candidate character. The likelihood of error is compared with a first threshold determined by a trade-off between speed and accuracy, and if the likelihood of error corresponding to the first width is lower than the threshold value, the candidate character is selected as the character meaning that a segment is known.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for object detection are disclosed. Methods can include, for each of a plurality of locations in one or more positive images, image filters are identified, each image filter representing visual features of a location in a positive image (e.g., an image that includes a particular object). Positive location feature scores and negative location feature scores are determined for locations within images. A positive location feature score is based on a similarity between the image filter and feature values for a positive image. A negative location feature score is determined based on a similarity between the image filter and feature values for a negative image. A distinctive location is identified based on the positive and negative location feature scores, and distinguishing feature values for identifying the particular object are identified for the distinctive location.

Abstract: A method for classification of samples comprising providing a trained statistical model based upon a set of initial samples. Receiving a set of first samples and training a first statistical model base upon the first set of samples, where the first statistical model is of the same class as the trained statistical model. Receiving a set of second samples and training a second statistical model base upon the second set of samples, where the second statistical model is of the same class as the trained statistical model. The trained statistical model, the first statistical model, and the second statistical model, being independent of each other and collectively used to classify another sample.

Abstract: Image information displayed on an electronic device can be modified based at least in part upon a relative position of a user with respect to a device. Mapping, topological or other types of positional data can be used to render image content from a perspective that is consistent with a viewing angle for the current relative position of the user. As that viewing angle changes, as a result of movement of the user and/or the device, the content can be re-rendered or otherwise updated to display the image content from a perspective that reflects the change in viewing angle. Simulations of effects such as parallax and occlusions can be used with the change in perspective to provide a consistent user experience that provides a sense of three-dimensional content even when that content is rendered on a two-dimensional display. Lighting, shading and/or other effects can be used to enhance the experience.

Abstract: A method for feature transformation of a data set includes: receiving a data set including original feature samples with corresponding class labels; splitting the data set into a direction optimization set and a training set; using the direction optimization set to calculate an optimum transformation vector that maximizes inter-class separability and minimizes intra-class variance of the feature samples with respect to corresponding class labels; using the optimum transformation vector to transform the rest of the original feature samples of the data set to new feature samples with enhanced discriminative characteristics; and training a classifier using the new feature samples, wherein the method is performed by one or more processors.

Abstract: The present invention discloses a method and apparatus of determining air quality. The method comprising: determining at least one key area; acquiring a reference clear image, a training image under poor air quality and corresponding actual air quality index in at least one location of the key area; and training an air quality model of the key area based on feature extracted from the reference clear image and the training image and based on the actual air quality index. With the method and apparatus of the invention, air quality can be determined based on image.

Abstract: Methods and systems for automatic detection of landmarks in digital images and annotation of those images are disclosed. A method for detecting and annotating landmarks in digital images includes the steps of automatically assigning a tag descriptive of a landmark to one or more images in a plurality of text-associated digital images to generate a set of landmark-tagged images, learning an appearance model for the landmark from the set of landmark-tagged images, and detecting the landmark in a new digital image using the appearance model. The method can also include a step of annotating the new image with the tag descriptive of the landmark.

Abstract: Systems and methods for object detection by receiving an image; segmenting the image and identifying candidate bounding boxes which may contain an object; for each candidate bounding box, dividing the box into overlapped small patches, and extracting dense features from the patches; during a training phase, applying a learning process to learn one or more discriminative classification models to classify negative boxes and positive boxes; and during an operational phase, for a new box generated from the image, applying the learned classification model to classify whether the box contains an object.

Abstract: A computerized rating tool is described that assists a user in efficiently and consistently assigning expert ratings (i.e., labels) to a large collection of training images representing samples of a given product. The rating tool provides mechanisms for visualizing the training images in an intuitive and configurable fashion, including clustering and ordering the training images. In some embodiments, the rating tool provides an easy-to-use interface for exploring multiple types of defects represented in the data and efficiently assigning expert ratings. In other embodiments, the computer automatically assigns ratings (i.e., labels) to the individual clusters containing the large collection of digital images representing the samples.

Abstract: An improved branch-and-bound process of interval arithmetic subdivision in furtherance of computation of rigorous error bounds on integrated digital scene information for two dimensional display is provided. More particularly, a first aspect of the subject process includes pseudo-randomly subdividing an interval domain comprising a set of interval variables in furtherance of ascertaining a characteristic contribution of the interval variables of said set of interval variables to an image space comprising at least a sub-pixel area. A further aspect, either alone or in combination with the first aspect contemplates pseudo-randomly discarding a select partitioning of interval variables of a set of interval variables of a geometric function from a computed solution of an interval arithmetic branch-and-bound process.

Abstract: A method for classifying a video regarding a subjective characteristic, the method comprising: measuring a plurality of basic features (11) per frame thus obtaining a plurality of basic features measurements; creating a plurality of second-level features by pooling (12) said basic features (11) measurements using a plurality of statistics of said basic features measurements in a determined period of time of footage; creating a plurality of video features by pooling (13) said plurality of second-level features using a plurality of statistics of said second level features along the duration of the video; choosing at least one video feature of said plurality of video features for classifying a video regarding a subjective characteristic.

Abstract: A method and a system for human action recognition are provided. In the method, a plurality of training data corresponding to a plurality of gestures are received and clustered into at least one group according to similarity between the training data, where the training data represent the gestures, and a corresponding relationship between the training data and the gestures may be one-to-one or many-to-one. An image sequence of human action is captured, and a data representing the human action to be identified is obtained there from. Then, a specific group having the highest similarity with the data to be identified is selected from the groups, and a ranking result of all the training data within the specific group is obtained through a rank classifier and the data to be identified. Finally, the human action is identified as the gesture represented by the first training data in the ranking result.

Abstract: Systems and methods are disclosed for generating a probability density to estimate the probability that an event will occur in a region of interest. The methods input spatial event data comprising one or more events occurring in the region of interest along with auxiliary data related to the region of interest. The auxiliary data comprises non-event data having spatial resolution such that the probability density estimate for the region of interest is calculated based on a function of the auxiliary data and the event data. In particular, the auxiliary data is used to generate a penalty functional used in the calculation of the probability density estimate.

Abstract: Estimating focus error in an image involves a training phase and an application phase. In the training phase, an optical system is represented by a point-spread function. An image sensor array is represented by one or more wavelength sensitivity functions, one or more noise functions, and one or more spatial sampling functions. The point-spread function is applied to image patches for each of multiple defocus levels within a specified range to produce training data. Each of the images for each defocus level (i.e. focus error) is sampled using the wavelength sensitivity and spatial sampling functions. Noise is added using the noise functions. The responses from the sensor array to the training data are used to generate defocus filters for estimating focus error within the specified range. The defocus filters are then applied to the image patches of the training data and joint probability distributions of filter responses to each defocus level are characterized.

Abstract: The present invention is to provide a head detecting method for detecting the head in an image correctly at high speed.

Abstract: A hybrid estimator system using visual and inertial sensors for real-time pose tracking on devices with limited processing power using at least one processor, a memory, a storage and communications through a protocol and one or more than one software module for a hybrid estimator, real-time algorithm selection to process different measurements, statistical learning for these characteristics to compute the expected device computing cost of any strategy for allocating measurements to algorithms, and algorithm selection based on the statistical learning module.

Abstract: A system, methods, and apparatus for generating pattern recognition classifiers are disclosed. An example method includes identifying graphical objects within an image of a card object, for each identified graphical object: i) creating a bounding region encompassing the graphical object such that a border of the bounding region is located at a predetermined distance from segments of the graphical object, ii) determining pixels within the bounding region that correspond to the graphical object, iii) determining an origin of the graphical object based on an origin rule, iv) determining a text coordinate relative to the origin for each determined pixel, and v) determining a statistical probability that features are present within the graphical object, each of the features including at least one pixel having text coordinates and for each graphical object type, combining the statistical probabilities for each of the features of the identified graphical objects into a classifier data structure.

Abstract: An automatic learning method for the automatic learning of the forms of appearance of objects in images in the form of object features (28) from training images (20) for using the learned object features (28) in an image processing system comprises determining a feature contribution (30) by a training image (20) to object features (28) by weighted summation of training image features by means of linear filter operations, applied to the feature image (24), by using a weight image (32) obtained at least from an annotation image (22) and a classification image (26). This allows faster learning processes and also the learning of a greater variance of forms of appearance of objects and backgrounds, which increases the robustness of the system in its application to untrained images.

Abstract: An apparatus and method are provided for recognizing an emotion of an individual based on Action Units. The method includes receiving an input AU string including one or more AUs that represents a facial expression of an individual from an AU detector; matching the input AU string with each of a plurality of AU strings, wherein each of the plurality of AU strings includes a set of highly discriminative AUs, each representing an emotion; identifying an AU string from the plurality of AU strings that best matches the input AU string; and outputting an emotion label corresponding to the best matching AU string that indicates the emotion of the individual.

Abstract: A dictionary creation device including a blurred image generation unit which outputs a blurred image generated by performing a blurring process to a learning image together with a blur parameter indicating a blurring state of the blurred image, a patch pair generation unit which generates a restoration patch and a blurred patch as a patch pair that is composed of the patches located at the corresponding positions of the learning image and the blurred image, and a registration unit which associates the patch pair with a blur parameter corresponding to the blurred patch in the patch pair and registers them in a dictionary.

Abstract: Systems and methods for metric learning include iteratively determining feature groups of images based on its derivative norm. Corresponding metrics of the feature groups are learned by gradient descent based on an expected loss. The corresponding metrics are combined to provide an intermediate metric matrix as a sparse representation of the images. A loss function of all metric parameters corresponding to features of the intermediate metric matrix are optimized, using a processor, to learn a final metric matrix. Eigenvalues of the final metric matrix are projected onto a simplex.

Abstract: A nearest-neighbor-based distance metric learning process includes applying an exponential-based loss function to provide a smooth objective; and determining an objective and a gradient of both hinge-based and exponential-based loss function in a quadratic time of the number of instances using a computer.

Abstract: A method and system for evaluating probabilistic boosting trees is disclosed. In an embodiment, input data is received at a graphics processing unit. A weighted empirical distribution associated with each node of the probabilistic boosting tree is determined using a stack implementation. The weighted empirical distribution associated with each node is added to a total posterior distribution value.

Abstract: Methods, systems, and processor-readable media for pruning a training dictionary for use in detecting anomalous events from surveillance video. Training samples can be received, which correspond to normal events. A dictionary can then be constructed, which includes two or more classes of normal events from the training samples. Sparse codes are then generated for selected training samples with respect to the dictionary derived from the two or more classes of normal events. The size of the dictionary can then be reduced by removing redundant dictionary columns from the dictionary via analysis of the sparse codes. The dictionary is then optimized to yield a low reconstruction error and a high-interclass discriminability.

Abstract: The techniques and systems described herein create and train a multiple clustered instance learning (MCIL) model based on image features and patterns extracted from training images. The techniques and systems separate each of the training images into a plurality of instances (or patches), and then learn multiple instance-level classifiers based on the extracted image features. The instance-level classifiers are then integrated into the MCIL model so that the MCIL model, when applied to unclassified images, can perform image-level classification, patch-level clustering, and pixel-level segmentation.

Abstract: Content-based retrieval of digital pathology images (DPI) is a fundamental component in an intelligent DPI processing and management system. The fundamental procedure of the retrieval is evaluating the similarity between the query image and every image in the database with some distance function, and sorting of the latter based on their distances to the query. A novel approach to optimally combine a set of existing distance functions into a stronger distance that is suitable for retrieving DPI in a way respecting human perception of image similarity is described herein.

Abstract: Systems and methods for object detection by receiving an image; segmenting the image and identifying candidate bounding boxes which may contain an object; for each candidate bounding box, dividing the box into overlapped small patches, and extracting dense features from the patches; during a training phase, applying a learning process to learn one or more discriminative classification models to classify negative boxes and positive boxes; and during an operational phase, for a new box generated from the image, applying the learned classification model to classify whether the box contains an object.

Abstract: An image processing apparatus includes: a gradient information calculating unit that calculates gradient information of each of pixels, based on pixel values of an intraluminal image; a closed region creating unit that, based on the gradient information, creates a closed region satisfying a condition where the closed region does not include, on the inside thereof, any pixel of which the gradient strength is equal to or higher than a predetermined value, and also, the boundary of the closed region does not curve toward the interior of the closed region, with a curvature equal to or larger than a predetermined value; and an abnormal part detecting unit that detects an abnormal part from the inside of the closed region.

Abstract: Quantification of metric or functional parameters often requires image segmentation. A crucial part of such method is the model of the surface characteristics of the object of interest (features), which drives the deformable surface towards the object boundary in the image. According to the present invention, sections of the mesh are assigned to different classes for different features. According to the present invention, the assignment of mesh sections to the classes is adapted by using actual feature information from the unseen image. Advantageously, this allows for an adaptation of the feature category to which the mesh section is assigned and thereby allows an improved segmentation of the object.

Abstract: Techniques for generating cross-modality semantic classifiers and using those cross-modality semantic classifiers for ground level photo geo-location using digital elevation are provided. In one aspect, a method for generating cross-modality semantic classifiers is provided. The method includes the steps of: (a) using Geographic Information Service (GIS) data to label satellite images; (b) using the satellite images labeled with the GIS data as training data to generate semantic classifiers for a satellite modality; (c) using the GIS data to label Global Positioning System (GPS) tagged ground level photos; (d) using the GPS tagged ground level photos labeled with the GIS data as training data to generate semantic classifiers for a ground level photo modality, wherein the semantic classifiers for the satellite modality and the ground level photo modality are the cross-modality semantic classifiers.

Abstract: A signal processing device includes a preprocessing unit interpolating a G color component to positions of a pixel of interest and a pixel having the same color component as the pixel of interest so as to produce a first G interpolation signal; a proximity G pixel G color difference and R/B pixel producing unit producing a first R-G/B-G color difference signal on the positions of the pixel of interest and the pixel having the same color component, producing a second R-G/B-G color difference signal on a position of a proximity G pixel, and interpolating the R/B color component to the position of the proximity G pixel; a G color difference re-constitution processing unit re-constituting a third R-G/B-G color difference signal on the position of the pixel of interest; and a G color difference interpolation processing unit interpolating an R-G/B-G color difference signal to a position of a predetermined pixel.

Abstract: A system for detecting a global harmful video includes: a video determination policy generation unit for determining harmfulness of learning video segments from video learning information to analyze occurrence information of harmful learning video segments, and generating a global harmfulness determination policy based on the occurrence information; and a video determination policy execution unit for determining harmfulness of input video segments from information of an input video to analyze occurrence information of harmful input video segments, and determining whether the input video is harmful or not based on the occurrence information of the harmful input video segments and the global harmfulness determination policy.

Abstract: The aesthetic quality of a picture is automatically inferred using visual content as a machine learning problem using, for example, a peer-rated, on-line photo sharing Website as data source. Certain visual features of images are extracted based on the intuition that they can discriminate between aesthetically pleasing and displeasing images. A one-dimensional support vector machine is used to identify features that have noticeable correlation with the community-based aesthetics ratings. Automated classifiers are constructed using the support vector machines and classification trees, with a simple feature selection heuristic being applied to eliminate irrelevant features. Linear regression on polynomial terms of the features is also applied to infer numerical aesthetics ratings.

Abstract: An information processing device includes a first calculation unit which calculates a score of each sample image including a positive image in which an object as an identification object is present and a negative image in which the object as the identification object is not present, for each weak identifier of an identifier including a plurality of weak identifiers, a second calculation unit which calculates the number of scores when the negative image is processed, which are scores less than a minimum score among scores when the positive image is processed; and an realignment unit which realigns the weak identifiers in order from a weak identifier in which the number calculated by the second calculation unit is a maximum.

Abstract: An image recognition information attaching apparatus includes a retrieving unit that retrieves image information on a per piece basis of identification information, from the image information having the identification information associated thereto in advance, a generator unit that generates feature information from the image information retrieved by the retrieving unit, and a learning unit that provides a learning result by learning a relation between the feature information generated by the generator unit and the identification information of the image information corresponding to the feature information, using a stochastic model including a mixture of a plurality of probability distributions.

Abstract: A learning unit 4 generates a function table indicating the relationship between the class number and position information of an object and the probability of appearance of the object for each small area image pattern of a code book, calculates a sharing matrix indicating the commonality of a feature amount between the classes, makes a tree diagram in which the classes with a similar feature amount are clustered, and calculates the weight of each node in the tree diagram for each small area image pattern. The recognition processing unit 7 compares image data captured by a camera 2 with the code book, selects the closest small area image pattern among a plurality of small area image patterns, extracts the class related to the node with the smallest weight among the nodes with a weight equal to or greater than a threshold value for the selected small area image pattern, and votes the position information of the small area image pattern for the class, thereby recognizing the object.

Abstract: An apparatus and method for detection of abnormal motion in video stream, having a training phase for defining normal motion and a detection phase for detecting abnormal motions in the video stream is provided. Motion is detected according to motion vectors and motion features extracted from video frames.

Abstract: A method Pan-sharpens a single panchromatic (Pan) image and a single multispectral (MS) image. A wavelet transform is applied to the Pan image and the MS image to obtain a wavelet transformed Pan image and a wavelet transformed MS image. Features, in the form of vectors, are extracted from the wavelet transformed Pan image and the wavelet transformed MS image. The features are separated into features without missing values and features with missing values. A dictionary is learned from features without missing values and used to predict the values for the features with the missing values. After the predicting, the features of the low frequency wavelet coefficients and the high frequency coefficients to form a fused wavelet coefficient map, and an inverse wavelet transform is applied to the fused wavelet coefficient map to obtain a fused MS image.

Abstract: A computer-implemented method for learning a tight frame includes acquiring undersampled k-space data over a time period using an interleaved process. An average of the undersampled k-space data is determined and a reference image is generated based on the average of the undersampled k-space data. Next, a tight frame operator is determined based on the reference image. Then, a reconstructed image data is generated from the undersampled k-space data via a sparse reconstruction which utilizes the tight frame operator.

Abstract: The invention includes a method for recognizing shapes using a preprocessing mechanism that decomposes a source signal into basic components called atoms and a recognition mechanism that is based on the result of the decomposition performed by the preprocessing mechanism. In the method, the preprocessing mechanism includes at least one learning phase culminating in a set of signals called kernels, the kernels being adapted to minimize a cost function representing the capacity of the kernels to correctly reconstruct the signals from the database while guaranteeing a sparse decomposition of the source signal while using a database of signals representative of the source to be processed and a coding phase for decomposing the source signal into atoms, the atoms being generated by shifting of the kernels according to their index, each of the atoms being associated with a decomposition coefficient. The invention also includes a shape recognition system for implementing the method.

Abstract: An image storage processing apparatus includes an image acquisition means for acquiring taken image data imaged at an imaging apparatus unit, a bio-information acquisition means for acquiring bio-information of a user of the imaging apparatus unit at the time of imaging the taken image data acquired by the image acquisition means, a subject information acquisition means for acquiring subject information as an image analysis result of the taken image data acquired by the image acquisition means and a storage processing means for performing processing of recording the taken image data acquired by the image acquisition means, the bio-information acquired by the bio-information acquisition means and the subject information acquired by the subject information acquisition means in a recording medium in a state in which they are associated with one another.

Abstract: Approaches for deciding what individuals in a population of visual system “neurons” are looking for using sparse overcomplete feature dictionaries are provided. A sparse overcomplete feature dictionary may be learned for an image dataset and a local sparse representation of the image dataset may be built using the learned feature dictionary. A local maximum pooling operation may be applied on the local sparse representation to produce a translation-tolerant representation of the image dataset. An object may then be classified and/or clustered within the translation-tolerant representation of the image dataset using a supervised classification algorithm and/or an unsupervised clustering algorithm.

Abstract: Implementations and techniques for measuring quality of experience associated with a mobile device are generally disclosed.