Abstract: Disclosed are an apparatus and a method of detecting a human component from an input image. The apparatus includes a training database (DB) to store positive and negative samples of a human component, an image processor to calculate a difference image for the input image, a sub-window processor to extract a feature population from a difference image that is calculated by the image processor for the positive and negative samples of a predetermined human component stored in the training DB, and a human classifier to detect a human component corresponding to a human component model using the human component model that is learned from the feature population.

Abstract: Often, information regarding images and/or other data may be incomplete. For example, an image may have depth information associated with a portion of the image, but not the entire image. It may be advantageous to extrapolate the values from the known points to the entire image. Accordingly, a dimensional array representing an image (or other data) may be traversed one or more times to generate an interpolated table. The interpolated table may comprise pixels of unknown value, where a pixel of unknown value may be associated with a number of pixels with known values. In this way, values for pixels may be specified based upon values of pixels having known values.

Abstract: A system and method are disclosed for learning a random multinomial logit (RML) classifier and applying the RML classifier for scene segmentation. The system includes an image textonization module, a feature selection module and a RML classifier. The image textonization module is configured to receive an image training set with the objects of the images being pre-labeled. The image textonization module is further configured to generate corresponding texton images from the image training set. The feature selection module is configured to randomly select one or more texture-layout features from the texton images. The RML classifier comprises multiple multinomial logistic regression models. The RML classifier is configured to learn each multinomial logistic regression model using the selected texture-layout features. The RML classifier is further configured to apply the learned regression models to an input image for scene segmentation.

Abstract: Embodiments of the present invention relate to a method and a system for personalized advertisement push based on user interest learning. The method may include: obtaining multiple user interest models through multitask sorting learning; extracting an object of interest in a video according to the user interest models; and extracting multiple visual features of the object of interest, and according to the visual features, retrieving related advertising information in an advertisement database. Through the method and the system provided in embodiments of the present invention, a push advertisement may be closely relevant to the content of the video, thereby meeting personalized requirements of a user to a certain extent and achieving personalized advertisement push.

Abstract: Methods for identifying and quantifying recurrent and deterministic patterns in digital images are provided. The methods, which are based on Recurrence Quantification Analysis (RQA), generate similarity or dissimilarity distance matrices for digital images that may be used to calculate a variety of quantitative characteristics for the images. Also provided are methods for identifying and imaging spatial distributions of time variable signals generated from dynamic systems. In these methods a time variable signal is recorded for a plurality of area or volume elements into which a dynamic system has been sectioned and RQA is used to calculate one or more RQA variables for each of the area or volume elements, which may then be used to generate a two or three dimensional image displaying the spatial distribution of the RQA variables across the system.

Abstract: The present invention discloses a system and method of transforming a sample of content data by utilizing known samples in a learning best to best determine coefficients for a linear combination of non-linear filter functions and applying the coefficients to the content data in an operational phase.

Abstract: A computer implemented method for extracting meaningful text from a document of unknown or unspecified format. In a particular embodiment, the method includes reading the document, thereby to extract raw encoded text, analysing the raw encoded text, thereby to identify one or more text chunks, and for a given chunk, performing compression identification analysis to determine whether compression is likely and, in the event that compression. The method can further include performing a decompression process, performing an encoding identification process thereby to identify a likely character encoding protocol, and converting the chunk using the identified likely character encoding protocol, thereby to output the chunk as readable text.

Abstract: A cognitive control framework system for automatically controlling execution of an application program having a graphical user interface includes a recording component, an execution scenario script, and a playback component. The recording component is adapted to capture user input data and images displayed by the graphical user interface during a recording phase of execution of the application program, and to analyze the captured user input data and displayed images to generate an execution scenario (script) during the recording phase. The execution scenario may be written in a selected high level language (e.g., XML). The playback component is adapted to generate simulated user input data based on the execution scenario during a playback phase of execution of the application program.

Abstract: A multi-class classifier is trained by selecting a query image from a set of active images based on a membership probability determined by the classifier, wherein the active images are unlabeled. A sample image is selected from a set of training image based on the membership probability of the query image, wherein the training images are labeled. The query image and the sample images are displayed to a user on an output device. A response from the user is obtained with an input device, wherein the response is a yes-match or a no-match. The query image with the label of the sample image is added to the training set if the yes-match is obtained, and otherwise repeating the selecting, displaying, and obtaining steps until a predetermined number of no-match is reached to obtain the multi-class classifier.

Abstract: A normal image representing a normal structure of a predetermined structure in an input medical image is generated with higher accuracy. Further, an abnormal component in the input medical image is separated with higher accuracy. A supervised learned filtering unit inputs an input image representing a predetermined structure to a supervised learned filter to generate an image representing a normal structure of the predetermined structure. The supervised learned filter is obtained through a learning process using supervisor images, each representing a normal structure of the predetermined structure in a subject (individual), and corresponding training images, each containing an abnormal component in the corresponding subject (individual). Further, a difference processing unit separates an abnormal component in the input image by calculating a difference between the input image and the image representing the normal structure.

Abstract: Embodiments described herein facilitate or enhance the implementation of image recognition processes which can perform recognition on images to identify objects and/or faces by class or by people.

Abstract: In a first exemplary embodiment of the present invention, an automated, computerized method is provided for processing an image. According to a feature of the present invention, the method comprises the steps of providing an image file depicting an image, in a computer memory, identifying a dominant region of single reflectance in the image and segregating the image into intrinsic images as a function of the dominant region of single reflectance.

Abstract: Described is a technology in which a low resolution image is processed into a high-resolution image, including by a two interpolation passes. In the first pass, missing in-block pixels, which are the pixels within a block formed by four neighboring original pixels, are given values by gradient diffusion based upon interpolation of the surrounding original pixels. In the second interpolation pass, missing on-block pixels, which are the pixels on a block edge formed by two adjacent original pixels, are given values by gradient diffusion based upon interpolation of the values of those adjacent original pixels and the previously interpolated values of their adjacent in-block pixels. Also described is a difference projection process that varies the values of the interpolated pixels according to a computed difference projection.

Abstract: In image processing for embedding additional information in an image, degradation of image quality due to consecution of the same code is suppressed. Hence, determination whether or not the same code consecutively appears in additional information is performed, an embedding condition for embedding a code in an image controlled based on the code consecution, and the additional information is embedded in the image in accordance with the embedding condition.

Abstract: An image processing apparatus for converting a first image into a second image having higher image quality than the first image. The image processing apparatus including first pixel value extracting means for extracting plural pixel values within the first image and estimate noise amount arithmetically operating means for obtaining estimate noise amounts for the plural pixel values. The image processing apparatus also including processing coefficient generating means for generating second processing coefficients in accordance with an arithmetic operation for first processing coefficients and the estimate noise amounts, second pixel value extracting means for extracting a plurality of pixel values, and predicting means for generating a pixel value of the pixel of interest.

Abstract: Described is a technology by which online recognition of handwritten input data is combined with offline recognition and processing to obtain a combined recognition result. In general, the combination improves overall recognition accuracy. In one aspect, online and offline recognition is separately performed to obtain online and offline character-level recognition scores for candidates (hypotheses). A statistical analysis-based combination algorithm, an AdaBoost algorithm, and/or a neural network-based combination may determine a combination function to combine the scores to produce a result set of one or more results. Online and offline radical-level recognition may be performed. For example, a HMM recognizer may generate online radical scores used to build a radical graph, which is then rescored using the offline radical recognition scores. Paths in the rescored graph are then searched to provide the combined recognition result, e.g., corresponding to the path with the highest score.

Abstract: Disclosed are apparatus and methods for detecting whether a video is adult or non-adult. In certain embodiments, a learning system is operable to generate one or more models for adult video detection. The model is generated based on a large set of known videos that have been defined as adult or non-adult. Adult detection is then based on this adult detection model. This adult detection model may be applied to selected key frames of an unknown video. In certain implementations, these key frames can be selected from the frames of the unknown video. Each key frame may generally correspond to a frame that contains key portions that are likely relevant for detecting pornographic or adult aspects of the unknown video. By way of examples, key frames may include moving objects, skin, people, etc. In alternative embodiments, a video is not divided into key frames and all frames are analyzed by a learning system to generate a model, as well as by an adult detection system based on such model.

Abstract: Techniques are disclosed for learning and modeling a background for a complex and/or dynamic scene over a period of observations without supervision. A background/foreground component of a computer vision engine may be configured to model a scene using an array of ART networks. The ART networks learn the regularity and periodicity of the scene by observing the scene over a period of time. Thus, the ART networks allow the computer vision engine to model complex and dynamic scene backgrounds in video.

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: The present invention relates to a data conversion apparatus and a learning device in which an image can be converted into a higher-quality image. A class-tap generating circuit (2) and a predictive-tap generating circuit (3) respectively generate from an SD image a class tap for classifying a specified pixel of an HD image and a predictive tap for determining the specified pixel. A classification circuit (4) classifies the specified pixel based on the class tap. A coefficient RAM (5) obtains a tap coefficient for the class of the specified pixel from tap coefficients obtained by using supervisor pixels and learner pixels and by learning the relationship between the supervisor pixels and the learner pixels for each of at least one class while applying weighting to the supervisor pixels and the learner pixels based on the activity of the predictive tap generated from the learner pixels. A predictive-computation circuit (6) determines the specified pixel by using the tap coefficient and the predictive tap.

Abstract: The present invention relates to an image processing apparatus, method, and program that can extract an object from an input image more easily and more accurately. A face detecting unit (31) detects a face from an input image, a mask area setting unit (33) sets a mask area which masks a person in the input image based on a position of the face detected by the face detecting unit (31), a background model updating unit (51) updates a background image by learning areas other than the mask area in the input image as the background image, and a separating unit (54) separates the input image into the background image and a foreground image which is an area of the person in the input image based on the background image updated by the background model updating unit (51) and the input image. The present invention can be applied to, for example, an image processing apparatus that extracts a person from an input image.

Abstract: For monitoring an image transformation such as aspect ratio conversion, an image feature is defined by identifying a position in the image having a local spatial maximum value and then identifying four other positions in the image having local spatial minimum values such that the four minimum value positions surround the position of the maximum, a first pair of the minimums lie on a first line passing through maximum and a second pair of the minimums lie on a second line passing through the maximum.

Abstract: An image processing device for converting an input image into an output image whose blur is reduced. The image processing device includes a first image extracting section that extracts a plurality of pixels composed of a pixel of the input image corresponding to a pixel of interest in the output image and predetermined pixels surrounding the pixel of the input image. A first feature quantity calculating section calculates a first feature quantity from the plurality of pixels extracted by the first image extracting section. A second pixel extracting section extracts a plurality of pixels composed of the pixel of interest and predetermined pixels surrounding the pixel of interest from the input image. A predicting section then generates a pixel value of the pixel of interest.

Abstract: A method for automatically selecting and organizing a subset of photos from a set of photos provided by a user, who has an account on at least one social network providing some context, for creating a summarized photo album with a storytelling structure. The method comprises: arranging the set of photos into a three level hierarchy, acts, scenes and shots; checking whether photos are photos with people or not; obtaining an aesthetic measure of the photos; creating and ranking face clusters; selecting the most aesthetic photo of each face cluster; selecting photos with people until complete a predefined number of photos of the summarized album picking the ones which optimize the function: Of(C,C*,S,CSN)=?fAf(S)??fd(HCharacter(S)HCharacter(C?CSN)??fd(HAct(S),HAct(C*)); and then selecting non-people photos which minimize the following function Oa: Oa(C*,S)=d(HAct,(S),HAct(C*)); ordering all the selected photos in chronological order; and finally discarding all the photos which have not been selected.

Abstract: Systems and methods are disclosed to recognize human action from one or more video frames by performing 3D convolutions to capture motion information encoded in multiple adjacent frames and extracting features from spatial and temporal dimensions therefrom; generating multiple channels of information from the video frames, combining information from all channels to obtain a feature representation for a 3D CNN model; and applying the 3D CNN model to recognize human actions.

Abstract: Down-sampling of an image may be performed in the DCT domain. A multiple layered network is used to select transform matrices for down-sampling a DCT image of size M×N to a DCT image of size I×J. A spatial domain down-sampling method is selected and applied to the DCT image to produce a down-sampled DCT reference image. A learning with forgetting algorithm is used to apply a decay to the elements of the transform matrix and select a transform matrices which solve an optimization problem. The optimization problem is a function of the visual quality of images obtained using the transform matrices and the computational complexity associated with using the transform matrices. The visual quality is a measure of the difference between the down-sampled DCT image obtained using the transform matrices and the visual quality of the DCT reference image obtained using a spatial domain down-sampling method.

Abstract: An image processing device includes: a smoothing section configured to extract a smoothing tap and smooth a target image on the basis of pixel values within the tap, the smoothing tap being of variable size and including plural pixels centered on each target pixel of the image; a class tap extracting section configured to extract a class tap including plural pixels centered on each target pixel in the smoothed image; a class code determining section configured to generate a code corresponding to a characteristic of variation of pixel values within the class tap, and determine a class code including a size of the smoothing tap and the code; and a pixel value computing section configured to read tap coefficients corresponding to the determined class code, and multiply pixel values forming a prediction tap extracted from the smoothed image, by the tap coefficients to calculate pixel values of a processed image.

Abstract: A system for intelligent goal-directed search in large volume visual imagery using a cognitive-neural subsystem is disclosed. The system comprises an imager, a display, a display processor, a cognitive-neural subsystem, a system controller, and operator controls. The cognitive-neural subsystem comprises a cognitive module, a neural module, and an adaptation module. The cognitive module is configured to extract a set of regions of interest from the image using a cognitive algorithm. The neural module is configured to refine the set of regions of interest using a neural processing algorithm. The adaptation module is configured to bias the cognitive algorithm with information gained from the neural module to improve future searches. The system functions in a plurality of operating modes, including: batch mode, semi-interactive mode, real-time mode, and roaming mode.

Abstract: A novel, linear modeling method to model a face recognition algorithm based on the match scores produced by the algorithm. Starting with a distance matrix representing the pair-wise match scores between face images, an iterative stress minimization algorithm is used to obtain an embedding of the distance matrix in a low-dimensional space. A linear transformation used to project new face images into the model space is divided into two sub-transformations: a rigid transformation of face images obtained through principal component analysis of face images and a non-rigid transformation responsible for preserving pair-wise distance relationships between face images. Also provided is a linear indexing method using the linear modeling method to perform the binning or algorithm-specific indexing task with little overhead.

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: Images are retrieved and ranked according to relevance to attributes of a multi-attribute query through training image attribute detectors for different attributes annotated in a training dataset. Pair-wise correlations are learned between pairs of the annotated attributes from the training dataset of images. Image datasets may then be searched via the trained attribute detectors for images comprising attributes in a multi-attribute query, wherein images are retrieved from the searching that each comprise one or more of the query attributes and also in response to information from the trained attribute detectors corresponding to attributes that are not a part of the query but are relevant to the query attributes as a function of the learned plurality of pair-wise correlations. The retrieved images are ranked as a function of respective total numbers of attributes within the query subset attributes.

Abstract: Various technologies and techniques are disclosed that improve handwriting recognition operations. Handwritten input is received in training mode and run through several base recognizers to generate several alternate lists. The alternate lists are unioned together into a combined alternate list. If the correct result is in the combined list, each correct/incorrect alternate pair is used to generate training patterns. The weights associated with the alternate pairs are stored. At runtime, the combined alternate list is generated just as training time. The trained comparator-net can be used to compare any two alternates in the combined list. A template matching base recognizer is used with one or more neural network base recognizers to improve recognition operations. The system provides comparator-net and reorder-net processes trained on print and cursive data, and ones that have been trained on cursive-only data. The respective comparator-net and reorder-net processes are used accordingly.

Abstract: An intrusion prevention/detection system filter (IPS filter) performance evaluation is provided. The performance evaluation is performed at both the security center and at the customer sites to derive a base confidence score and local confidence scores. Existence of new vulnerability is disclosed and its attributes are used in the generation of new IPS filter or updates. The generated IPS filter is first tested to determine its base confidence score from test confidence attributes prior to deploying it to a customer site. A deep security manager and deep security agent, at the customer site, collect local confidence attributes that are used for determining the local confidence score. The local confidence score and the base confidence score are aggregated to form a global confidence score. The local and global confidence scores are then compared to deployment thresholds to determine whether the IPS filter should be deployed in prevention or detection mode or sent back to the security center for improvement.

Abstract: Local models learned from anomaly detection are used to rank detected anomalies. The local models include image feature values extracted from an image field of video image data with respect to different predefined spatial and temporal local units, wherein anomaly results are determined by failures to fit to applied anomaly detection module local models. Image features values extracted from the image field local units associated with anomaly results are normalized, and image feature values extracted from the image field local units are clustered. Weights for anomaly results are learned as a function of the relations of the normalized extracted image feature values to the clustered image feature values. The normalized values are multiplied by the learned weights to generate ranking values to rank the anomalies.

Abstract: A prediction calculation unit calculates a pixel value of a pixel of interest for each color component by a calculation of a learned predictive coefficient and a predictive tap, and outputs an output image including the pixel value of the pixel of interest of each color component. For example, the present technology can be applied to an image processing apparatus.

Abstract: There is provided an image processing apparatus including a model-based processing unit that executes model-based processing for converting resolution and converting an image on the basis of a camera model and a predetermined model having aligning, with respect to a high-resolution image output one frame before, and a prediction operation unit that performs a prediction operation on a pixel value of a high-resolution image to be output, on the basis of parameters stored in advance, an observed low-resolution image that is an input low-resolution image, and an image obtained by executing the model-based processing.

Abstract: A system and method for performing multi-image training for pattern recognition and registration is provided. A machine vision system first obtains N training images of the scene. Each of the N images is used as a baseline image and the N?1 images are registered to the baseline. Features that represent a set of corresponding image features are added to the model. The feature to be added to the model may comprise an average of the features from each of the images in which the feature appears. The process continues until every feature that meets a threshold requirement is accounted for. The model that results from the present invention represents those stable features that are found in at least the threshold number of the N training images. The model may then be used to train an alignment/inspection tool with the set of features.

Abstract: An extraction-pattern storing unit stores therein information related to a plurality of different extraction patterns for extracting a predetermined number of pixels from pixels surrounding a pixel that is a target for detecting a face part image. A face-part-image detecting unit extracts a pixel using the different extraction patterns stored in the extraction-pattern storing unit, and detects the face part image included in an image using a feature amount of an extracted pixel. A face-image detecting unit detects a face image from the image based on the face part image detected by the face-part-image detecting unit.

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: A sketch generating system and a method for generating a sketch based on an image are provided. The system includes: a sketch database and a generating subsystem. The sketch database stores local image samples and corresponding local sketch units in different categories. The generating subsystem extracts geometrical features from an input image, retrieves local image units from the input image according to the geometrical features; as to each local image unit retrieved, searches the sketch database for a local sketch unit corresponding to a local image sample having a largest similarity value with the local image unit, and combines all local sketch units found to form one sketch.

Abstract: A tracker component for a computer vision engine of a machine-learning based behavior-recognition system is disclosed. The behavior-recognition system may be configured to learn, identify, and recognize patterns of behavior by observing a video stream (i.e., a sequence of individual video frames). The tracker component may be configured to track objects depicted in the sequence of video frames and to generate, search, match, and update computational models of such objects.

Abstract: The invention concerns image data processing, through noise reduction comprising the following steps: associating a learning zone (ZA) with a reference point (Pref) of the image (IM); for each variable point (PC, PC?) of the learning zone, evaluating a distance (d,d?,) between: values of points in a first window (f1) of the image, centered on the reference point, and values of points in a second window (f2, f2,), of similar format as the format of the first window and centered on the variable point; repeating said distance calculation for all the points of the learning zone as successive variable points and estimating an average value to assign to the reference point, said average being weighted on the basis of the distances evaluated for each variable point.

Abstract: A sequence layer in a machine-learning engine configured to learn from the observations of a computer vision engine. In one embodiment, the machine-learning engine uses the voting experts to segment adaptive resonance theory (ART) network label sequences for different objects observed in a scene. The sequence layer may be configured to observe the ART label sequences and incrementally build, update, and trim, and reorganize an ngram trie for those label sequences. The sequence layer computes the entropies for the nodes in the ngram trie and determines a sliding window length and vote count parameters. Once determined, the sequence layer may segment newly observed sequences to estimate the primitive events observed in the scene as well as issue alerts for inter-sequence and intra-sequence anomalies.

Abstract: A method and system for creating a form template for a form are disclosed. The method comprises analyzing an image of a form to detect object demarcations in the form. The method also comprises classifying the object demarcations into one of a plurality of predefined object categories and processing each object demarcation based on the object category into which it has been classified, thereby to create the form template automatically.

Abstract: Down-sampling of an image may be performed in the DCT domain. Transform matrices are obtained for down-sampling a DCT image of size M×N to a down-sampled DCT image of size I×J. The transform matrices may be used to down-sample the DCT image directly in the DCT domain. A spatial domain down-sampling method is selected and applied to the DCT image to produce a down-sampled DCT reference image. The transform matrices are selected by solving an optimization problem, leading to transform matrices which achieve a desired trade-off between the visual quality of images obtained using the transform matrices and the computational complexity associated with using the transform matrices. The visual quality is a measure of the difference between the down-sampled DCT image obtained using the transform matrices and the visual quality of the DCT reference image obtained using a spatial domain down-sampling method.

Abstract: An image-based pattern recognizer and a method and apparatus for making such a pattern recognizer are disclosed. By employing positional coding, the meaning of any feature present in an image can be defined implicitly in space. The pattern recognizer can be a neural network including a plurality of stages of observers. The observers are configured to cooperate to identify the presence of features in the input image and to recognize a pattern in the input image based on the features. Each of the observers includes a plurality of neurons. The input image includes a plurality of units, and each of the observers is configured to generate a separate output set that includes zero or more coordinates of such units.

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: An information processing device includes a learning unit that performs, using an action performed by an object and an observation value of an image as learning data, learning of a separation learning model that includes a background model that is a model of the background of the image and one or more foreground model(s) that is a model of a foreground of the image, which can move on the background, in which the background model includes a background appearance model indicating the appearance of the background, and at least one among the one or more foreground model(s) includes a transition probability, with which a state corresponding to the position of the foreground on the background is transitioned by an action performed by the object corresponding to the foreground, for each action, and a foreground appearance model indicating the appearance of the foreground.

Abstract: Techniques are disclosed for visually conveying a trajectory map. The trajectory map provides users with a visualization of data observed by a machine-learning engine of a behavior recognition system. Further, the visualization may provide an interface used to guide system behavior. For example, the interface may be used to specify that the behavior recognition system should alert (or not alert) when a particular trajectory is observed to occur.

Abstract: A mobile platform efficiently processes sensor data, including image data, using distributed processing in which latency sensitive operations are performed on the mobile platform, while latency insensitive, but computationally intensive operations are performed on a remote server. The mobile platform acquires sensor data, such as image data, and determines whether there is a trigger event to transmit the sensor data to the server. The trigger event may be a change in the sensor data relative to previously acquired sensor data, e.g., a scene change in an image. When a change is present, the sensor data may be transmitted to the server for processing. The server processes the sensor data and returns information related to the sensor data, such as identification of an object in an image or a reference image or model. The mobile platform may then perform reference based tracking using the identified object or reference image or model.