462 results about "Eigenvector computation" patented technology

In realizing a user recognition system and a method thereof, a user feature vector is extracted from an input facial image, and at least one cluster is generated and a user feature template is enrolled. The cluster includes the feature vector as a member of it. When a user facial image is inputted, a user feature vector is extracted from the image, and a similarity between the feature vector and the user feature template is calculated. When the similarity is greater than a predetermined threshold value, a user of the user feature template is recognized as the user of the input image.

A method and system for detecting and tracking objects near a vehicle using a three dimensional laser rangefinder. The method receives points from the laser rangefinder, where the points represent locations in space where the rangefinder senses that some object exists. An algorithm first estimates the location of a ground plane, based on a previous ground plane location, data from onboard sensors, and an eigenvector calculation applied to the point data. Next, a plan view occupancy map and elevation map are computed for stationary objects, based on point data in relation to the ground plane. Finally, dynamic objects are detected and tracked, sensing objects which are moving, such as other vehicles, pedestrians, and animals. The output of the method is a set of stationary and dynamic objects, including their shape, range, and velocity. This output can be used by downstream applications.

An object recognition apparatus includes an input section for inputting a first image set including a predetermined number of images including a first image representing a first target object using a first attribute and a second image representing the first target object using a second attribute different from the first attribute, a feature vector calculation section for calculating a first feature vector in a feature space having at least one filter output value as a vector component, the at least one filter output value being obtained by applying at least one predetermined image filter to at least one predetermined position in the predetermined number of images in the first image set, and a judgment section for judging whether or not the first target object is categorized into a specific category based on a relationship between the first feature vector and predetermined discriminant parameters.

The invention discloses a face identification method and apparatus. The method includes the steps of extracting face characteristic vectors of a face image to be identified by using a convolution neural network model. The face characteristic vectors comprise image characteristics and facial attribute characteristics. Matching degree between the face image to be identified and face images in a face template database is calculated according to the face characteristic vectors, and the identification result of the face image to be identified is determined according to the matching degree. During a face identification process, the face identification accuracy is effectively improved based on comprehensive comparison of the face image characteristics and facial attribute characteristics.

The invention discloses an internet information product recommending method based on matrix decomposition. The method comprises the following steps of: 1) obtaining the user scoring record to the information product; 2) obtaining the social relationship record between internet users; 3) respectively building a scoring matrix and a social matrix according to the types of a target user and a target product; 4) learning a user feature vector and a product feature vector through a matrix decomposition technology; 5) calculating the scores of different products scored by the target user according to the feature vectors, so as to recommending the favorite products of the user according to the scores. In the method, analysis on user social relationship is introduced, and personalized product recommendation is provided for the target user based on the production type information. The calculation is simple and quick, and the method has better expandability and adaptability, so that the method is suitable for highly dynamic and immense amount of product-oriented recommendation for the internet users.

A system for generating a word sequence includes one or more processors in connection with a memory and one or more storage devices storing instructions causing operations that include receiving first and second input vectors, extracting first and second feature vectors, estimating a first set of weights and a second set of weights, calculating a first content vector from the first set of weights and the first feature vectors, and calculating a second content vector, transforming the first content vector into a first modal content vector having a predetermined dimension and transforming the second content vector into a second modal content vector having the predetermined dimension, estimating a set of modal attention weights, generating a weighted content vector having the predetermined dimension from the set of modal attention weights and the first and second modal content vectors, and generating a predicted word using the sequence generator.

A biometric user authentication method, includes enrolling a user based on user's biometric samples to generate user's reference data; and authenticating the user based on a user's live biometric sample and the user's reference data; wherein enrolling a user includes acquiring the user's biometric samples; extracting an enrollment feature vector from each user's biometric sample; computing a biometric reference template vector as a mean vector based on the enrollment feature vectors; computing a variation vector based on the enrollment feature vectors and the mean vector; randomly generating an enrollment secret vector; computing an enrollment code vector based on the enrollment secret vector and the variation vector; computing a difference vector as a wrap-around difference between the enrollment code vector and the mean vector; computing an error correction vector based on the enrollment secret vector to enable error correction during the user authentication phase according to a given error tolerance level, wherein the error correction vector is not computed if the error tolerance level is equal to zero; and storing the variation vector, the difference vector, and the error correction vector as a part of the user's reference data to be used during the user authentication phase.

The invention discloses a method for identifying virtual malicious nodes and a virtual malicious node network in social networks. The method includes the following steps: (1) obtaining the attribute data, behavioral data and communication data of unidentified accounts from target social networks; (2) calculating the creditworthiness of each unidentified account according to the credibility model feature vector calculated using the extracted data; (3) making a comparison between the user behavioral habit statistic data of each unidentified account with the credibility being lower than a set threshold value and that of a normal user, and judging whether the unidentified account is a virtual malicious node or not; (4) sorting the virtual malicious node set and establishing association for the virtual malicious node in each sort result, so as to form a virtual malicious node network; then evaluating by using the Bayesian network algorithm to determine the final virtual malicious node network. The method provided by the invention can be used to effectively identify highly-disguised malicious nodes and a collaborative virtual malicious node network.

A multi-view face recognition method and system are provided. In the multi-view face recognition method, two images to be recognized are input, a linear projection matrix is calculated based on grouped images in a training set, two feature vectors corresponding to the two input images are extracted based on the linear projection matrix, a distance between the two extracted feature vectors is calculated, and it is determined based on the distance between the two feature vectors whether the two input images belong to a same person.

The invention relates to an information pushing method. The method includes the following steps that feature information of contents to be pushed and historical operation information of a target user are extracted from a database, and a content feature vector and a user feature vector are constructed according to the feature information and the historical operation information respectively; similarity between the target user and other users is calculated according to the user feature vector, a plurality of similar users similar to the target user are selected from the other users according to the similarity, weights to all the contents to be pushed of the similar users are obtained, and correlation coefficients to all the contents to be pushed of the target user are calculated according to the weights; heat predication is conducted on all the contents to be pushed according to the content feature vector, and the correlation coefficients are corrected according to the heat predication result; all the contents to be pushed are ranked according to the corrected correlation coefficients, and the corresponding contents to be pushed are pushed to the target user according to the ranking result.

A method and system for detecting and tracking objects near a vehicle using a three dimensional laser rangefinder. The method receives points from the laser rangefinder, where the points represent locations in space where the rangefinder senses that some object exists. An algorithm first estimates the location of a ground plane, based on a previous ground plane location, data from onboard sensors, and an eigenvector calculation applied to the point data. Next, a plan view occupancy map and elevation map are computed for stationary objects, based on point data in relation to the ground plane. Finally, dynamic objects are detected and tracked, sensing objects which are moving, such as other vehicles, pedestrians, and animals. The output of the method is a set of stationary and dynamic objects, including their shape, range, and velocity. This output can be used by downstream applications, such as collision avoidance or semi-autonomous driving systems.

The invention provides a commodity recommendation method. The method comprises steps: a user information vector corresponding to associated user information of each user in a target user set and a commodity information vector corresponding to associated commodity information of each commodity in a target commodity set are extracted; the commodity information vector and the user information vectorare inputted to a corresponding depth neural network model trained and obtained according to user historical behavior data for feature extraction; according to the extracted commodity feature vector and a user feature vector, the correlation coefficient between each user and each commodity is calculated; and according to all correlation coefficients associated with each user, a commodity is recommended for each user. Correspondingly, the invention also corresponding provides a commodity recommendation device, computer equipment and a computer readable storage medium. Through the technical scheme of the invention, the cold start problem in the existing recommendation scheme can be effectively solved, the feature extraction difficulties are effectively reduced, and the commodity recommendation efficiency and the commodity recommendation effects are thus improved.

The invention belongs to the technical field of neural network, and provides a method and apparatus for increasing generalization capability of a convolutional neural network. The method includes the following steps: reading a group of images from a training set, mapping the group of images to a plurality of image character vectors, dividing the plurality of image characteristic vectors into a plurality of classes based on the types of the images; based on the image characteristic vectors of each class, calculating the intra-class loss function of all the characteristic vectors; based on the image characteristic vectors of each class, calculating the inter-class loss function of all the characteristic vectors; based on the intra-class loss function of all the characteristic vectors, using the back propagation algorithm to update the weight of each node of the convolutional neural network; repeating the above mentioned steps until the convolutional neural network converges on the training set or reaches preset repeating times. According to the invention, the method and the apparatus can save all data in long-tailed distribution, makes full usage of abundant inter-class information of tail data, and increases the generalization capability of the convolutional neural network.

The invention relates to a magnetic resonance diffusion tensor imaging fiber bundle tracking device. The process is respectively completed by each component through the following steps of (1) collecting magnetic resonance diffusion tensor images; (2) carrying out brain issue dividing on a sequence without a diffusion gradient magnetic field and any sequence with the diffusion gradient magnetic field, and using the sequence without the diffusion gradient magnetic field as an image reference template of the brain tissues; (3) carrying out three-dimensional affine conversion on the two image sequences of the brain tissues after being extracted, to obtain a space conversion relationship; (4) carrying out space conversion on the remained diffusion tensor images by an optimum conversion relationship; (5) calculating tensor fields and feature vectors; (6) setting the interested areas and tracking conditions; and (7) carrying out the bidirectional tracking and displaying based on a fiber bundle with variable step size, so as to quickly and effectively carry out fiber bundle tracking and displaying on the white matters of human brain. In the diffusion tensor imaging process, the image deviation caused by space positions can be corrected, and the elastic step size is adopted in the tracking process, so as to ensure the reliable fiber bundle tracking.

A biometric user authentication method, includes enrolling a user based on user's biometric samples to generate user's reference data; and authenticating the user based on a user's live biometric sample and the user's reference data; wherein enrolling a user includes acquiring the user's biometric samples; extracting an enrollment feature vector from each user's biometric sample; computing a biometric reference template vector as a mean vector based on the enrollment feature vectors; computing a variation vector based on the enrollment feature vectors and the mean vector; randomly generating an enrollment secret vector; computing an enrollment code vector based on the enrollment secret vector and the variation vector; computing a difference vector as a wrap-around difference between the enrollment code vector and the mean vector; computing an error correction vector based on the enrollment secret vector to enable error correction during the user authentication phase according to a given error tolerance level, wherein the error correction vector is not computed if the error tolerance level is equal to zero; and storing the variation vector, the difference vector, and the error correction vector as a part of the user's reference data to be used during the user authentication phase.

The invention provides an identification method of characters of a licence plate, mainly solving the problems of low identification speed and complex realization in the prior art. The identification process of the identification method comprises the following step of: individually taking out the pretreated characters of the licence plate, uniformizing, and scanning in vertical and horizontal directions; recording horizontal and vertical characteristic vectors, and respectively calculating similarity with characteristic vectors already well stored in a standard character library; finding a certain character of the standard character library, which has maximum similarity, through similarity magnitude comparison, wherein the character is a first-time identification result of characters to be identified; dividing partial similar characters in the first-time identification result into an up and down part and a left and right part, and respectively scanning the characteristic vectors of the similar characters; and then carrying out similarity calculation with a characteristic vector library of the characters, and finding a certain character of the standard character library, which has the maximum similarity, as a final identification result. The invention has high identification speed and identification accuracy rate and can be used for the real-time detection of the licence plate of a high-speed running vehicle.

A system, method and device for object identification is provided. The method of identifying objects includes, but is not limited to, calculating feature vectors of the object, calculating feature vectors of the object's context and surroundings, combining feature vectors of the object, calculating likelihood metrics of combined feature vectors, calculating verification likelihood metrics against contact list entries, calculating a joint verification likelihood metric using the verification likelihood metrics, and identifying the object based on the joint verification likelihood metric.

A sound source localization apparatus for localizing a sound source using an eigenvector, includes, a sound signal input unit inputting a sound signal, a correlation matrix calculation unit calculating a correlation matrix of the input sound signal, and an eigenvector calculation unit calculating an eigenvalue of the correlation matrix using the calculated correlation matrix, wherein the eigenvector calculation unit calculates the eigenvector using the correlation matrix of the input sound signal and one or more predetermined correlation matrices.

A mechanomyography (MMG) signal input device, a human-machine operating system and an identification method thereof are provided. The system includes a mechanomyography (MMG) signal input device, a signal processing unit, a motion database and a calculating unit. The MMG signal input device is mounted on a measuring portion of a testing body, wherein the measuring portion has a plurality of muscle groups. The signal processing unit is used for receiving the MMG signals of the muscle groups, and performing integration and pre-processing on the MMG signals to obtain a processed signal. The motion database is used for storing a motion mode. The calculating unit is used for performing signal intensity computation, data segmentation, feature vector calculation, and testing body's motion recognition, and outputting a corresponding control signal according to the result of recognition.

The invention discloses a content-based question automatic classification method and a content-based question automatic classification system. The system comprises a question key word acquisition module, a characteristic space construction module and a semantic mapping module, wherein, the question key word acquisition module is used for acquiring a question key word of a novel question according to a key word label and / or a fillable content label in a template, setting a weight for the question key word and obtaining a question vector of the novel question; the characteristic space construction module is used for acquiring a characteristic vector of each class according to questions of all the prior classes and weights and constructing a characteristic space; the semantic mapping module is connected with the characteristic space construction module and the question key word acquisition module and used for mapping the question vector of the novel problem to the characteristic space, calculating the similarity of the novel problem and each class according to the question vector of the novel question after mapping and the characteristic vector of each class, and returning classes which are most related to the novel problem according to the similarity. The content-based question automatic classification method and the content-based question automatic classification system realize automatic classification of the novel problem which is put forward by a user and return most probable results to the user for selection.

The embodiment of the invention provides a method and device for region division. The method comprises: a regional boundary of a to-be-divided region is obtained and a rectangular boundary of the to-be-divided region is calculated; location information of a transportation junction point in the to-be-divided region is obtained and a shortest-path polygon is drawn by using the transportation junction point as a discrete point; regional combination processing is carried out on the shortest-path polygon and the rectangular boundary of the to-be-divided region to obtain a preliminary regional division result; feature vector calculation parameters of interest points in the to-be-divided region are obtained and feature vectors of all interest points in all regions in the preliminary regional division result are calculated based on the feature vector calculation parameters; on the basis of the feature vectors of all interest points in all regions, regional similarity degrees between all regions and spatial adjacent regions are calculated; and according to the regional similarity degrees, regions in the preliminary regional division result are combined to generate a final regional divisionresult. According to the invention, with full consideration of characteristics of a road network and regional interest points, regional division becomes reasonable.

The invention discloses a face recognition method by an independent component analysis based on a multi-scale total variational derivative image, which belongs to the face recognition technical field; the method is as follows: a contrast gradient is strengthened; TV-L<1> is used to carry out scale decomposition to a face image to obtain a large-scale image comprising a skeleton contour and muscle information and a small-scale image comprising the details of mouth, eyes and nose; quotient balance is carried out to the small-scale image to obtain the feature of unchanged illumination; feature fusion technology is selected to fuse the features of large scale and unchanged illumination into a new face image; Gabor is used to analyze and extract the features of the new face image in a specific scale and direction to generate a multi-scale Gabor face; the eigenvectors of all the samples are extracted by an information maximization independent component analysis algorithm; the similarity of the eigenvectors of the face which is to be treated with recognizing is calculated by the eigenvectors of the known face; according to the similarity, the eigenvectors are sorted to acquire a final recognition result. The face recognition method achieves high recognition rate and strong robustness to illumination, expression, make-up and other external interference.

A feature vector computation section 24 of an image processing apparatus computes a feature vector expressing gradient histograms for each of plural child regions that have been further partitioned from plural parent regions partitioned from a discrimination-subject image. A feature relative vector computation section 26 of the image processing apparatus computes for each parent region a feature relative vector expressing relative values computed from respective combinations of the same or different elements across feature vectors computed for each child region, and relative values computed from respective combinations of the same or different elements within one of the feature vectors. A discrimination section 30 of the image processing apparatus, based on the feature relative vector computed for each parent region, discriminates whether or not the image subject to processing is an image in which a processing target object appears.

The invention provides a control method for calculating Chinese word semantic similarity. The method is combined with a method based on Chinese thesaurus and based on statistic. The method is characterized by comprising the following steps of: a, obtaining a word pair, and obtaining a corresponding feature vector according to the word pair; b, carrying out semantic expansion on the feature vector so as to generate an expanded feature vector; c, carrying out semantic mapping on the feature vector so as to generate a mapped feature vector; and d, calculating the word similarity according to the expanded feature vector and the mapped feature vector.

The embodiment of the invention provides an image recognition method and a device. The method comprises the following steps: obtaining an image to be recognized, dividing the image to be recognized into a plurality of sub-images according to a preset size, and obtaining a first gray feature vector corresponding to each sub-image; acquiring a second gray feature vector corresponding to the matchingtemplate, calculating similarity between the sub-image and the matching template according to the first gray feature vector and the second gray feature vector; the sub-images whose similarity is greater than the preset threshold are selected as the target sub-images, and the target sub-images are recognized by the classification model according to the target sub-images. The apparatus is configured to perform the method. An embodiment of the present invention calculates a similarity between a sub-image and a matching template according to a first gray feature vector and a second gray feature vector, and the target sub-images whose similarity is greater than the preset threshold are selected as the target sub-images, and then the target sub-images are input into the classification model forrecognition, thereby reducing the recognition time and improving the recognition efficiency of the images to be recognized.

The invention discloses a method and a system for the automatic issue of network advertisements. The method is as follows: step 1, the page layout content of a current web page is analyzed to establish a characteristic vector of the current web page; step 2, according to the characteristic vector, the correlation degree between the current web page and each first issued advertisement is calculated, and the scoring of each first advertisement is carried out according to the correlation degree; step 3, according to a scoring value of each advertisement, a first advertisement is selected and issued on the current web page; step 4, when the system jumps to a new web page from the current web page, a second issued advertisement is calculated for the new web page, and the selected first advertisement or second advertisement are issued on the new web page according to scores on the new web page of the selected first advertisement and the selected second advertisement. When a user browses information from network, the method can automatically issue related advertisements according to the current web page information and a user model; moreover, when a user skips between web pages, the method ensures that the advertisement can track the user to the new web page.

The invention relates to the field of information processing, and discloses a text classification method and a terminal device. According to the embodiment, the method comprises the steps of calculating a word vector matrix containing all word vectors after training samples of N text types are segmented; calculating out the feature vector of the training sample of each text type based on the word vector matrix; calculating the input amount of a back-propagating neural network according to the calculated feature vectors of the training samples; then determining a text classifier according to the back-propagating neural network; finally determining the type of a text to be tested according to the feature vector of the text to be tested and the text classifier. In CBOW pair word vectorization, the relation of a current word with a few front words and a few rear words is taken into consideration, the whole network training also has a semantic feature by being combined with the classical back-propagating neural network, the whole network can further recognize and understand the text content, and the better training effect is achieved.

A method for learning parameters used to recognize characters included in a text in a scene text image of training set is provided. The method includes steps of: (a) a training apparatus generating each feature vector corresponding to each of the segmented character images; (b) the training apparatus processing feature vectors ci+j of neighboring character images to thereby generate a support vector to be used for a recognition of a specific character image; (c) the training apparatus obtaining a merged vector by executing a computation with the support vector and a feature vector ci of the specific character image; and (d) the training apparatus (i) performing a classification of the specific character image as a letter included in a predetermined set of letters by referring to the merged vector; and (ii) adjusting the parameters by referring to a result of the classification.

A method and apparatus of a device that classifies an image is described. In an exemplary embodiment, the method includes tiling at least one region of interest of the input image into a set of tiles. For each tile, the method includes extracting a feature vector of the tile by applying a convolutional neural network, wherein a feature is a local descriptor of the tile; and computing a score of the tile from the extracted feature vector, said tile score being representative of a contribution of the tile into a classification of the input image. The method also includes sorting a set of the tile scores and selecting a subset of the tile scores based on their value and / or their rank in the sorted set. The method also includes applying a classifier to the selected tile scores in order to classify the input image

The invention relates to a time-domain fault identifying method based on coefficient variations of a regression model. The time-domain fault identifying method includes the following steps of firstly, setting a reference eigenvector, and extracting a set of a reference eigenvector formed by coefficients of the regression model; secondly, setting an eigenvector to be evaluated: extracting another set of a to-be-evaluated eigenvector formed by coefficients of the regression model according to the same way of the first step when in need of fault diagnosis of equipment; thirdly, computing difference degree between the eigenvector to be evaluated and the reference eigenvector, and indicating the difference degree as vector distance; fourthly, evaluating significance level as time limit value; and fifthly, comparing and judging equipment conditions: comparing the vector distance with the limit value, and analyzing and judging whether faults exist or not. The time-domain fault identifying method is low in computing cost and high in reliability and applicability.