138 results about "Mean vector" patented technology

A speech recognizer operating in both ambient noise (additive distortion) and microphone changes (convolutive distortion) is provided. For each utterance to be recognized the recognizer system adapts HMM mean vectors with noise estimates calculated from pre-utterance pause and a channel estimate calculated using an Estimation Maximization algorithm from previous utterances.

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.

A speech recognizer operating in both ambient noise (additive distortion) and microphone changes (convolutive distortion) is provided. For each utterance to be recognized the recognizer system adapts HMM mean vectors with noise estimates calculated from pre-utterance pause and a channel estimate calculated using an Estimation Maximization algorithm from previous utterances.

The invention discloses a tracking method of a video frequency moving object target self-adapting window, relating to machine vision and pattern recognition technique. A frame image of a video frequency sequence is read into a postpositional memory zone, and staring position and size information of a tracked object is obtained in the frame, then distribution statistical information of a target signature is extracted to build a gauss mixed model as an object template, a mean vector and a covariance matrix in gauss mixed distribution are used to describe the position and the size of the object, then the next frame image of the video frequency sequence is read into the postpositional memory zone. In a new frame video frequency image, a parameter estimation method is used to obtain object gauss mized model parameters in the current frame in iterative computation and find candidate template similar to the object template, and the final model parameters obtained in iteration are used to update the tracking window for realizing self-adaption of the tracked window. According to the tracking method of the video frequency moving object target self-adapting window, tracking reliability is greatly increased, which is widely applied in the fields of robot technology, vision navigation, automatic supervision, traffic control and the like.

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.

Ultrasonic inspection equipment facilitates alignment of display positions of three-dimensional ultrasonic inspection data and three-dimensional shape data, and quickly discriminates between a defect echo and an inner-wall echo. A computer 102A has a position correction function of correcting a relative display position between three-dimensional shape data and three-dimensional ultrasonic inspection data. A display position of the three-dimensional ultrasonic inspection data or that of the three-dimensional shape data is moved by a norm of a mean vector along the mean vector that is calculated from a plurality of vectors defined by a plurality of points selected in the three-dimensional ultrasonic inspection data and by a plurality of points selected in the three-dimensional shape data. The three-dimensional shape data and the three-dimensional ultrasonic inspection data are displayed in such a manner as to be superimposed on each other on a three-dimensional display unit 103C.

A method of extracting feature vectors of an image by using local linear transformation functions, and a method and apparatus for image recognition employing the extracting method. The method of extracting feature vectors by using local linear transformation functions includes: dividing learning images formed with a first predetermined number of classes, into a second predetermined number of local groups, generating and storing a mean vector and a set of local linear transformation functions for each of the divided local groups comparing input image vectors with the mean vector of each local group and allocating one of the local groups to the input image; and extracting feature vectors by vector-projecting the local linear transformation functions of the allocated local group on the input image. According to the method, the data structure that has many modality distributions because of a great degree of variance with respect to poses or illumination is divided into a predetermined number of local groups, and a local linear transformation function for each local group is obtained through learning. Then, by using the local linear transformation functions, feature vectors of registered images and recognized images are extracted such that the images can be recognized with higher accuracy.

The invention provides a recording device clustering method based on Gaussian mean super vectors and spectral clustering. The method comprises the steps that the Melch frequency cepstrum coefficient MFCC characteristic which characterizes the recording device characteristic is extracted from a speech sample; the MFCC characteristics of all speech samples are used as input, and a common background model UBM is trained through an expectation maximization EM algorithm; the MFCC characteristic of each speech sample is used as input, and UBM parameters are updated through a maximum posteriori probability MAP algorithm to acquire the Gaussian mixture model GMM of each speech sample; the mean vector of all Gaussian components of each GMM is spliced in turn to form a Gaussian mean super vector; a spectral clustering algorithm is used to cluster the Gaussian mean super vectors of all speech samples; the number of recording devices is estimated; and the speech samples of the same recording device are merged. According to the invention, the speech samples collected by the same recording device can be found out without knowing the prior knowledge of the type, the number and the like of the recording devices, and the application scope of the method is wide.

A method for adapting a codebook for speech recognition, wherein the codebook is from a set of codebooks comprising a speaker-independent codebook and at least one speaker-dependent codebook is disclosed. A speech input is received and a feature vector based on the received speech input is determined. For each of the Gaussian densities, a first mean vector is estimated using an expectation process and taking into account the determined feature vector. For each of the Gaussian densities, a second mean vector using an Eigenvoice adaptation is determined taking into account the determined feature vector. For each of the Gaussian densities, the mean vector is set to a convex combination of the first and the second mean vector. Thus, this process allows for adaptation during operation and does not require a lengthy training phase.

An image classification method based on a deep learning feature and a maximum confidence path belongs to the field of mode identification. The method comprises the steps of: training a convolutional neural network on a large enough image library; extracting an image feature by means of the trained convolutional neural network; calculating a mean vector of each class; performing iteration clustering on the mean vector that represents each class by means of a spectrum clustering algorithm so as to construct a visual tree; training svm for each non-leaf node of the tree; and for a given test image, from top to bottom, judging the probability of the test image to the corresponding node, and finding a leaf node with the biggest path probability, namely a final target class. The image feature is extracted by means of CNN, thereby achieving very good discrimination and robustness; a distance calculation formula of two classes is given out, the complexity of calculation is greatly optimized by means of derivation and the similarity of the classes is obtained, so that the visual tree is constructed by iteratively using the spectrum clustering algorithm; and the use of a visual relationship between the classes can achieve very good effects for large image classification.

A method of speech recognition with compensation is provided by modifying HMM models trained on clean speech with cepstral mean normalization. For all speech utterances the MFCC vector is calculated for the clean database. This mean MFCC vector is added to the original models. An estimate of the background noise is determined for a given speech utterance. The model mean vectors adapted to the noise are determined. The mean vector of the noisy data over the noisy speech space is determined and this is removed from model mean vectors adapted to noise to get the target model.

A knowledge map representation learning method based on structure information and text description aims at mapping entities and relationships in a triple into a low-dimensional continuous real value space. The invention aims at improving vector representation of entities in knowledge representation; The corresponding text description information of the entity is obtained from the existing knowledge base Freebase, Word2vec is used to represent each description with a word vector, Then, the description vector is represented by word addition and mean vector, and the description vector is represented by doc2vec sentence vector. Then, the word vector is used as the input of CNN text encoder to obtain the description text-based representation vector of each entity. In the joint representation, weights are used to evaluate the symbol-based representation vectors in the knowledge base, and the effects of the representation vectors based on network structure and description text on the final representation vectors of entities are analyzed, so as to achieve the fusion of structure information and text information, and improve the accuracy of knowledge map representation.

Reducing mismatch between HMMs trained with clean speech and speech signals recorded under background noise can be approached by distribution adaptation using parallel model combination (PMC). Accurate PMC has no closed-form expression, therefore simplification assumptions must be made in implementation. Under a new log-max assumption, adaptation formula for log-spectral parameters are presented, both for static and dynamic parameters. The system takes the mean vector (<HIL><PDAT>41< / BOLD><PDAT>) made up of the static part (<HIL><PDAT>41< / BOLD><HIL><PDAT>a< / ITALIC><PDAT>) and the dynamic part (<HIL><PDAT>41< / BOLD><HIL><PDAT>b< / ITALIC><PDAT>) and the noise vector (<HIL><PDAT>43< / BOLD><PDAT>) made up of the static part (<HIL><PDAT>43< / BOLD><HIL><PDAT>a< / ITALIC><PDAT>) and dynamic part (<HIL><PDAT>43< / BOLD><HIL><PDAT>b< / ITALIC><PDAT>) and applies to a decision circuit (<HIL><PDAT>45< / BOLD><PDAT>) to determine if the quiet vector plus the gain of speech produced in noise with respect to clean speech is greater than the noisy mean vector and if so the static part is equal to the gain plus the clean speech vector and the dynamic part is the change in the the quiet speech vector and if not greater than the noise vector then the static part equals the noise vector and the dynamic part is zero.< / PTEXT>

The invention provides a dynamic non-uniformity correction method for a linear scanned image based on image sequence analysis. The dynamic non-uniformity correction method comprises the following steps: firstly, selecting k (k is equal to 1, 2,..., and belongs to Z) images for the existing image to be corrected, and setting an ideal desired value q; then, calculating the average value of pixel values on each row of the k images to acquire a mean vector; and finally, using a ratio of the mean vector to the set ideal desired value as a correction coefficient to correct the images, so that a novel image which is uniform in background, clear in defect contour and convenient to position a subsequent defect target is acquired. The method can effectively overcome influences of changes of such factors as parameters including material specification, thickness and the like of the detected object, detection angle, plate type consistency and light source illumination attenuation, so as to dynamically acquire in real time an ideal image which is uniform in grey distribution, high in definition and remarkably enhanced in the detected object, thus greatly simplifying subsequent image processing difficulty and effectively improving the real-time performance of the integral system.

The invention belongs to the technical field of emergency processing of environmental engineering, and particularly relates to a pipe explosion pre-warning method based on metering zone flow monitoring data. The method comprises the steps of collecting flow data for forming a historical data matrix and performing a pipe explosion simulation experiment to obtain pipe explosion data for forming a test matrix; presetting an initialization parameter k1 and an abnormal value detection parameter k2, performing clustering analysis on the historical data matrix, removing abnormal vectors, generating an initialization matrix, forming a detection matrix by the initialization matrix and test vectors, performing clustering analysis on the detection matrix, removing abnormal test vectors, performing difference operation for the vectors and mean vectors of the detection matrix, and if each difference value is greater than zero or less than zero, not giving a pipe explosion pre-warning, otherwise, giving the pipe explosion pre-warning; and computing a missing alarm rate and a false alarm rate, adjusting the preset initialization parameter and the preset abnormal value detection parameter until optimal missing alarm rate and false alarm rate are obtained, and applying the optimal missing alarm rate and false alarm rate to real-time pipe explosion pre-warning. The pre-warning can be accurately given during pipe explosion or leakage, the hardware cost is low, and the pre-warning effect is ideal.

Ultrasonic inspection equipment facilitates alignment of display positions of three-dimensional ultrasonic inspection data and three-dimensional shape data, and quickly discriminates between a defect echo and an inner-wall echo.A computer 102A has a position correction function of correcting a relative display position between three-dimensional shape data and three-dimensional ultrasonic inspection data. A display position of the three-dimensional ultrasonic inspection data or that of the three-dimensional shape data is moved by a norm of a mean vector along the mean vector that is calculated from a plurality of vectors defined by a plurality of points selected in the three-dimensional ultrasonic inspection data and by a plurality of points selected in the three-dimensional shape data. The three-dimensional shape data and the three-dimensional ultrasonic inspection data are displayed in such a manner as to be superimposed on each other on a three-dimensional display unit 103C.

An automatic image-labeling method of fusing pseudo-correlation feedback with index technique includes indexing unlabeled query image in labeled imagebank to obtain k numbers of correlation images and label word set of correlation image, calculating posterior probability (PP) of each labeled word labeled query image (ELWQI), picking up mean value vector as new query vector according to k numbers of correlation image, repeating said steps till maximum iteration frequency N is obtained, calculating stability factor (SF) of each label word in each cycle of labeled word set, calculating sequence probability of ELWQI according to PP and value of SF in order to label query image.

The embodiment of the invention discloses a facial recognition method and device, and belongs to the technical field of image processing. The method comprises the steps that a facial image to be recognized is acquired and the target feature vectors of the facial image are extracted; the distance between each target feature vector and the mean vector corresponding to each facial identifier is calculated so as to obtain first distance sets, and the corresponding first recognition result and the corresponding first confidence coefficient of the facial image are determined according to each firstdistance set; the distance between each target feature vector and each feature vector corresponding to each facial identifier is calculated so as to obtain second distance sets, and the correspondingsecond recognition result and the corresponding second confidence coefficient of the facial image are determined according to each second distance set; and the corresponding facial recognition resultof the facial image and the corresponding confidence coefficient of the facial recognition result are determined according to the first recognition result, the first confidence coefficient and the second recognition result and the second confidence coefficient. With application of the facial recognition method and device, the efficiency and the accuracy of facial recognition can be enhanced.

The invention discloses an increment subspace target tracking method based on Lp norm regularization. The increment subspace target tracking method comprises the following steps: reading a video sequence image; manually marking an image of a first frame target; acquiring an observation value through utilizing a Ross method; directly calculating residual errors of front five frames of images to determine a target state of the current frame, and collecting a sample; acquiring an initial subspace and a mean vector by adopting the Ross method; establishing a target function for observed samples under an augmented Lagrangian framework from the sixth frame, and carrying out minimization calculation; estimating the confidence coefficient that each candidate sample becomes a current frame real target, and collecting the sample; and when five samples are collected, updating the subspace and the mean vector by carrying out increment through the Ross method. Under the condition of guaranteeing certain instantaneity, the increment subspace target tracking method has relatively strong interference resistance under challenges, such as shielding, light illumination and motion blur, on the target; and the influences on tracking by different sparseness can be observed under one uniform framework.

The invention discloses a public sentiment event detection method and device. The method comprises the following steps that: obtaining the feature word vector of a text to be detected; obtaining vectors corresponding to all feature words, and obtaining a sensitive term-of-meaning vector; calculating a similarity between the feature word vector of the text to be detected and the feature word vectors corresponding to all feature words; obtaining a corresponding first sensitive term of meaning when the similarity is maximum, and obtaining the amount of the first sensitive terms of meaning and the amount of the feature words in the text to be detected; according to a first preset weight value and a second preset weight value, calculating the weighted sum of the amount of the first sensitive terms of meaning and the amount of the feature words; and when the weighted sum is greater than a threshold value, determining a described event in the text to be detected as a public sentiment event. By use of the method, through the vectorization of the text to be detected, effective semantic constraint can be achieved; and meanwhile, the similarity between the feature word vector of the text to be detected and the feature word vectors corresponding to all feature words can be calculated to accurately detect the problems of the public sentiment event which needs to be paid attention.

The mismatch between the distributions of acoustic models and features in speech recognition may cause performance degradation. A sequential bias adaptation (SBA) applies state or class dependent biases to the original mean vectors in acoustic models to take into account the mismatch between features and the acoustic models.

The invention discloses a two-dimensional Otsu image segmentation method in combination with a fruit fly optimization algorithm. The method adopts a two-dimensional histogram area divided by two straight lines passing through a threshold vector point and a gray-scale axis to form angles alpha and beta, and comprises the steps of calculating a joint probability density first, then calculating probabilities and mean vectors of a target area and a background area, calculating a between-class variance and a within-class variance of the target area and the background area, weighting the background area in a threshold solving formula according to the proportions of the target and the background in the image, and seeking an optimal two-dimensional threshold vector by adopting the fruit fly optimization algorithm. The method has the advantages of more accurate segmentation on pictures, better noise suppression and low running time consumption.

The invention relates to a data classification system and method based on KL divergence optimizationmethod for classifying data based on KL divergence optimization. The method comprises the steps thatdata preprocessing is conducted on original images, t, texts and other data, and objects are modeled into multi-dimensional distribution; S; selecting a certain amount of triple from the tagged training data to carry out model training; T; the selected triple serves as training data, a linear mapping A is applied to all the mean vectors, t, the optimal linear mapping is learned through iterativeoptimization, and the learning process is based on the basic assumption of metric learning, t, that is, t, the distance between samples of the same kind becomes smaller, and the distance between samples of different kinds becomes larger; A; an intrinsic gradient descent algorithm is adopted for optimization, and after the gradient of an objective function is projected to the tangent space of the same manifold, Riemannian gradient descent is executed on the manifold of an SPD matrix with given affine invariant Riemannian metric; A; and calculating the KL divergence between the test set and thetraining set, and classifying the samples by adopting a K-nearest neighbor (KNN) classifier. The method can effectively improve the classification precision of the system, and has more stable performance.

The invention provides an abnormal detection method and device for a hyper-spectral image. The method includes the steps that the hyper-spectral image to be detected is obtained, abnormal detection is performed on the hyper-spectral image to be detected through an RX algorithm, so that an abnormal detection result is obtained, threshold segmentation is performed on the abnormal detection result to obtain a target image element set and a background image element set, the mean vector and the covariance matrix of the target image element set are calculated, the mean vector and the covariance matrix of the background image element set are calculated, the distance between each image element to be detected in the hyper-spectral image and each background image element is calculated through the mean vector and the covariance matrix of the background image element set, the distance between each image element to be detected in the hyper-spectral image and each target image element is calculated through the mean vector and the covariance matrix of the target image element set, and the distance between each image element to be detected and each background image element is subtracted from the distance between each image element to be detected and each target image element to obtain the abnormal detection result. The method and device can improve the detection rate of abnormal objects and reduce the false alarm rate.

The invention discloses a remote-sensing image semi-supervised projection dimension reducing method based on local consistency. The method includes the following steps: (1) dividing a remote-sensing image data set; (2) generating a semantics similar matrix, a neighbor matrix and a location consistency matrix; (3) mixing a label matrix and the neighbor matrix; (4) generating neighbor mean vector; (5) generating an alien divergence matrix, a similar divergence matrix and a local consistency divergence matrix; (6) calculating an optimum projection matrix; and (7) conducting projection and dimension reducing. The method adopts the semi-supervised learning based on local consistency binding and improves recognition rate under small sample learning condition.

The invention provides an anomaly detection method and device. The method includes the steps that a hyper-spectral image to be detected is acquired, anomaly detection is carried out on the hyper-spectral image to be detected through the RX algorithm, an anomaly detection result matrix is acquired, and a weight matrix of the anomaly detection result matrix is acquired; normalization processing is carried out on the weight matrix, a normalized weight matrix of the weight matrix is acquired, the mean vector and covariance matrix of the background in the hyper-spectral image to be detected are calculated by using the normalized weight matrix, anomaly detection is carried out on the hyper-spectral image to be detected by using the mean vector and covariance matrix of the background through the RX algorithm, and anomaly detection results are acquired. According to the anomaly detection method, the mean vector and covariance matrix of the background are estimated through the weighting method, abnormal information and noise interference can be reduced, background information can be highlighted and better meet a hypothesis of the RX algorithm on the background information, and therefore detection rate is improved, and false alarm rate is reduced.

The present invention relates to a fast face recognition method in an application environment of a massive face database. The Bisecting K-Means (BKM) algorithm is combined with the PCA (principal component analysis) +LDA (linear discriminant analysis) algorithm to perform characteristic extraction from a large-scale face database. The method comprises the steps of: firstly, calculating a mean vector for each sample in the face database and then mapping the mean vector to a space with a specific dimension according to the principle that the intra-class aggregation is maximum, and the inter-class aggregation is minimum; then aggregating the mean vector in the dimension space; recombining the samples and grouped according to the cluster attribute after an aggregation result is obtained; and finally, extracting the characteristics on each cluster according to the linear subspace face characteristics extraction algorithm.

The invention discloses an optimal sensor layout method for system parameter identification. The method comprises the following steps: firstly supposing that to-be-identified parameters are in a probability distribution form (such as obey normal distribution), sampling the to-be-identified parameters based on the Monte Carlo method, computing the response vector at each candidate sensor, and performing decentration operation on the response vector, namely, subtracting the mean vector by the response vector; secondly, counting the response variance at each candidate sensor, taking the locationwith the maximum response variance as the optimal sensor layout location; thirdly, performing orthogonalization operation on the response vector at the residual candidate sensor and the response vector at the selected sensor to remove the related information, acquiring an independent component of the response vector at the residual candidate sensor, and selecting the location with the maximum independent variance as the suboptimal sensor layout location; and finally, repeating the above-mentioned selection process until the mode of the independent component of the response vector at the residual candidate sensor tends to zero. Through the method disclosed by the invention, the location and number of the optimal sensor required for identifying the system parameter can be determined, and theresponse information of the residual sensor can be completely expressed by the response information at the selected sensor.

The invention discloses a method for forecasting continuous casting crystallizer steel leakage by adopting spatial density-based clustering DBSCAN, and belongs to the technical field of ferrous metallurgy continuous casting detection. In the method, continuous casting steel leakage forecast and clustering analysis are organically combined, temperature data series at different times are analyzed through density-based clustering, and crystallizer steel leakage is recognized online and forecasted accurately. The method particularly comprises the steps that feature extraction is conducted on crystallizer copper plate thermocouple temperature data through density-based clustering, a steel leakage sample accumulation area and a mean vector are obtained, whether the temperature series detected inreal time are located in the accumulation area or not is judged online on the basis, a typical steel leakage temperature mode is recognized, and crystallizer steel leakage is forecasted. According tothe method, typical change features of thermocouple temperature in the time and space are extracted and fused according to the single and approximate features of the temperature mode during steel leakage, the similarity of a large quantity of steel leakage samples is excavated through clustering analysis, temperature fluctuation and misforecast caused by factors such as normal working condition and manual operation are effectively rejected, and the steel leakage forecast accuracy rate is significantly increased.