1850 results about "Mixture model" patented technology

A method for distance definition in a text-to-speech conversion system by applying Gaussian Mixture Model (GMM) to a distance definition. According to an embodiment, the text that is to be subjected to text-to-speech conversion is analyzed to obtain a text with descriptive prosody annotation; clustering is performed for samples in the obtained text; and a GMM model is generated for each cluster, to determine the distance between the sample and the corresponding GMM model.

Audio/Visual data is classified into semantic classes such as News, Sports, Music video or the like by providing class models for each class and comparing input audio visual data to the models. The class models are generated by extracting feature vectors from training samples, and then subjecting the feature vectors to kernel discriminant analysis or principal component analysis to give discriminatory basis vectors. These vectors are then used to obtain further feature vector of much lower dimension than the original feature vectors, which may then be used directly as a class model, or used to train a Gaussian Mixture Model or the like. During classification of unknown input data, the same feature extraction and analysis steps are performed to obtain the low-dimensional feature vectors, which are then fed into the previously created class models to identify the data genre.

Methods and systems are provided for analyzing the business rules, business metrics, and decision parameters for a firm or organization, processing a subset of such data to form output, and offering access to selective views of such output including evaluation and comparative data regarding execution of such business rules, information on corresponding business metrics or sets of business metrics, information on corresponding decision parameters or sets of decision parameters or scenarios, and other useful analytic information which can help a firm or organization evaluate and modify business policies based on said rules, metrics, and parameters. In addition to said rules, metrics and/or parameters, the data for the business rule analysis can include conventional historical data or hypothetical data based on simulations which the current system and method provide based on prescribed random and non-random algorithms. The simulated or hypothetical data enables users to conduct rule analysis based on historical data, simulated data, or hybrid models. In this manner, the methods and systems described in this invention provide for both an evaluation of a firm's current policies as well as an evaluation of policy modifications not actually executed but for which hypothetical data can be provided and analyzed.

The invention discloses a method for analyzing high precision 4D flight trajectory of an airplane based on real-time radar data, comprising the following steps: establishing a 4D fight trajectory theoretical model based on the airplane performance; establishing a 4D fight trajectory empirical model by mining and analyzing based on the historical fight trajectory data of the airplane types in the above step; combining the theoretical model and the empirical model and quantizing various influencing factors in the flight process into adjustable parameters; and correcting the 4D flight trajectory mixed models by radar data to form the final flight trajectory of the airplane. The invention takes the airplane performance into consideration and establishes the theoretical model by performance library parameters on the basis of a standard flight procedure, thus having high reliability; the invention forms the empirical model by analyzing the historical data, thus having high reality; the invention generates the 4D flight trajectory most similar to the actual flight and can adjust the parameters to form a plan trajectory aiming at the environment of each flight, thus high sensitivity; and the invention correct the plan trajectory by introducing the real-time radar data, thus having high accuracy.

The invention provides a voiceprint identification method based on a Gauss mixing model and a system thereof. The method comprises the following steps: voice signal acquisition; voice signal pretreatment; voice signal characteristic parameter extraction: employing a Mel Frequency Cepstrum Coefficient (MFCC), wherein an order number of the MFCC usually is 12-16; model training: employing an EM algorithm to train a Gauss mixing model (GMM) for a voice signal characteristic parameter of a speaker, wherein a k-means algorithm is selected as a parameter initialization method of the model; voiceprint identification: comparing a collected voice signal characteristic parameter to be identified with an established speaker voice model, carrying out determination according to a maximum posterior probability method, and if a corresponding speaker model enables a speaker voice characteristic vector X to be identified to has maximum posterior probability, identifying the speaker. According to the method, the Gauss mixing model based on probability statistics is employed, characteristic distribution of the speaker in characteristic space can be reflected well, a probability density function is common, a parameter in the model is easy to estimate and train, and the method has good identification performance and anti-noise capability.

Object comparison is disclosed, including: adapting N universal mixture model components to a first object to generate N corresponding first object mixture model components, where N is an integer greater than or equal to two; and generating a similarity measure based on component-by-component comparison of the N first object mixture model components with corresponding N second object mixture model components obtained by adaptation of the N universal mixture model components to a second object.

The invention discloses a track predication method based on a Gauss mixture time series model (GMTSM for short) to carry out model regression and analysis of road vehicle flow on a large quantity of vehicle historical tracks to realize vehicle track predication. The method mainly comprises the following steps: (1) carrying out unsupervised learning cluster on vehicle historical tracks through a k-means algorithm; (2) constructing a historical track probability distribution model through a Gauss mixture time series model; and (3) predicating tracks of mobile objects through a regressive processof the mixture model. An experiment result shows that the Gauss mixture time series model automatically adjusts the weight of sub-models and selects a predication track with the largest probability in a condition that road vehicle flows change suddenly.

In the invention, the detecting system comprises a video camera and a signal processor. It uses an improved Gaussian mixture distribution model to characterize each pixel in image frame; the characterized feature only uses a brightness feature; if there isn't a moving object (vehicle), the video image is relatively static; each pixel obeys a statistic model along with time variation; when getting a new image frame, the Gaussian mixture distribution model is renewed, if the pixel in current image match the Gaussian mixture distribution model, then determining the pixel is a background point; otherwise, determining the pixel is a foreground point.

The present invention relates to a method for the real-time identification of seizures in an Electroencephalogram (EEG) signal. The method provides for patient-independent seizure identification by use of a multi-patient trained generic Support Vector Machine (SVM) classifier. The SVM classifier is operates on a large feature vector combining features from a wide variety of signal processing and analysis techniques. The method operates sufficiently accurately to be suitable for use in a clinical environment. The method may also be combined with additional classifiers, such a Gaussian Mixture Model (GMM) classifier, for improved robustness, and one or more dynamic classifiers such as an SVM using sequential kernels for improved temporal analysis of the EEG signal.

An “Image Segmenter” provides a variational energy formulation for segmentation of natural objects from images. In general, the Image Segmenter operates by adopting Gaussian mixture models (GMM) to capture the appearance variation of objects in one or more images. A global image data likelihood potential is then computed and combined with local region potentials to obtain a robust and accurate estimation of pixel foreground and background distributions. Iterative minimization of a “global-local energy function” is then accomplished by evolution of a foreground/background boundary curve by level set, and estimation of a foreground/background model by fixed-point iteration, termed “quasi-semi-supervised EM.” In various embodiments, this process is further improved by providing general object shape information for use in rectifying objects segmented from the image.

The invention provides a method and device for detecting road traffic abnormal events in real time. The method includes the steps of monitoring a road, obtaining a plurality of frames of continuous monitor images, extracting bright white segments from the monitor images, obtaining lane lines and lane end points through processing, building a lane model, determining a bidirectional detection area of a lane according to the lane model, detecting a moving object in the bidirectional detection area according to a Gaussian mixture model background subtraction method, determining the position of the moving object, building the mapping relation between the moving target and an actual vehicle according to the position of the moving target in the multiple frames of continuous monitor images by the adoption of a posterior probability splitting and merging algorithm and a feature point matching and tracking method, obtaining the running track and running speed of the actual vehicle, detecting the lane model and the running track and running speed of the actual vehicle according to a prestored road traffic abnormal behavior semantic model, and judging whether the road traffic abnormal events exist or not. The method has the advantages of being intelligent, high in accuracy and the like.

The invention discloses a video flame-detecting method based on multi-features fusion, which is characterized in that: a fast Gaussian hybrid model is established and a pixel counter is arranged when a computer reads a first frame image a monitoring site captured by a colored camera; when the computer reads new images of a second frame and the following frames, a fast Gaussian hybrid method is firstly used for carrying out motion detection so as to extract a moving object and then a flame color decision is carried out to the moving object to obtain an alternative flame area; and finally, flicker analysis is carried out on the alternative flame area so as to identify a real flame object. The video flame-detecting method carries out modeling to flame color characteristics, motion characteristics and flickering characteristics in flame areas; various flame detection interfering resources are gradually and rapidly eliminated in accordance with the recognition sequence of motion, color and flickering. The test results show that the method has relatively strong robustness; in addition, based on an AMD 2.04GHz processor, the processing speed can reach 22 frames per second for video images with pixels of 320 multiplied by 240.

The present disclosure provides systems and methods for virtual facial makeup simulation through virtual makeup removal and virtual makeup add-ons, virtual end effects and simulated textures. In one aspect, the present disclosure provides a method for virtually removing facial makeup, the method comprising providing a facial image of a user with makeups being applied thereto, locating facial landmarks from the facial image of the user in one or more regions, decomposing some regions into first channels which are fed to histogram matching to obtain a first image without makeup in that region and transferring other regions into color channels which are fed into histogram matching under different lighting conditions to obtain a second image without makeup in that region, and combining the images to form a resultant image with makeups removed in the facial regions. The disclosure also provides systems and methods for virtually generating output effects on an input image having a face, for creating dynamic texturing to a lip region of a facial image, for a virtual eye makeup add-on that may include multiple layers, a makeup recommendation system based on a trained neural network model, a method for providing a virtual makeup tutorial, a method for fast facial detection and landmark tracking which may also reduce lag associated with fast movement and to reduce shaking from lack of movement, a method of adjusting brightness and of calibrating a color and a method for advanced landmark location and feature detection using a Gaussian mixture model.

A humming transcription system and methodology is capable of transcribing an input humming signal into a standard notational representation. The disclosed humming transcription technique uses a statistical music recognition approach to recognize an input humming signal, model the humming signal into musical notes, and decide the pitch of each music note in the humming signal. The humming transcription system includes an input means accepting a humming signal, a humming database recording a sequence of humming data for training note models and pitch models, and a statistical humming transcription block that transcribes the input humming signal into musical notations in which the note symbols in the humming signal is segmented by phone-level Hidden Markov Models (HMMs) and the pitch value of each note symbol is modeled by Gaussian Mixture Models (GMMs), and thereby output a musical query sequence for music retrieval in later music search steps.

The invention discloses a barrier detection method in the vegetation environment based on multispectral and 3D feature fusion. The method comprises that 3D point cloud data is collected, and color and near-infrared images are collected at the same time; 3D point cloud data is registered with multispectral data to obtain multispectral 3D point cloud; a grid map is established, and candidate barrier grids are obtained via a height threshold; the scale of the near-infrared image is adjusted to normalize the infrared light intensity; the near-infrared intensity value and the visible light intensity of RGB color information within the grids are combined and serve as feature information, and 2D feature is obtained via calculation; and the candidates barrier grids are processed and filtered by utilizing a Gaussian mixture model to obtain a final barrier detection result. According to the method, influence caused by change of illumination condition is reduced by normalizing the infrared light intensity, and multispectral 3D features are sued to realize effective barrier detection.

The invention relates to a special audio event layered and generalized identification method based on a combination of an SVM (Support Vector Machine) and a GMM (Gaussian Mixture Model), and belongs to the technical field of a computer and audio event identification. The special audio event layered and generalized identification method comprises the following steps of: firstly, obtaining an audio characteristic vector file of a training sample; secondly, respectively carrying out model training on a great quantity of audio characteristic vector files (of the training samples) with various types by using a GMM method and an SVM method, so as to obtain the GMM model with generalization capability and an SVM classifier, and complete offline training; and finally, carrying out layered identification on the audio characteristic vector files to be identified by using the GMM model and the SVM classifier. With the adoption of the method provided by the invention, the problems that the conventional special audio event identification is low in identification efficiency on a continuous audio stream, very short in continuing time, high in audio event false dismissal probability can be solved. The method can be applied to searching a special audio and monitoring a network audio based on contents.

The invention provides a forest point cloud sorting method based on mode recognition, and belongs to the research field of forest canopy structure parameter obtaining methods. The method comprises the following steps of: obtaining three-dimensional laser point cloud data of the forest, selecting training sample of each sort, and computing obvious feature of each training sample; initially sorting the point cloud data by using an algorithm for obtaining Gauss mixed model based on expectation maximization algorithm, sorting the initial sorting result by using a filter and then processing. Compared with the traditional and existing method of obtaining the canopy structure parameter by using LIDAR, the method provided by the invention is time-saving and labor-saving, free from damaging the forest canopy structure, strong in applicability and high in precision.

The invention discloses a deep learning based medical image classification device and a construction method thereof. The device comprises an input module, a coarse classification module, a fine classification module, an integration module and a display module; the coarse classification module comprises a regional convolutional neural network RCNN; the fine classification module comprises a first cyclic convolutional neural network rCNN1 for identifying an original image, a HOG (Histogram of Oriented Gradient) model which converts an image to an HOG, a support vector machine (SVM) for identifying the HOG, a Gaussian mixture model (GMM) and a second cyclic convolutional neural network rCNN2; and the integration module comprises an integrated classifier as a GMM, integrates identification confidence scores of different areas output by the four classifiers into one input vector, inputs the input vector after weighting, and obtains a final identification confidence score of the different areas.

The present invention is directed to computer-based methods and system to perform root-cause analysis on an industrial process. The methods and system load process data for an industrial process from a historian database and build a hybrid first-principles and inferential model. The methods and system then executes the hybrid model to generate KPIs for the industrial process using the loaded process variables. The methods and system then selects a subset of the KPIs to represent an event occurring in the industrial process, and divides the data for the subset into multiple subset of time series. The system and methods select time intervals from the time series based on the data variability in the selected time intervals and perform a cross-correlation between the loaded process variables and the selected time interval, resulting in a cross-correlation score for each loaded process variable. The methods and system then select precursor candidates from the loaded process variables based on the cross-correlation scores and execute a parametric model for performing quantitative analysis of the selected precursor candidates, resulting in a strength of correlation score for each precursor candidate. The methods and system select root-cause variables from the selected precursor candidates based on the strength of correlation scores for analyzing the root-cause of the event.

The invention provides a video analysis based abnormal behavior detection method and system. The video analysis based abnormal behavior detection method comprises the following steps of extracting pedestrian foreground images from video frames; performing mesh generation on the video frames to divide the video frames into a plurality of mesh areas and setting the mesh areas in which the pedestrian foreground images are arranged to be movement areas; marking the movement areas through a nearest neighbor method and correlating the movement areas of the adjacent video frames; calculating light stream characteristics of the marked movement areas; obtaining a weighting direction histogram according to the light stream characteristics; calculating entropy of the weighting direction histogram; selecting a detection threshold value through a Gaussian mixture model, detecting whether an abnormal behavior is generated or not according to the detection threshold value and the entropy of the weighting direction histogram and updating the detection threshold value. The nested state machine based deduction process control method can automatically detect the abnormal behavior in a video scene, avoid the abnormal behavior disturbing the public plate order and endangering the public security and personal safety, reduce personnel workload and avoid potential risks caused by leak detection and error detection.

A method and system for detecting certain types of content, such as advertisements, using acoustical means from a media stream. The method uses two matching processes to detect and identify repeated content, the starting and end boundaries of which are then found. This content is used as the basis to find non-repeated content (such as less-frequently repeated advertisements) that are typically located in proximity to repeated content and can be evaluated using Gaussian mixture models (GMMs). The system that implements this method can be used for advertisement detection and monitoring for traditional media, such as television and radio, as well as for Internet-based media, such as streaming video, streaming audio and podcasts. The system can also be used to detect and identify copyrighted material in Internet traffic.

A plurality of image chips (202) (over 100), each of the chips containing the same, known target of interest, such as, for example an M109 tank are presented to the system for training. Each image chip of the known target is slightly different than the next, showing the known target at different aspect angles and rotation with respect to the moving platform acquiring the image chip.

The system extract multiple features of the known target from the plurality of image chips (202) presented for storage and analysis, or training. These features distinguish a known target of interest from the nearest similar target to the M109 tank, for example a Caterpillar D7 bulldozer. These features are stored for use during unknown target identification. When an unknown target chip is presented, the recognition algorithm relies on the features stored during training to attempt to identify the target.

The tools used for extracting features of the known target of interest as well as the unknown target presented for identification are the same and include the Haar Transform (404), and entropy measurements (410) generating coefficient locations. Using the Karhunen-Loeve (KL) transform 406, eigenvectors are computed. A Gaussian mixture model (GMM) (507) is used to compare the extracted coefficients and eigenfeatures from the known target chips with that of the unknown target chips. Thus the system is trained initially by presenting to it known target chips for classification. Subsequently, the system uses the training in the form of stored eigenfeatures and entropy coefficients fused with multiscale features to identify unknown targets.

The invention discloses a high-efficiency voice detecting method. The method comprises the following steps: analyzing the short-time energy and the short-time zero-crossing rate of an original audio frequency on a time domain and removing parts of non-voice signals; analyzing the spectral envelop characteristic and the entropy characteristic of a preserved audio frequency signal subband on a frequency domain and further removing parts of non-voice signals; forming an audio frequency segment by continuous frames with similar characteristics in each preserved frame of audio frequency signals; calculating the average value of Mel-frequency Cepstral coefficient of each frame in each audio frequency, respectively inputting the average values into a voice gaussian mixture model and various non-voice gaussian mixture models, and performing band-level judgment on whether the audio frequency segment contains voice data according to the output probability of each model, thereby finally obtaining a voice detecting result. The method is capable of detecting voice signals from audio frequency data streams under various complex environments, and positioning the boundary between voice segment data and non-voice segment data relatively correctly.

A speech enhancement method (and concomitant computer-readable medium comprising computer software encoded thereon) comprising receiving samples of a user's speech, determining mel-frequency cepstral coefficients of the samples, constructing a Gaussian mixture model of the coefficients, receiving speech from a noisy environment, determining mel-frequency cepstral coefficients of the noisy speech, estimating mel-frequency cepstral coefficients of clean speech from the mel-frequency cepstral coefficients of the noisy speech and from the Gaussian mixture model, and outputting a time-domain waveform of enhanced speech computed from the estimated mel-frequency cepstral coefficients.