185 results about "Random field ising model" patented technology

Apparatus and methods according to some embodiments of the present invention use a graphical model, such as a Markov random field model, to represent principal driving directions within an environment. The model has a plurality of nodes representing spatial locations within the environment, and the principal direction for each node is determined probabilistically using linear features detected within an image of the environment. Apparatus and methods according to embodiments of the present invention can be used in improved autonomous navigation systems, such as robotic vehicles.

The invention discloses a non-uniform image motion blur removing method based on a deep neural network. The non-uniform image motion blur removing method comprises two main steps of training deep neural network parameters, and estimating and removing motion blur by applying a deep network to an image. The deep network structure consists of a convolution network layer and a forward network layer; the network model training process comprises the following steps: generating a blur image block and a blur kernel pair by using a natural image, and training the deep neural network model parameters by using the data; providing a motion blur image, decomposing the image into an image block set with overlapped areas, inputting each image block into motion blur probability distribution corresponding to deep neural network output image blocks, and further acquiring different motion blur kernels per pixel points of the image by optimizing a Markov random field model; finally, acquiring a blur removed image by using a deconvolution algorithm based on the estimated motion blur kernels. Due to good learning capability of the deep neural network, non-uniform motion blur of the image can be precisely estimated, and clear images can be further obtained.

The invention discloses an automatic sleep staging method of a single-lead electroencephalogram. A training model comprises a feature extraction module and a staging optimization module, the feature extraction module comprises CNNs (convolutional neural networks) (1) and an Softmax layer (2), the staging optimization module comprises bi-directional LSTM (long-short term memory) recurrent neural networks (3) and a CRF (corticotropin releasing factor) conditional random field model (4), and the CNNs (1), the Softmax layer (2), the LSTM recurrent neural networks (3) and the CRF conditional random field model (4) are sequentially connected. The method only needs the single-lead sleep electroencephalogram, portable and comfortable sleep monitoring requirements are met, temporal and spatial characteristics of the electroencephalogram are sufficiently excavated according to the convolutional neural networks and the recurrent neural networks, the method has dynamic learning capacity and can adapt to great changed environments of diseases, the staging optimization module sufficiently considers relation between the front and the back of N 30s of electroencephalogram data, and the staging accuracy and the generalization ability of the model are improved.

The invention discloses a method for segmenting HMT images which is based on the nonsubsampled Contourlet transformation. The method mainly solves the problem that the prior segmentation method has poor area consistency and edge preservation, and comprises the following steps: (1) performing the nonsubsampled Contourlet transformation to images to be segmented and training images of all categories to obtain multi-scale transformation coefficients; (2) according to the nonsubsampled Contourlet transformation coefficients of the training images and the hidden markov tree which represents the one-to-one father and son state relationship, reckoning the model parameters; (3) calculating the corresponding likelihood values of the images to be segmented in all scale coefficient subbands, and classifying by examining possibility after integrating a labeled tree with a multi-scale likelihood function to obtain the maximum multi-scale; (4) updating category labels for each scale based on the context information context-5 model; and (5) with the consideration of the markov random field model and the information about correlation between two adjacent pixel spaces in the images to be segmented, updating the category labels to obtain the final segmentation results. The invention has the advantages of good area consistency and edge preservation, and can be applied to the segmentation for synthesizing grainy images.

A method and apparatus for detecting vascular landmarks in a 3D image volume, such as a CT volume, is disclosed. One or more guide slices are detected in a 3D image volume. A set of landmark candidates for multiple target vascular landmarks are then detected based on the guide slices. A node potential value for each landmark candidate is generated based on an error value determined using spatial histogram-based error regression, and edge potential values for pairs of landmark candidates are generated based on a bifurcation analysis of the image volume using vessel tracing. The optimal landmark candidate for each target landmark is then determined using a Markov random field model based on the node potential values and the edge potential values.

The invention discloses an image significance detection method combining color and depth information. The method comprises the following steps: performing superpixel segmentation on a to-be-detected color image, calculating a region contrast image in each segmented area through combining depth and color features, and obtaining a depth prior image and a direction prior image by utilizing depth information; integrating the region contrast image, the depth prior image and the direction prior image, and obtaining a contrast image integrated with prior information through calculation; performing overall optimization on the contrast image integrated with prior information: executing the normal inner product weighted webpage ranking algorithm, selecting an area with high confidence coefficient as a sampling area, designing an image restoration problem based on a Markov random field model, and solving to obtain a final significance detection image. According to the invention, the influence of the depth and direction information on significance is explored, and compared with the existing image significance detection method combining color and depth information, the method provided by the invention achieves a better effect.

The present invention is a method of generating a model of a random field which has directionally varying continuity. First, a tentative model for the random field is specified. Second, connected strings of nodes within the model are identified. Third, a spectral simulation on each of the strings of nodes is performed to determine updated values for the random field. Finally, the tentative model is updated with the data values from the spectral simulations.

The invention provides a method and a system for discovering new words in open fields. The method includes receiving corpora to be processed, conducting format conversion and word segmentation processing on the corpora to obtain a plurality of text messages, extracting characteristic messages of the plurality of text messages, judging whether combinations of adjacent text messages of a part of the text messages in the plurality of text messages are new words, conducting new word boundary labeling on the adjacent text messages on yes judgment, estimating parameters of a conditional random field model according to the labeled text messages and characteristic messages and identifying the surplus text messages according to the estimated conditional random field model to obtain new words of the surplus text messages. By means of the method, the new word boundary labeling is conducted on the text messages, the parameters of the conditional random field model are estimated, the plurality of text messages are identified to obtain the new words in the text messages, the new words in various fields can be identified, and meanwhile identification efficiency is improved.

An image processing system automatically segments and labels an image using a hierarchical classification model. A global classification model determines initial labels for an image based on features of the image. A label-based descriptor is generated based on the initial labels. A local classification model is then selected from a plurality of learned local classification model based on the label-based descriptor. The local classification model is applied to the features of the input image to determined refined labels. The refined labels are stored in association with the input image.

The invention discloses a method for detecting remote sensing image change based on non-parametric density estimation, which mainly solves the problem that the estimation to the statistic items which relevant to a change type and a non-change type in a differential chart in the prior art has error. The realizing process of the method is that inputting two remote sensing images with different time-phase, removing noise of each channel of each image, obtaining noise-removing images of the two time-phase, and constructing difference images through adopting the change time-vector method, gathering the difference images into change type and a non-change type through applying K-means clustering algorism, obtaining the initial sorting results, and estimating the statistic items relevant to the change type and the non-change type in differential images through adopting non-parameter density estimation, carrying out the self-adapting space restriction combining the variable weight markov random field model, and obtaining the final change detecting results. The experimentation shows that the invention can effectively keeps the structure information of the images, removes insulation noise, improves the change detection processing efficiency, and can be used for the fields of disaster surveillance, land utilization and agriculture investigation.

The invention, which belongs to the technical field of image processing, discloses a video segmentation method based on strong target constraint video saliency. According to the method disclosed by the invention, strong target constraint is introduced based on image saliency. The location and scale constraint of a target are obtained by a multi-scale tracking algorithm and optical flow correction,color constraint information of the target is obtained based on a historical frame segmentation result, and calculation is carried out obtain a video saliency result; histogram classification is carried out on the video saliency result to obtain a tag mask graph, and foreground / background prior probability models of a current frame are calculated; a super-pixel-based time-space continuum full connection condition random field model is constructed at the current frame, data items are defined by using the prior probability models, an intra-frame smooth item and an inter-frame smooth item are defined by combining color distances, space distances and edge relationships between super pixels, and optimized solution is carried out by using a fast high-dimensional Gaussian filter algorithm to complete video target segmentation. Therefore, the accuracy and the time efficiency of video segmentation are improved.

The invention discloses an extraction method of a visual saliency area based on a monocular depth map. The extraction method of the visual saliency area comprises the following sequential steps: dividing an original image to obtain super pixels; building a characteristic vector of each super pixel to estimate an absolute depth characteristic of absolute depth in a scene in the image; building a probability model by using a Gaussian-Markov random field model, calculating a distance relation between the super pixel characteristic vector and an adjacent super pixel characteristic vector by virtue of the probability model to obtain a relative depth characteristic on the basis of the absolute depth characteristic and obtain depth values and depth maps of the super pixels simultaneously; calculating saliency values of the super pixels; calculating a gain coefficient according to the depth values and correcting the saliency values by using the gain coefficient. The extraction method can be used for quickly and automatically identifying the saliency objects in the image without any prior knowledge, is high in universality and can also be used for accurately detecting the saliency area.

The invention provides a metal artifact correction method, comprising: first, employing a method based on a Markov random field model to divide metal areas, thereby solving the problem of inaccurate division of metal areas by a traditional metal artifact removal algorithm, and avoiding more secondary artifacts caused by inaccurate positioning of metal projection borders; and employing a prior interpolation algorithm to solve the problem of unsmooth borders of metal projection areas after linear interpolation; in addition, metal edge framework structure information is fed back to corrected projection data, thereby solving the problem that metal edge high contrast substance structures may be lost in correction results by a traditional interpolation algorithm.

A method for detecting early indications of equipment failure in an industrial system includes receiving sensor training data collected from industrial equipment under normal conditions and identifying periods of time in the sensor training data when the equipment was functioning normally; finding a pattern for each identified period of time to initialize a plurality of mixture models; learning weighting factors, mean and variance of each of the plurality of mixture models, and removing unimportant models from the plurality of mixture models; determining a Gaussian Markov random field model from surviving mixture models by calculating gating functions for each of the variables and individual mixture models; determining a threshold value of an anomaly score for each variable from the sensor training data; and deploying the model to monitor sensor data from industrial equipment using the threshold values to detect anomalous sensor data values indicative of an impending system failure.

The invention belongs to the field of remote sensing image and particularly relates to a RPC model parameter extraction method and a RPC model geometric correction method. The technique scheme of the invention simulates elevation change to a random field and constructs Kriging equations according to the observed values of a plurality of random variables based on stationary random field theory, thus obtaining the optimal bias-free interpolation of unobserved points. The RPC model parameter extraction method provided by the invention can acquire high-accuracy PRC model parameters based on a stationary random field model to ensure the RPC geometric correction accuracy. The RPC model geometric correction method provided by the invention can ensure the accuracy of the final geometric correction result by carrying out crude correction based on a common RPC model and carrying out fine correction based on the stationary random field model. By adopting the stationary random field model to assist the RPC geometric correction, both the two methods can improve the robustness and the accuracy of geometric correction of the universal sensor for remote sensing image even if the assistant means are different.

The invention belongs to the technical field of SAR (synthetic aperture radar) image change detection, and discloses an SAR image change detection method based on a support vector machine and a discriminative random field. The SAR image change detection method based on the support vector machine and the discriminative random field includes the steps: normalizing gray values of two original time phase images, and extracting corresponding gray characteristic differences and textural characteristic differences in the processed images; forming difference characteristic vectors; extracting boundary strength of each pixel in a difference image by the aid of weighted average ratio operators; selecting training samples in the difference image, and expressing the training samples by the aid of the corresponding difference characteristic vectors to obtain initial category labels of testing samples and posterior probabilities of the category labels of the testing samples by the aid of the training support vector machine; obtaining initial support vector machine-discriminative random field models; updating the support vector machine-discriminative random field models to obtain final category labels and change detection results of the corresponding testing samples.

The invention provides a high-resolution SAR image classification method based on sparse features and conditional random fields. The method mainly solves the problems of low classification precision and non-accurate boundary retention in complicated scenes in the prior art. The method comprises the steps of: firstly, inputting high-resolution SAR images, selecting images to build a training data block set, and training system parameters of a sparse feature extraction algorithm; secondly, extracting SAR image block sparse features and training a logistics classifier to obtain the classificationposterior probability of the images and build a univariate potential energy function; thirdly, building a bivariate potential energy function by using a boundary constraint map obtained after fusionof a binary edge partition map and an edge strength map; forming a complete full connection conditional random field model by using the univariate potential energy function and the bivariate potentialenergy function and performing reasoning on the model to obtain a classification result. The method increases the classification precision of complicated scenes and edge details of high-resolution SAR images and can be used for SAR image terrain classification.

The present application relates to a method and device for constructing a news entity identification model based on migration learning, a computer device and a storage medium. The method comprises thefollowing steps of constructing a named entity recognition model; extracting the neural network parameters of the second neural network model in the pre-trained part-of-speech tagging model, and initializing the first neural network model of the named entity recognition model according to the neural network parameters; acquiring a news corpus training sample, wherein the first Chinese character in the news corpus training sample is labeled with a corresponding label; converting the first Chinese character into a first character vector, and inputting the first character vector into a first neural network model to obtain a first feature vector of the Chinese character; by using the first eigenvector corresponding to the first Chinese character and the corresponding label, carrying out the supervised training on the target conditional random field model, and obtaining the news entity recognition model. The method can improve the recognition effect of the news entity recognition model.

The invention discloses a method for detecting two-time-phase remote sensing image change. In the method for detecting two-time-phase remote sensing image change, a gaussian process classifier is combined with a Markov random field model, effective utilization of image space information is reinforced, and change detecting precision is improved.

Systems and methods for pre-training a sequence tagger with unlabeled data, such as a hidden layered conditional random field model are provided. Additionally, systems and methods for transfer learning are provided. Accordingly, the systems and methods build more accurate, more reliable, and / or more efficient sequence taggers than previously utilized sequence taggers that are not pre-trained with unlabeled data and / or that are not capable of transfer learning / training.

Systems and methods for using social network information to predict complex phenomena. According to one embodiment the system or method comprises a Support Vector Machine classifier utilized to infer a pre-determined state of an individual, location, or event based on information gathered from a social network dataset. A conditional random field model can then be used to predict an individual's propensity toward that pre-determined state using features derived from the social network dataset. Performance of the conditional random field model can be enhanced by including features that are not only based on the status of net work connections, but are also based on the estimated encounters with individuals having the pre-determined state, including individuals other than network connections.

A method and apparatus are provided for training and using a hidden conditional random field model for speech recognition and phonetic classification. The hidden conditional random field model uses features, at least one of which is based on a hidden state in a phonetic unit. Values for the features are determined from a segment of speech, and these values are used to identify a phonetic unit for the segment of speech.

The invention provides a target color recognition method for video surveillance. The method comprises the steps that a color recognition target area is extracted from a video image; colors of the target area are initialized; color category detection is carried out on all color components h, s and l of the target area; the proportion of each color of the color recognition target area is calculated, and the color with the largest proportion serves as the color of the color recognition target area. According to the method, the Markov random field model is adopted, influence on color recognition from environment changes and target posture changes can be reduced according to the neighborhood position relation of the color recognition target area, and therefore the color recognition rate can be increased, a color target can be effectively positioned in intelligent surveillance, and strong evidence can be provided for solving criminal cases like violation of laws and rules.

The invention is applicable to the technical field of image processing, and provides a depth image processing method and a depth image processing device. The method comprises the following steps: inputting a depth image and a corresponding color image, wherein the depth image and the color image are pre-registered and are of the same size; extracting a boundary image from the color image, and determining a boundary mask based on the boundary image; filtering the depth image in multiple preset directions through a preset filter based on the boundary mask to get corresponding multiple filtering candidate results; and fusing the multiple filtering candidate results into a depth information completion result of the depth image through a Markov random field model. According to the invention, the boundary of the color image is extracted, the boundary mask is introduced, the depth image is filled in multiple preset directions through an improved local bilateral filter, and the filling results of the multiple preset directions are fused into a final information completion result, and thus, the integrity of depth information is guaranteed.

The invention discloses a named entity identification method based on adversarial training. The method comprises the following steps: obtaining relevance features among judicial field characters through RoBERTa model training and Bi-LSTM training; splicing the two relevance features together, and predicting a training sample by using a conditional random field model to obtain a prediction result;according to the method, external word vectors and word vectors of different dimensions can be introduced to be combined with judicial domain text word mixing vectors of different dimensions, confrontation disturbance is carried out on the mixing word vectors in a judicial domain text, and the accuracy of model recognition is improved.

The invention discloses a face portrait synthesis method based on greedy search. The face portrait synthesis method comprises the following steps: dividing a dictionary training picture sample set A, synthesizing a training picture sample set Bp, synthesizing a training portrait sample set Bs and testing a picture sample set; blocking so as to obtain a dictionary training picture block set SA, synthesizing a training picture block set Sp, synthesizing a training portrait block set Ss and testing a picture block set; learning a training picture block characteristic dictionary Dp from the dictionary training picture block set SA; solving a sparse representation set Cp corresponding to Sp and testing sparse representation corresponding to the picture block S by using Dp; performing greedy search in Cp by using sparse representation so as to find neighbors; establishing a Markov random field model by using the searched neighbor portrait-picture blocks so as to synthesize portraits. Compared with a conventional method, the face portrait synthesis method does not need to limit the background of a tested picture, is capable of combining non-face components in the tested picture, and can be applied to criminal investigation and digital entertainment.

The invention relates to a semantic segmentation optimization method for an edge image. The method comprises the steps of selecting image data; Using the image data to train and verify an image semantic segmentation model and a full connection condition random field model; Obtaining a semantic segmentation result of the image by using the trained image semantic segmentation model; Obtaining a superpixel segmentation result of the image edge information by using a superpixel segmentation algorithm; Optimizing the semantic segmentation result by using the superpixel segmentation result to form afirst optimization result; And optimizing the first optimization result by using the trained full connection condition random field model. According to the method provided by the invention, the advanced semantic information in the image can be effectively extracted, the image edge information is reserved through the super-pixel segmentation algorithm, the semantic segmentation accuracy of an existing segmentation model on the image edge is improved through the local edge optimization algorithm, and the method is flexible to implement, high in compatibility and relatively high in robustness.

The invention discloses an automatic filling method for a medical health record based on speech recognition. The method comprises that a training set file used for speech recognition training is obtained in advance; a speech recognition model of a deep neural network is called, and the model is trained to obtain an improved speech recognition model; a condition random field model is called and trained to obtain an information extraction model; voice data to be recognized is input, recognized by the improved speech recognition model automatically, and converted into text data; and feature of each word is obtained to form text feature data, the text feature data is marked, labels of each word in the text feature data is obtained, the words are logged into corresponding data according to the labels, and a record filled automatically is obtained. Compared with an automatic filling result in the prior art, objective facts and main diagnosis are combined, the reliability is very high, medical staff is effectively freed from tedious work of writing the health records of patients manually, the method is simple, and the accuracy is high.