30 results about "Earth mover's distance" patented technology

The invention relates to a method for automatically classifying text documents by utilizing a body, comprising the following steps: firstly expressing the characteristic information of a text document by utilizing a weighted key word set; and then expressing the characteristic information of a classifying catalogue by a body which is subject to body disambiguation and body expansion; transforming the body into a weighted word meaning set through analyzing the body structural characteristic; finally calculating the semantic similar value between the key word set of the text document and the body weighted word meaning set by utilizing a Earth Mover's Distance method; further calculating the similar value between the text document and the classifying catalogue; and classifying and sequencing the text document according to the similar value between the text document and the classifying catalogue. By utilizing the method of the invention, the text document can be automatically classified, and the accuracy of the text document classification can be improved.

A new measure to compare histogram-based descriptors, a diffusion distance, is disclosed. The difference between two histograms is defined to be a temperature field. The relationship between histogram similarity and diffusion process is discussed and it is shown how the diffusion handles deformation as well as quantization effects. As a result, the diffusion distance is derived as the sum of dissimilarities over scales. Being a cross-bin histogram distance, the diffusion distance is robust to deformation, lighting change and noise in histogram-based local descriptors. In addition, it enjoys linear computational complexity which significantly improves previously proposed cross-bin distances with quadratic complexity or higher The proposed approach is tested on both shape recognition and interest point matching tasks using several multi-dimensional histogram-based descriptors including shape context, SIFT and spin images. In all experiments, the diffusion distance performs excellently in both accuracy and efficiency in comparison with other state-of-the-art distance measures. In particular, it performs as accurate as the Earth Mover's Distance with a much greater efficiency.

A method of perceptual 3-dimensional (3D) shape description and a method and apparatus for searching a perceptual 3D graphics model database established using the description method are provided. The description method includes: generating nodes that respectively correspond to parts of a part-based representation of a 3D shape model, the nodes including unary attributes of the parts; generating edges that include relational attributes between the nodes; and generating an attributed relational graph of the 3D shape model that is comprised of the nodes and the edges. The search method includes: receiving a predetermined 3D graphics model; transforming the received 3D graphics model into a perceptual 3D shape descriptor; and comparing the perceptual 3D shape descriptor with each of the perceptual 3D graphics models stored in the database to retrieve the 3D graphic models that are similar to the perceptual 3D shape descriptor. The searching apparatus includes: a query input unit that receives a query that is a 3D graphics model; a model / shape descriptor transforming unit that transforms the 3D graphic model received as the query into a perceptual 3D shape descriptor; a matching unit that compares the perceptual 3D shape descriptor with each of the perceptual 3D graphics models stored in the database to retrieve the models that are similar to the perceptual 3D shape descriptor; and a model output unit that outputs the retrieved model. A query by sketch or a query by editing is available, and the models that are similar to a query can be more accurately retrieved due to a double earth mover's distance method used to match query and model graphs.

The invention discloses a video event recognition method based on a top-down motion attention mechanism, which comprises the following steps of: 1, detecting points of interest of each frame in each video in a video set on a computer by using Gaussian difference detector, wherein the video set comprises a training video set and a testing video set; 2, extracting scale-invariant characteristic description sub-characteristics and light stream characteristics from the detected points of interest of each frame; 3, establishing an apparent word list and a motion word list; 4, learning the probability of each motion word about each type of events on the training video set and establishing a motion information-based attention histogram; 5, calculating the similarity between videos in the video set by using the distance of a bulldozer, and generating a kernel function matrix; and 6, training a support vector machine classifier by using the obtained kernel function matrix so as to obtain classifier parameters, classifying the tested video sets, and outputting classification results.

The invention discloses a large-scale image data similarity searching method based on an EMD (earth mover's distance). The method comprises the following steps that an image data mapping function f used for mapping to a one-dimension real number key value space Omega(phi) is designed; an operation MR1 is started, and a load of each key value in the Omega(phi) is estimated; the operation MR2 is started, the cutting is carried out on the Omega(phi) through a Map task on the basis of the estimated key value load, and data corresponding to the cutting region are sent to a Reduce task in a segmented way; image data received by each Reduce task is mapped to the key values in the Omega(phi) on the basis of f, and an index structure oriented to the EMD is built on the basis of the key values; the similarity searching based on the EMD is executed on the basis of the index structure; execution results of each Reduce task based on EMD similarity searching in the MR2 are subjected to union set taking and output. The large-scale image data similarity searching method has the advantages that the network transmission data quantity is lower, the calculation load distribution is more balanced, the similarity searching efficiency is higher, and the big data set analysis and processing expandability is better.

The invention relates to a wood identification method. The method comprises the following steps of: (1) generating characteristic quantities, namely partitioning a wood image, extracting local characteristics of each partitioned image, combining the local characteristics of a plurality of partitioned images to form the characteristics of the wood image, clustering and merging the characteristics of the wood image by clustering, extracting clustered information, and on the basis, screening the clustered information to form the characteristic quantities of the wood image; (2) repeating the step (1) for each image in an N original wood images-containing training sample set so as to generate an N characteristic quantities-containing characteristic set; (3) repeating the step (1) for the wood images to be identified so as to generate the characteristic quantities of the wood images to be identified; and (4) calculating the characteristic quantities of the wood images to be identified and the characteristic set by an earth mover's distance (EMD) algorithm to generate N EMDs, and classifying the N EMDs to obtain an identification result. The method has higher identification rate.

A method of aligning image having the steps of obtaining a first image, the first image having a corresponding first data set, obtaining a second image, the second having a corresponding second data set; learning a joint intensity distribution from a pair of prealigned images, and aligning the first image and the second image by computing Earth Mover's Distance between their observed joint intensity distribution and the learned joint intensity distribution.

The invention provides a labelling image scene clustering method based on vision and labelling character related information. The method comprises the following steps of: dividing a training image and a test image respectively by using a NCut (Normalized Cut) image dividing algorithm; constructing a vision nearest-neighbour graph G(C)(V, E) of all images {J1, ., Jl} PCtrain for learning, wherein in a training image set, each image has one group of initial normalized labelling character weight vectors; spreading the labelling character of each training image among the vision nearest neighbours, receiving the accepted images according to the degree of normalized EMD (Earth Mover's Distance) among the accepted images; for each training image, normalizing the accumulated labelling character weights; after the vision characteristics of the image are converted into a group of labelling characters with weights, carrying out the scene semantic clustering by using a PLSA (Probabilistic Latent Semantic Analysis) model; learning each scene semantic vision space by using a Gaussian mixture model; and carrying out the scene classification by using the vision characteristics. With the invention, the coupling precision between the vision characteristics of the image and the labelling character can be increased, and the method can be directly used for the automatic semantic labelling of the image.

The invention provides a target tracking method and a target tracking device in video monitoring. The method comprises the steps that the state value corresponding to a target in a previous frame is obtained; a current frame is subjected to target detection, and the observed value corresponding to the target observing in the current frame is obtained; a cost matrix is built according to the state value and the observing value, in addition, an EMD (earth mover's distance) algorithm is applied for solving the cost matrix, and an allocation matrix is obtained; and the target in the current frame is recognized according to the allocation matrix. The device comprises an obtaining module, a target detecting module, an operation module and a recognition module, wherein the obtaining module is used for obtaining the state value corresponding to the target in the previous frame, the target detecting module is used for carrying out the target detection on the current frame to obtain the observed value corresponding to the target observing in the current frame, the operation module is used for building the cost matrix according to the state value and the observing valve and applying the EMD algorithm for solving the cost matrix to obtain the allocation matrix, and the recognition module is used for recognizing the target in the current frame according to the allocation matrix. When the target tracking method and the target tracking device are adopted, the accuracy can be improved.

The invention relates to a parallel sentence pair screening method and system. The method comprises the following steps: a source language statement and a target language statement of each to-be-screened sentence pair are divided into words; the word vector of each word obtained through dividing is determined with a bilingual word vector model; the weight value of each word in the source language statement is calculated; the weight value of each word in the target language statement is calculated; an objective function is established, the optimal solution of the objective function is calculated, and the minimum earth mover's distance of each to-be-screened sentence pair is determined according to the optimal solution; the parallel sentence pair screening standard is determined according to the minimum earth mover's distances of a plurality of to-be-screened sentence pairs, and parallel sentence pair screening is performed according to the parallel sentence pair screening standard. The method is specially designed for parallel sentence pair screening work, the condition that all corpora are inter-translated is not supposed, large quantity of rough bilingual corpora on the internet can be screened, and high-quality and reliable bilingual corpora can be obtained.

The invention relates to a data stream similarity connection method. According to a preset time span value P, a B+tree forest set index is constructed on a data stream R; and when the timestamps of data tuples on the data stream R and a data stream S are within a time range of a current sliding window, on the basis of the B+tree forest set index, similarity connection between the data stream R and the data stream S on the basis of an EMD (Earth Mover's Distance) under sliding window semantics is carried out. By use of the method, on the basis of the +tree forest set index, a similarity connection method based on the sliding window semantics and the EMD on the data stream can be designed on the basis of the B+tree forest set index, a solution is put forward for the data stream similarity connection on the basis of the EMD under the sliding window semantics, and the processing efficiency and performance of the similarity connection is obviously improved.

A Compressible Earth Mover's Distance (CEMD) better matches how humans perceive distance between distributions. Earth and holes are able to be compressed when moving from the earth to the holes thus minimizing the distance and improving the quality of image comparison. CEMD is utilized in a number of implementations, for instance, content based image retrieval, color query and other applications in multimedia. Another implementation includes using CEMD with a content recognition system for indexing occurrences of objects within an audio / video content data stream which processes the stream of data to generate a content index database corresponding to the content stream. CEMD is usable with in a variety of systems to assist in image recognition.

A Compressible Earth Mover's Distance (CEMD) better matches how humans perceive distance between distributions. Earth and holes are able to be compressed when moving from the earth to the holes thus minimizing the distance and improving the quality of image comparison. CEMD is utilized in a number of implementations, for instance, content based image retrieval, color query and other applications in multimedia. Another implementation includes using CEMD with a content recognition system for indexing occurrences of objects within an audio / video content data stream which processes the stream of data to generate a content index database corresponding to the content stream. CEMD is usable with in a variety of systems to assist in image recognition.

A method of measuring distance between two distributions of data points in a space is described. The distances between data points are ordered in a plurality of heaps stored within 2-D heaps, which is constructed by giving heap structure to a list of one-dimensional heaps. Each plurality of heaps is ordered so that the minimum distance is the root node. The plurality of heaps within the 2-D heaps are ordered to determine the maximum of the root nodes. The maximum of the 2-D heaps are compared to determine the Hausdorff distance which is the greater of the maximums. The corresponding heaps are removed from their respective 2-D heaps. The Hausdorff distance is utilized in a set of calculations to generate an EMD. Determining the Hausdorff distance and generating the EMD is repeated until no more earths and no more holes remain. After all earths are moved to all holes, the work is established, and the similarity of the plurality of images is determined.

A method of aligning image having the steps of obtaining a first image, the first image having a corresponding first data set, obtaining a second image, the second having a corresponding second data set; learning a joint intensity distribution from a pair of prealigned images, and aligning the first image and the second image by computing Earth Mover's Distance between their observed joint intensity distribution and the learned joint intensity distribution.

The invention provides a labelling image scene clustering method based on vision and labelling character related information. The method comprises the following steps of: dividing a training image and a test image respectively by using a NCut (Normalized Cut) image dividing algorithm; constructing a vision nearest-neighbour graph G(C)(V, E) of all images {J1, ., Jl} PCtrain for learning, wherein in a training image set, each image has one group of initial normalized labelling character weight vectors; spreading the labelling character of each training image among the vision nearest neighbours, receiving the accepted images according to the degree of normalized EMD (Earth Mover's Distance) among the accepted images; for each training image, normalizing the accumulated labelling character weights; after the vision characteristics of the image are converted into a group of labelling characters with weights, carrying out the scene semantic clustering by using a PLSA (Probabilistic Latent Semantic Analysis) model; learning each scene semantic vision space by using a Gaussian mixture model; and carrying out the scene classification by using the vision characteristics. With the invention, the coupling precision between the vision characteristics of the image and the labelling character can be increased, and the method can be directly used for the automatic semantic labelling of the image.

A method of measuring distance between two distributions of data points in a space is described. The distances between data points are ordered in a plurality of heaps stored within 2-D heaps, which is constructed by giving heap structure to a list of one-dimensional heaps. Each plurality of heaps is ordered so that the minimum distance is the root node. The plurality of heaps within the 2-D heaps are ordered to determine the maximum of the root nodes. The maximum of the 2-D heaps are compared to determine the Hausdorff distance which is the greater of the maximums. The corresponding heaps are removed from their respective 2-D heaps. The Hausdorff distance is utilized in a set of calculations to generate an EMD. Determining the Hausdorff distance and generating the EMD is repeated until no more earths and no more holes remain. After all earths are moved to all holes, the work is established, and the similarity of the plurality of images is determined.

The invention discloses a method for measuring dynamic characteristic similarity of a complex system. A recursive analysis result has rigidity and binary value property due to the fact that a Heaviside step function is used, so an unreliable research result is produced, the slight change of a critical distance or the change of the length and the position of a time sequence can cause the obvious change of a research result; a local binary model (LBP) and a texture similarity measurement Earth Mover's Distance model (EMD) are used to propose a new idea for texture analysis of a Gaussian recursive graph for complex system dynamics characteristic analysis, and on the basis, a method system for measuring the dynamic characteristic similarity of the complex system is constructed.

The invention relates to a method for automatically classifying text documents by utilizing a body, comprising the following steps: firstly expressing the characteristic information of a text document by utilizing a weighted key word set; and then expressing the characteristic information of a classifying catalogue by a body which is subject to body disambiguation and body expansion; transforming the body into a weighted word meaning set through analyzing the body structural characteristic; finally calculating the semantic similar value between the key word set of the text document and the body weighted word meaning set by utilizing a Earth Mover's Distance method; further calculating the similar value between the text document and the classifying catalogue; and classifying and sequencing the text document according to the similar value between the text document and the classifying catalogue. By utilizing the method of the invention, the text document can be automatically classified, and the accuracy of the text document classification can be improved.

The invention relates to a data stream similarity connection method. According to a preset time span value P, a B+tree forest set index is constructed on a data stream R; and when the timestamps of data tuples on the data stream R and a data stream S are within a time range of a current sliding window, on the basis of the B+tree forest set index, similarity connection between the data stream R and the data stream S on the basis of an EMD (Earth Mover's Distance) under sliding window semantics is carried out. By use of the method, on the basis of the +tree forest set index, a similarity connection method based on the sliding window semantics and the EMD on the data stream can be designed on the basis of the B+tree forest set index, a solution is put forward for the data stream similarity connection on the basis of the EMD under the sliding window semantics, and the processing efficiency and performance of the similarity connection is obviously improved.

The invention discloses an LTE distributed system antenna fault detection method. Firstly rapid database creating and adaptive partition of the space region are completed by using pedestrian track reckoning and the K-means clustering algorithm so as to ensure that the regional data have significant characteristics and then the regional database is constructed; then different distribution distancemodels are constructed according to the regional signal distribution situation and EMD between the online measured data and the database is completed by using the improved Earth Mover's Distance (EMD)algorithm so as to complete preliminary selection of the similar regions; and finally the user regional location estimation is realized by the EMD-based multi-layer classification algorithm and antenna troubleshooting is performed in the final location region by using the related test equipment so as to find the location of the fault antenna.

The invention relates to a parallel sentence pair screening method and system. The method comprises the following steps: a source language statement and a target language statement of each to-be-screened sentence pair are divided into words; the word vector of each word obtained through dividing is determined with a bilingual word vector model; the weight value of each word in the source language statement is calculated; the weight value of each word in the target language statement is calculated; an objective function is established, the optimal solution of the objective function is calculated, and the minimum earth mover's distance of each to-be-screened sentence pair is determined according to the optimal solution; the parallel sentence pair screening standard is determined according to the minimum earth mover's distances of a plurality of to-be-screened sentence pairs, and parallel sentence pair screening is performed according to the parallel sentence pair screening standard. The method is specially designed for parallel sentence pair screening work, the condition that all corpora are inter-translated is not supposed, large quantity of rough bilingual corpora on the internet can be screened, and high-quality and reliable bilingual corpora can be obtained.

The present invention claims to protect a feature-aligned Chinese word segmentation method, which includes: 101 extracting features of bigrams from marked data and unmarked data; 102 combining marked data with unmarked data by the Earth Mover's Distance (hereinafter referred to as EMD) method 103 Use the classifier xgboost to train the features of the labeled data after feature alignment, so as to predict the probability of big words in unlabeled data becoming words; 104 Extract some big words and steps from the results of the classifier 101 The bigram integration of labeled data is used as the feature of the conditional random field and trained; 105 The unlabeled data is sequence-labeled and segmented through the established model. The invention mainly performs feature alignment on marked data and unmarked data through EMD, predicts the word formation probability of binary words through classifier learning, and then integrates conditional random fields in a stacking manner to form a new tokenizer.

The invention discloses a blind hyperspectral unmixing model construction method based on Sinkhorn distance, and the method comprises the steps: replacing an Euclidean distance with an Earth Mover's Distance (EMD), and overcoming the noise influence; improving the EMD into a Sinkhorn distance through entropy regularization constraint, and modeling the relationship between different feature dimensions, so that the correlation between the features is ensured; and based on a manifold learning theory, introducing a graph regularization term to maintain a local geometric structure between data. According to the method, the problems that the traditional Euclidean distance is easily influenced by noise and correlation characteristics in an image space are ignored are solved by constructing the unmixing framework based on the Sinkhorn distance. According to the characteristic that the EMD is insensitive to the relation between the features of different dimensions, the model takes the Sinkhorn distance as the standard of error measurement, the features on different dimensions can be effectively modeled, and the correlation between the features is fully developed and utilized. The unmixing performance of the proposed model is quantitatively evaluated by adopting a Lagrange function method and a KKT condition, and the feasibility and superiority of the unmixing model are proved.

The invention discloses an EMD (earth mover's distance) algorithm based individual trip mobile phone switching sequence road matching method. The method includes firstly, calibrating a main mobile phone base station switching sequence corresponding to each road, secondly, calculating EMD values between a to-be-matched switching sequence generated by individual trip and the known calibrated main switching sequence, and finally selecting one calibration rod corresponding to the minimum EMD value, namely the road matched with the individual trip. The method has the advantages that mobile phone switching sequence data are constructed into the corresponding EMD modes by utilization of data information such as the mobile phone base station switching sequence, switching time and the like provided by communication operators, then the mobile phone switching modes of different roads are analyzed and classified according to the EMD values, and after new unknown individual mobile phone switching data are acquired, the user travel path can be acquired and road map matching is realized by only mode classification of the new unknown individual mobile phone switching data. Under the current background of high popularizing rate of mobile phones, the method can be applied to large-range resident trip path information recognition and acquisition.