141 results about "Spatial similarity" patented technology

Provided are a method and apparatus for estimating a motion vector using a plurality of motion vector predictors, an encoder, a decoder, and a decoding method. The method includes calculating spatial similarities between the current block and the plurality of neighboring partitions around the current block, selecting at least one of the neighboring partitions based on the calculated spatial similarities, and estimating a motion vector of the selected partition as the motion vector of the current block.

A method for matching documents based on spatial layout of regions based on a shape similarity model for detecting similarity between general 2D objects. The method uses the shape similarity model to determine if two documents are similar by logical region generation in which logical regions are automatically derived from information in the documents to be matched, region correspondence, in which a correspondence is established between the regions on the documents, pose computation in which the individual transforms relating corresponding regions are recovered, and pose verification in which the extent of spatial similarity is measured by projecting one document onto the other using the computed pose parameters.

The invention discloses video noise reduction device and a video noise reduction method. The method comprises the following steps of: obtaining a brightness difference histogram of a current image by using a denoising result of a previous frame of image and a gradient magnitude histogram of the current image; carrying out noise level evaluation on the current image according to the brightness difference histogram; calculating the spatial distance of any two pixel points in the current image, so as to obtain the spatial similarity of the any two pixel points; carrying out denoising on the current image according to the spatial similarity; calculating a pixel time domain distance between any pixel point in the current image and the pixel point at the position corresponding to the previous frame of denoised image, and calculating the corresponding time domain similarity; carrying out three-dimensional recursive denoising on the video image according to the obtained time domain similarity, the spatial similarity denoising result and the previous frame of denoising result. By adopting the device and method disclosed by the invention, three-dimensional recursive denoising is carried out by using the pertinence of the pixel in space and time, so that strong complicated noise can be removed; an image detail can be kept; the stability of the denoising effect also can be ensured.

Methods and systems for knowledge extraction and prediction are described. In an example, a computerized method, and system for performing the method, can include receiving historical data pertaining to a domain of interest, receiving predetermined heuristics design data associated with the domain of interest, and using the predetermined heuristics design and historical data, automatically creating causal maps including a hierarchy of nodes, each node of the hierarchy of nodes being associated with a plurality of quantization points and reference temporal patterns, the plurality quantization points being known reference spatial patterns. In an example the computerized method, and system for performing the method, can further include receiving, at each node, a plurality of unknown patterns pertaining to a cause associated with the domain of interest, automatically mapping the plurality of unknown patterns to the quantization points using spatial similarities of the unknown patterns and the quantization points, automatically pooling the quantization points into a temporal pattern, the temporal pattern being a sequence of spatial patterns that represent the cause, automatically mapping the temporal pattern to a reference temporal pattern, automatically creating a sequence of the temporal patterns, and automatically recognizing the cause using the sequence of temporal patterns.

A method for generating spatial-temporal consistency depth map sequences based on convolutional neural networks for 2D-3D conversion of television works includes steps of: 1) collecting a training set, wherein each training sample thereof includes a sequence of continuous RGB images, and a corresponding depth map sequence; 2) processing each image sequence in the training set with spatial-temporal consistency superpixel segmentation, and establishing a spatial similarity matrix and a temporal similarity matrix; 3) establishing the convolution neural network including a single superpixel depth regression network and a spatial-temporal consistency condition random field loss layer; 4) training the convolution neural network; and 5) recovering a depth maps of a RGB image sequence of unknown depth through forward propagation with the trained convolution neural network; which avoids that clue-based depth recovery method is greatly depended on scenario assumptions, and inter-frame discontinuity between depth maps generated by conventional neural networks.

The invention discloses a robot positioning error graded compensation method. The method includes: calculating terminal pose errors before compensation; building a kinematic model and a kinematic error model; calculating the optimal solution of each kinematic parameter error, the corrected kinematic model and the terminal pose under the corrected kinematic model; calculating input compensation quantity after positioning error compensation; calculating actual arrival pose to obtain residual errors after parameter calibration; building a residual error estimation model, and performing expected target point compensation; predicting positioning errors; acquiring predicted target point residual errors; calculating input compensation quantity; calculating terminal actual arrival pose to completethe compensation. By the method, industrial absolute positioning precision can be increased effectively, and the anisotropy of positioning errors in the space can be shown by using the residual errormodel based on space similarity.

The invention discloses a method for generating a spatial-temporal consistency depth map sequence based on a convolution neural network, which can be used in a film and television work 2D-to-3D technology. The method comprises the following steps: (1) collecting a training set, wherein each training sample in the training set is composed of a continuous RGB image sequence and a corresponding depth map sequence; (2) carrying out spatial-temporal consistency super pixel segmentation on each image sequence in the training set, and constructing a spatial similarity matrix and a temporal similarity matrix; (3) constructing a convolution neural network composed of a single super pixel depth regression network and a spatial-temporal consistency conditional random field loss layer; (4) training the convolution neural network; and (5) for an RGB image sequence of unknown depth, using the trained neural network to recover a corresponding depth map sequence through forward propagation. The problem that a depth recovery method based on clues relies too much on the scene hypothesis and the problem that the frames of a depth map generated by the existing depth recovery method based on a convolution neural network are discontinuous are avoided.

The invention discloses a hyperspectral remote sensing surface feature classification method based on superpixel-tensor sparse coding, and irons out the defects that the prior art cannot makes the most of the spatial information of a hyperspectral image to carry out classification and is low in classification speed. The method comprises the steps: (1) inputting a to-be-classified hyperspectral image; (2) building a hierarchical spatial similarity matrix; (3) obtaining a superpixel set; (4) building a mark sample dictionary; (5) solving a sparse coefficient matrix; (6) classifying superpixels; (7) outputting the classification results of the to-be-classified hyperspectral image. The method is advantageous in maintaining the spatial consistency of homogeneous regions of the hyperspectral image and being high in classification speed, and can be used for the rapid classification of the hyperspectral image.

The invention relates to a stereo matching method based on a self-adaptive weight and local entropy, belonging to the technical field of binocular stereo vision. The method includes redefining the calculation of the window weight according to the gray similarity and the spatial similarity under a self-adaptive weight theory framework; introducing a data smoothing item into the cost function, and determining the penalty coefficient of the data smoothing item by applying the image local entropy; and performing parallax refinement by applying left-right consistency check and weighted median filtering on the initial parallax map obtained by cost aggregation to obtain a high-precision parallax image. The accurate parallax map can be obtained, and the parallax accuracy is improved.

The invention discloses a pedestrian re-identification model optimization method based on a fusion loss function, and the method comprises the steps: (1) obtaining a pedestrian re-identification modelthrough the training of a neural network, and calculating the feature spatial similarity of each pedestrian image in the model through a cross entropy loss function; (2) setting a loss threshold value according to the spatial similarity, and constructing a triple object by using the threshold value; and (3) constructing triple loss function iterative optimization by using the triple object to obtain an optimized pedestrian re-identification model. According to the method, the cross entropy loss function and the triple loss function are combined, deep learning and feature recognition are applied to the pedestrian re-recognition problem, and the important significance is achieved for improving public security management.

Provided are a method and apparatus for estimating a motion vector using a plurality of motion vector predictors, an encoder, a decoder, and a decoding method. The method includes calculating spatial similarities between the current block and the plurality of neighboring partitions around the current block, selecting at least one of the neighboring partitions based on the calculated spatial similarities, and estimating a motion vector of the selected partition as the motion vector of the current block.

The present invention discloses an HEVC fast coding method. During the coding process, the temporal similarity between optimal coding units of the same coordinate position in forward and backward reference frames, the spatial similarity between adjacent optimal coding units in the left and right direction, and the spatial similarity between adjacent optimal coding units in the up and down direction, are fully utilized to figure out the depth predicted values of P-frame or B-frame optimal coding units. After that, based on the one-to-one correspondence relationship between the depth predicted values of the optimal coding units and depth ergodic intervals, the depth ergodic intervals of the optimal coding units are determined. The optimal coding units and all coding units in the optimal coding units are coded within the depth ergodic intervals. During the coding process, qualified P-frame or B-frame optimal coding units and qualified coding units are selected quickly in the prediction mode, so that the unnecessary depth traversal and the unnecessary prediction-mode traversal are avoided. Therefore, on the premise that the video quality is ensured, the calculation complexity of the video coding is lowered.

A system for clustering a plurality of documents having input and output space data is disclosed that uses both input and output space criteria. The system can aggregate documents into clusters based on input and / or output space similarity measures, and then refine the clusters based on further input and / or output space similarity measures. Aggregation of documents into clusters can include forming a hierarchical tree based on the input and / or output space similarity measures where the hierarchical tree has a root node, branching into intermediate nodes, and branching into leaf nodes covering individual documents, where the hierarchical tree includes a leaf node for each document of the plurality of documents. The system can include forming a forest of sub-trees of the hierarchical tree based on cluster criteria. Textual and numeric similarity measures can be used depending on the type and distribution of data in the input and output spaces.

The invention discloses a Chinese POI (Point Of Interest) matching method based on natural language understanding, which is used for solving the problem of low Chinese POI matching precision caused byfrequent incomplete Chinese addresses, inconsistent Chinese character expressions and one-word polysemy of multi-source Chinese POI data. According to the method, six similarity measurement characteristics of geometric positions, category structures, POI names and address attributes are integrated and input into the deep neural network to calculate the matching probability, so that the influenceof subjective factors caused by artificial weight allocation is avoided; a natural language understanding technology is introduced to fully consider the semantic relationship between address attributes and name attributes in the similarity calculation process of the six attribute characteristics to overcome the defects of a traditional algorithm; short text similarity calculation is improved by using a neural network of a twin attention mechanism, and POI data accurate matching is achieved in combination with grammar, semantics and spatial similarity calculation.

The invention discloses a cloud model based sensor fault identifying system and method and belongs to a field of agricultural IoT. According to the invention, fault identification of abnormal points is realized through space similarity of sensor node environment information and node information of the current moment is predicated according to time relevance of abnormal point history data. The node information is predicated by utilizing the space relevance among homogeneous sensor nodes and heterogeneous sensor nodes. An estimation value of the sensor node of the current moment is generated through combination of the predication information by utilizing a cloud model based data fusion method, so that fault diagnosis and data restoration of the sensor node are realized. The invention can realize abnormal point identification for sensor nodes in facility agriculture and can realize diagnosis and data restoration by utilizing the fault diagnosis method; data collection accuracy of the sensor nodes can be improved effectively and easiness and quickness are achieved. The invention can be applied to fault identification and diagnosis of abnormal data of the sensor nodes and realizes estimation of sensor abnormal data.

The invention provides a method for extracting the dimension characteristics of an underwater target based on space similarity. The method comprises the following steps of calculating the conventional beamforming space power spectrum of a distribution source based on ripple signals of the underwater target, utilizing point target space power spectra obtained by a space similarity theory and a classic theory to perform space similarity measurement, and finishing classification and recognition according to previously established space measurement threshold standards so as to extract the dimension characteristics of the underwater target. The method for extracting the dimension characteristics has the advantages that the influence of irradiation angles, base arrays and target distance parameters on the dimension characteristics is considered so as to conform with the visual recognition of the eyes to the dimensional target; three space similarity measurement methods are adopted in the process of extracting the dimension characteristics of the target, the defects of a single method are overcome, and the extracting and recognition reliability of the dimension characteristics of the underwater target is improved; and the conventional beamforming (CBF) method is used for calculating the space power spectrum of the distribution source target, and the space similarity calculation methods are relatively simpler and favorable to engineering realization.

A method of clustering a plurality of documents having input and output space data is disclosed that uses both input and output space criteria. The method can include aggregating documents into clusters based on input and / or output space similarity measures, and then refining the clusters based on further input and / or output space similarity measures. Aggregating the documents into clusters can include forming a hierarchical tree based on the input and / or output space similarity measures where the hierarchical tree has a root node, branching into intermediate nodes, and branching into leaf nodes covering individual documents, where the hierarchical tree includes a leaf node for each document of the plurality of documents. The method can then include forming a forest of sub-trees of the hierarchical tree based on cluster criteria. Textual and numeric similarity measures can be used depending on the type and distribution of data in the input and output spaces.

The invention discloses an oil and gas reservoir comprehensive assessment oriented spatial case-based reasoning method. The method comprises the following steps: (1) aiming to a position of a to-be-assessed area of the oil and gas reservoir, determining types of various drilled reservoirs, interpolating basic parameters of the drilling, and exporting the result; (2) partitioning the to-be-assessed area into grids and generating a grid pattern layer, wherein each of the grids is a case expression object; (3) extracting the basic parameter attribute, a spatial feature value and the reservoir type (a part of grids have) corresponding to each of the grids in the pattern layer; (4) respectively exporting all and provided grid information of the reservoir types in the grid pattern layer, and constituting a reservoir to-be-assessed case set and a case library; (5) setting weights, finding the case with the highest similarity from the case library according to the spatial similarity and / or attribute similarity of each case in the to-be-assessed case set, and attaching the reservoir type to the to-be-assessed case; (6) correlating all the obtained cases of the reservoir type to the grid pattern layer, coloring the grids according to the type, and at last obtaining a reservoir comprehensive assessment classification chart.

The invention provides an indoor locating method oriented to non-specific Wi-Fi equipment. The method comprises the steps that reference points are arranged in an indoor environment, and first Wi-Fi equipment is used for detecting received signal strength of all Wi-Fi access points at the reference points and multiple related points on the peripheries of the reference points to establish a Wi-Fi position fingerprint database; second Wi-Fi equipment is used at the position to be located to collect the received signal strength data from all the Wi-Fi access points, and data processing is carried out; the similarity of data space distribution of received signal strength nearby the position to be located and the data space distribution of received signal strength of each reference point is calculated, and the position coordinates of the position to be located are estimated. According to the indoor locating method oriented to the non-specific Wi-Fi equipment, the received signal strength data in eight directions around the reference points are stored when the Wi-Fi position fingerprint database is established, and the precision of locating is improved. When locating calculation is carried out, the spatial similarity degree of the position to be located and each reference point in the Wi-Fi position fingerprint database is taken into consideration, the locating precision problem of Wi-Fi equipment of different types is solved effectively, and the robustness of a system is improved.

The invention discloses a three-dimensional image visual saliency map extraction method. The method includes the steps of firstly, establishing a relation model of the visual comfort degree characteristics and the visual comfort degree prediction values of a three-dimensional image through training; secondly, predicting the visual comfort degree of the three-dimensional image through the relation model so as to obtain a visual comfort degree saliency map; thirdly, extracting a two-dimensional saliency map of the three-dimensional image through a visual saliency model based on the graph theory; fourthly, obtaining a deep saliency map of the three-dimensional image according to the space similarity and the parallax similarity between regions; fifthly, fusing the two-dimensional saliency map, the deep saliency map and the visual comfort degree saliency map to obtain the final three-dimensional visual saliency map. The method has the advantage that the obtained three-dimensional visual saliency map can well conform to the characteristics of the saliency semanteme.

The invention belongs to the technical field of monitoring systems, and particularly relates to a distributed network visual monitoring system. A frame difference image processing module is used for performing image foreground segmentation on an image picked up by a monitoring terminal, forming a foreground image, performing inter-frame RGB spatial similarity processing and inter-frame foreground pixel area difference processing on the foreground image, forming frame difference image processing data by statistical data of inter-frame RGB spatial similarity and statistical data of foreground pixel area difference data, and analyzing and processing the frame difference image processing data and a frame difference image threshold to form a motion target detection result. Through adoption of the distributed network visual monitoring system, the key problem of how to enhance the real-time performance of an embedded visual system while ensuring the recognition and tracking efficiency in the prior art is solved. The distributed network visual monitoring system has the beneficial technical effects of enhancing the capability of video monitoring on the aspects of finding abnormal behaviors in real time and transmitting alarm information immediately, greatly lowering the labor cost, generating good economical and social benefits, and increasing the efficiency of security and protection monitoring.

The invention discloses a large-scale forest scene quick generation method based on space similarity. The method comprises the steps of 1) setting initialization parameters of a large-scale forest simulation scene, and acquiring scene distribution information of visible data and forest space of the whole scene; 2) partitioning the space data of the large-scale forest scene based on a quadtree, and dividing the scene into blocks with the same size; 3) calculating the space similarity of the forest scene; 4) judging the similarity between the scene blocks to quickly obtain biological amounts of plants; and 5) if the similarity between the scene blocks does not reach a certain proportion, calculating a growth model of the scene blocks, determining basic plants according to initial parameter data transmitted by a user, calculating an influence range and the biological amounts of the basic plants, then introducing a three-dimensional tree model into simulation for realizing the forest scene, and displaying a visible result to the user. The method is high in speed and high in precision.

The invention discloses a vector element change comparison method based on similarity degree. The method comprises the following steps: first, data being normalized so as to have the same spatial reference; second, establishing a knowledge rule base and determining knowledge rules, which mainly includes data analysis and change definition; third, data alignment and updating. The invention has thebeneficial effects that the accuracy and efficiency of the target identification are improved by combining the buffer area, the spatial index, the key attribute and the spatial similarity, and the reliability of the change detection result is ensured; at the same time, the invention proposes a whole set of element comparison and linkage update flow based on similarity, which can directly guide vector element comparison and update production activities.

The invention relates to a graph cut stereo matching method based on adaptive weight, and belongs to the neighborhood of binocular stereoscopic vision technology. The method comprises firstly redefining the data items of an energy function according to grayscale similarity and spatial similarity under the theoretical framework of the adaptive weight, and using the gradient information of an imageas the smooth item of the energy function; then solving a model by using a graph cut theory and an [alpha] extended algorithm; and finally, subjecting a disparity map to disparity refinement by usingleft and right consistency test and weighted median filtering to obtain a high-precision disparity map. Four international standard stereo image pairs provided by the Middlebury website are used for testing and experiments show that the method can obtain a more accurate disparity map.

The invention is applicable to the field of information technology. Provided is a parking lot data patching method and device, an apparatus and a storage medium. The method comprises: according to thespatial characteristics and spatial similarity measurement model of parking lot, calculating the spatial similarity between two parking lots; according to the spatial similarity, cluistering the parking lots into different parking lots, calculating the spatial similarity probability corresponding to the parking lot cluster is calculated, and calculating the data similarity between two sample parking lots in the same parking lot cluster is calculated; according to the data similarity and the spatial similarity probability corresponding to the parking lot cluster, calculating the data similarity probability of the parking lot cluster under the condition of spatial similarity; when the probability of spatial similarity exceeds the probability threshold, repairing the data of the parking lotto be repaired in the parking lot cluster by the cyclic generation countermeasure network. The data of the parking lot to be repaired can be repaired accurately without relying on the prior knowledgeof the parking lot to be repaired, and the data collection cost of the parking lot can be effectively saved.

The invention discloses a malicious code classification method based on depth forest, belonging to the technical field of network security. The method comprises the following steps: performing data preprocessing, mapping malicious code binary file to image, zooming and standardizing the image; Extracting HOG eigenvector set of the image; The HOG eigenvector set is transformed into matrix, which isdivided into training set and test set. Constructing random forest; Constructing completely random forest; constructing Deep forest.wherein Deep forest is composed of multi-granularity scanning and cascade forests. Each layer is composed of equal number of random forests and completely random forests. Classifying The images with depth forests. The present application makes full use of the spatialsimilarity of malicious code images. The accuracy of classification and the accuracy of macro F1 measurement of the malicious code classification method based on the depth forest of the present application are high, and the problem of binary file reallocation is solved by scanning HOG feature matrices with windows of different sizes of the depth forest of the present application.

An improved hyperspectral image denoising method disclosed by the present invention comprises the following steps of (1) transforming a to-be-denoised hyperspectral image into a spatial spectrum jointed two dimensional matrix; (2) according to the spatial spectrum jointed two dimensional matrix, adopting an euclidean distance-based neighbor similarity calculation strategy to calculate the spatial similarity of the pixel points of the hyperspectral image and the adjacent pixel points; (3) combining the pixel point spatial local similarity and the low rank property between the spectrums, establishing a denoising model, and recovering the original noise-free data; (4) adopting the original noise-free data to recover a three dimensional noise-free hyperspectral image. The improved hyperspectral image denoising method of the present invention enables a denoising effect to be improved remarkably, and enables the hyperspectral images to be classified reliably aiming at the application scenes of different real-time response and high-precision demands.

The invention discloses a user activity anomaly detection and traffic prediction method based on big data analysis, which comprises the following steps: 1) obtaining regional number and time period information of an undetermined abnormal region; 2) using historical big data to evaluate an empirical cumulative probability curve of the undetermined abnormal region at time period activeness; then, using the empirical cumulative probability curve to determine a traffic activeness abnormal value of each moment in a prediction region; 3) finding eight regions adjacent to the space of the predictionregion with the prediction region as the center; then, using measurement sim (i, j) of spatial similarity of two adjacent regions i, j to find a spatial similar region k of the prediction region; andfilling a vacancy value yi, t of the time sequence of the prediction region; 4) performing traffic prediction on the time sequence of the prediction region obtained in the step 3) and finishing user activity anomaly detection and traffic prediction based on big data analysis. The method can realize accurate traffic prediction of the prediction region.

The invention relates to a method for obtaining a Euclidean space coordinate conversion parameter based on a minimum point set. The method comprises the following steps: 1, randomly extracting the minimum point set in point cloud; 2, calculating a coordinate conversion scaling proportionality coefficient lambda; 3, calculating a spatial similarity transformation model rotating matrix R; 4, calculating and drawing a translation vector t between a local coordinate system and world coordinate system; 5, counting correct points and gross errors; 6, executing judgment; and 7, outputting the Euclidean space coordinate conversion parameter. According to the method provided by the invention, the initial value of a spatial similarity transformation model is not needed, and the Euclidean space coordinate conversion parameter can be directly obtained according to the minimum point set, thereby avoiding the limitation that the traditional least square method requires the initial value. In combination with the RANSAC algorithm, the influence of the gross errors in control points to the calculation of the Euclidean space coordinate conversion parameter can be avoided, high-precision Euclidean space coordinate conversion parameter can be obtained, and the spatial similarity transformation between the coordinate systems can be calculated accurately.

Determining a location of a user on a social network platform is difficult due to incorrect information or lack of information associated with the user. A system and method are provided to compute contextual similarity. This includes, for example, computing content similarity between seed users and followers / friends, as well as computing an engagement score between seed users and followers / friends. The system also computes geo-social-spatial similarity. The similarity scores are used in any inference computation to infer the geo-locations of the followers of the seed users, and subject users who share common friends with the seed users. The user geo-location inference database is updated using the result. Other seed users are selected, and the process is repeated.