48 results about "Distance classifier" patented technology

Techniques and structures are disclosed in which one or more distortion categories are identified for an image or video, and a quality of the image or video is determined based on the one or more distortion categories. The image or video may be of a natural scene, and may be of unknown provenance. Identifying a distortion category and / or determining a quality may be performed without any corresponding reference (e.g., undistorted) image or video. Identifying a distortion category may be performed using a distortion classifier. Quality may be determined with respect to a plurality of human opinion scores that correspond to a particular distortion category to which an image or video of unknown provenance is identified as belonging. Various statistical methods may be used in performing said identifying and said determining, including use of generalized Gaussian distribution density models and natural scene statistics.

Disclosed is a hyperspectral data classification method which is combined spectrum and spatial information. The steps comprises (1) reading the hypersectral data, (2) confirming the minimum size of structural element, (3) calculating differentiation between picture elements in neighborhood of each structural element by extended mathematical morphology expansion and corrosion operation, (4) obtaining exponential value of morphology eccentricity by the extended expansion and the corrosion operation of step (3), (5), constantly repeating the above steps with the adding of the size of the structural element to achieve the maximum size of the structural element, (6), constantly updating the exponential value MEI of morphology eccentricity in iteration process via the obtained new value, and generating a final exponential value MEI of morphology eccentricity after the iteration process is finished, (7) realizing the extraction of the data characteristic by the image of the exponential value MEI of morphology eccentricity, namely generating ground object type information, and realizing sophisticated category of the ground object by a minimum-distance classifier. The method is an unsupervised classification method for hyperspectral ground object with strong stability, high reliability and high accuracy.

The invention discloses an illumination-classification-based adaptive image segmentation method which is used for accurately segmenting a target object under different illumination conditions. The illumination conditions are divided into two types, namely a frontlighting type and a backlighting type, by extracting color characteristics of an image to be processed in a red, green and blue (RGB) space and a hue, saturation and value (HSV) space and adopting a minimum euclidean distance classifier; a proper color characteristic quantity serving as a segmenting parameter is extracted from the image in the two illumination types and imported into a two-dimensional histogram; neighbor information of each pixel point is increased, so the interference resistance capacity is improved; and the acquired image is subjected to intelligent illumination judgment and precise segmentation. In the illumination-classification-based adaptive image segmentation method, a mode of judging the illumination condition first and then selecting a segmenting algorithm is adopted, so the algorithm has higher pertinence and the effectiveness of the algorithm is improved; meanwhile, illumination correction is not required, so the computing cost is reduced greatly; and a favorable condition is created for the subsequent image processing and analysis.

A handwriting authentication and verification method based on single character characterized by that, four traverse line element characteristic is extracted to better reflect the feature of the Chinese characters, then one of the following methods is used to reflect the optical diagnostic character of the difference between different writers, one is to extract the most distinctive feature through direct LDA (linear differentiation analysis) transformation, the other is to first extract the most distinctive feature through direct PCA (principal component analysis) transformation dimensionality reduction, then using LDA transformation to obtain the most distinctive diagnostic character. In accordance with the invention, an Euler's distance classifier is used for categorization authentication.íí

A high spectroscopic data supervision classification method based on information quantity dimensionality sequence includes the following steps: reading the high spectrum image data of a certain region; selecting a reference spectrum from the spectrum library or selecting the training sample from images to execute wave band average acquiring reference spectrum; calculating one by one the reference spectrum and all test spectral information quantity dimensionality sequence; executing vector angle matching one by one the test spectrum with all reference spectral information quantity dimensionality sequence, and using minimum distance classification machine to classify; result binarization matching, the matching result of each series of field culture is represented by the binary images, each image only includes a series of field culture. The high spectroscopic data classification method based on information quantity dimensionality sequence introduces the information quantity dimensionality into the spectrum domain analysis, synthetizes the advantages of the all band matching and partial quantization characteristic matching, can obtain higher classification effectiveness and classification accuracy, and has important value in the high spectroscopic data classification and object identification.

The invention provides an automatic decomposition method of an array type sEMG (surface EMG) signal. The automatic decomposition method comprises the following steps: firstly, pretreating the array type sEMG signal, and calculating an initial distribution sequence vector; determining the number of kinematic units, namely, the cluster number, by using time domain subtraction; according to the cluster number, clustering array type sEMG waveforms by using a minimum distance classifier; finally, recalculating a new kinematic unit distribution sequence vector, circulating the executive routine till decomposition is finished, classifying all the distribution sequences, and optimizing the result. The decomposition method is high in accuracy, quick in calculation speed and easy to operate.

The invention provides a radar target one-dimensional range profile local optimal sub-space recognition method which effectively improves the performance of recognizing a radar target. According to the method, firstly, a nearest intra-class distance scattering matrix and a nearest between-class distance scattering matrix are calculated through training data; then, a local optimal sub-space is set up according to an optimal ratio criteria, the characteristics of the target are extracted, and the characteristics are classified through a minimum distance classifier; finally, the classification where the input target belongs is determined. The method specifically includes the steps of determining a vector X<W>ij and a vector Xij through a radar target one-dimensional range profile training vector Xij; determining a vector d<W>ij and a vector dij; determining a matrix DW and a matrix DB; determining m vectors a1, a2...and am of the local optimal sub-space; determining the equation (please see the equation in the specification) of the local optimal sub-space according to lambdai and a vector ai, wherein i is 1 or 2 or 3...or m; determining a template library according to the projection of the training vectors in the sub-space A; determining the local optimal sub-image of an input target one-dimensional range profile Xt; determining the distance between the local optimal sub-image and a library template vector, and determining the classification where the input target range profile belongs through the minimum distance classifier.

The invention discloses an external insulation filth status identification method based on visible-light images. According to the method, images of an insulator are photographed, the images of the insulator are extracted by utilization of an image segmentation technology, related characteristic quantities are calculated and selected, and the insulator filth degree is determined by utilization of a distance classifier. The method has advantages comprising capability of detecting the insulator filth degree under a condition of no power cut, no need of installing any equipment in a pole or a tower, low operation cost and maintenance cost, simple operation, safe and reliable usage and high detection precision.

The invention discloses a human behavior recognition and classification method based on an accelerometer. The method includes the first step of collecting human behavior samples as a training set, the second step of searching for a projection matrix U which is optimal for the recognition and classification of the training set, the third step of carrying projection on no-labeled data, and the fourth step of classifying the projected data by using a minimum distance classifier to obtain a recognition result. According to the human behavior recognition method, a partial approximate linear hypothesis is carried out on adjacent blocks formed by labeled data so as to enable the distance between different types of samples on the blocks to be large enough, positional sequence information of the same type of samples is reserved as far as possible through class sigmoid function penalty factors, and finally a global objective function is established on the basis of the objective functions on all blocks. The human behavior recognition method can reserve the information of the distance between the samples in a higher dimensional space properly, and reduces dependence of recognition models on artificial tagging samples, and the recognition effect is superior to a representative human behavior recognition method based on linear discriminant analysis.

The invention discloses a scene recognition method based on a layered Gaussian hybrid model. The method comprises the following steps: extracting the layered Gaussian hybrid model corresponding to all images in a training image library and obtaining lie algebra descriptors of the layered Gaussian hybrid model; receiving images to be sorted and extracting the lie algebra descriptors of the layered Gaussian hybrid model of the images to be sorted; obtaining the distances from the lie algebra descriptors of the layered Gaussian hybrid model of the images to be sorted to the lie algebra descriptors of the layered Gaussian hybrid model of all images in the training image library; and sorting the images to be sorted according to the obtained distances by a minimum mid-value distance sorter and returning sorting results. According to the method provided by the invention, character representation of the layered Gaussian hybrid model of the images is optimized, so that the scene recognition performance is greatly improved.

The invention relates to the technical field of image information processing, in particular to a fine category classification method based on component polygons. The polygons based on component points are adopted for effectively correcting targets under different postures, fisher encoding features adopted in the classification process have robustness for changes in dimensions and directions of the targets, the adopted greedy algorithm can be used for finding out a set of component point combinations highest in discrimination power, and fine classifiers can be used for distinguishing categories very similar to one another. According to the fine category classification method, the component points of target images are connected, the multiple polygons based on the component points are constructed, and the number of errors caused by correction can be reduced effectively.

The invention provides an increment study classification method under limited storage resources, which belongs to the technical field of pattern recognition. The method adopts a minimum distance classifier and comprises the following concrete steps: firstly, pre-classifying new samples, adding the new samples which are correctly pre-classified to the corresponding subset of the classifier, adding the new samples which are wrongly pre-classified to the set of wrong samples, and carrying out K-mean clustering on the samples in the set of wrong samples; then respectively selecting representative samples for each subset in the classifier and the set of wrong samples, adding the subsets in the set of wrong samples to the classifier after selection, and updating the classifier; and finally, adopting the updated classifier to classify the new samples. In the invention, by selecting the representative samples, not only the studied knowledge is saved, but also the new knowledge is acquired, and higher sample recognizing accuracy is achieved on the basis of reducing the storage cost and calculating the cost.

The invention discloses a computing method and a computing method system for crop disease index. The computing method comprises the following steps: acquiring image of crop and pre-processing the image; dividing the pre-processed image into multiple sub-images and extracting color characters of the sub-images; using a minimum distance classifier for identifying the disease types and disease level of the sub-images according to the color characters of the sub-images; computing the crop disease index according to the disease types and disease level of the sub-images. The method solves that the disease index computing of the prior art depends on experience and knowledge of pathology of field crop protection persons, the statistical result is inaccurate and manual computation is required. The method is conducive to improve the prevention and treatment level for crop disease index and promote the implementation of precision agriculture, and novel method and technology are provided for the crop disease prevention and treatment field.

The invention provides an identification-oriented integral-local feature fusion-recognition system in the technical field of pattern recognition. In the system, a weighted adjacency graph construction module establishes a weighted adjacency graph, and acquires the similarity weight between any two vertices according to category information; a matrix construction module of a training sample receives the similarity weight, and establishes a similar matrix, a degree matrix, a laplacian matrix of the graph, an intra-class scatter matrix and an inter-class scatter matrix for the training sample according to the nearest neighbor principle; a selection module for an optimal regulatory factor selects a regulatory factor value allowing the recognition rate of the training sample to reach maximum as an optimal value of regulatory factors; a projection-matrix acquisition module selects eigenvectors corresponding to top minimum feature values in all feature values as base vectors, so as to form a projection matrix; and a data classification module adopts a minimum distance classifier to recognize the category of test data. The system utilizes the category information of data to depict data relation more accurately, thereby gaining higher recognition performance.

The invention provides a multi-classifier integration-based image character recognition method. The method comprises the following steps of: converting a colored to-be-identified image into a grayscale image; carrying out binary processing on the grayscale image and segmenting an image region with character information; segmenting each Chinese character from a whole character image; extracting grid features and direction features of each Chinese character; selecting stroke density total length features to carry out first-layer rough classification by adoption of a minimum distance classifier; and respectively selecting peripheral features, the grid features and the direction features to complete second-layer classification matching by adoption of a nearest-neighbor classifier. The method has the advantages that the character recognition has relatively strong anti-jamming capability and relatively strong character local structure description capability, and is less influenced by stroke widths; by adoption of a classifier integration technology of complementing and combining the minimum distance classifier and the nearest-neighbor classifier, a system is more reliable; and the characters can be intelligently recognized, so that the adaptability of the system is improved and the recognition rate is high.

The invention provides a plastic part defect detection method based on multiple threshold intervals. The method comprises the following steps: S1, inputting a grayed image of a surface picture of a detected part; S2, obtaining a defect eigenvalue vector of the grayed image of the detected part and a defect threshold interval in multiple threshold intervals of the defect; and S3, calculating a defect type corresponding to the defect eigenvalue vector by using a least distance classifier according to the defect threshold interval. Through the plastic part defect detection method, multiple threshold intervals are set, and the least distance classifier is used; the automation detection of the surface defects of the plastic parts is achieved; the detection efficiency of the surface defects of the plastic parts is improved; the manpower, the material resources and the financial resources are reduced.

The method of classifying raw EEG signals uses a classification method based on nuclear features extracted as dominant singular values from an EEG signal segment using singular value decomposition (SVD) and a class means-based minimum distance classifier (CMMDC) to classify a patient's EEG signals. From a mean EEG signal, a set of zero-centered EEG signals are calculated, and from the zero-centered EEG signals and a standard deviation of the EEG signals, a unit variance is calculated for each component. Using the standardized component signals a nuclear matrix is calculated, to which singular value decomposition is applied to generate a set of singular values. The CMMDC is applied to class means associated with first and second classes and a nuclear feature vector to classify the patient's EEG signals as belonging in either the first or second class.

A hand-written number identification method based on nuclear function, including character extraction and identification of the hand-written number, at the phase of the extraction, firstly map the low-dimension character to the high-dimension space via nonlinear mapping function to make the problem of the character space become linear dividable or near linear dividable and further uses the nuclear function to extract the characters; at the character identification phase, use simple k-NN distance classifier and shortest distance classifier to identify.

The invention relates to a bearing fault diagnosis method which is characterized by including the following steps: (1) obtaining multiple groups of vibration acceleration signal data of the spindle bearing of a wind generating set in different states, randomly extracting a plurality of groups of vibration acceleration signal data as standard sample data, and taking the other groups of vibration acceleration signal data as to-be-detected sample data; (2) adaptively decomposing the standard sample data and the to-be-detected sample data to obtain a series of intrinsic rotational components; (3)making a time-domain analysis of the instantaneous amplitude and instantaneous phase information of the first intrinsic rotational components in the standard sample data and the to-be-detected sampledata, and extracting bearing fault feature vectors; (4) inputting the extracted fault feature vector of the standard sample data to a neighbor distance classifier for training to get a trained fault diagnosis model; and (5) inputting the extracted bearing fault feature vector of the to-be-detected sample data to the trained fault diagnosis model for fault identification to obtain the fault state of the bearing. The bearing fault diagnosis method of the invention can be widely applied to bearing fault diagnosis.

A discrimination-oriented correlated characters fusion recognition system of an image belongs to the technical field of telecommunication, comprising the following steps: first, determining a projection matrix via an iterative process according to a training image; second, respectively projecting the training image and test image to a characteristic space by utilizing the projection matrix, extracting the training characteristic and test characteristic and then carrying out parallel fusion respectively on the characteristics to form a new characteristic; at last, extracting characteristics with more discriminability by utilizing linear discriminant analysis method and adopting minimum distance classifier, thus identifying the category to which the test image belongs. The system can more naturally depict the relation between row vectors and the relation between column vectors of an image matrix; the extracted characteristics can more intensively reflect the image information; further discrimination-oriented processing can extract characteristics with more discriminability; and the system can improve not only processing speed but also recognition performance by being applied to the image recognition.

The invention discloses a big data image classification method. The big data image classification method comprises a first step of enabling image samples to be collected to serve as a training set, a second step of searching a projection matrix optimal in big data image classification, a third step of performing projection on data without marks and a fourth step of adopting a minimum distance classifier to classify the samples after projection. According to the method, local geometric information of sample distribution can be effectively utilized, classified discrimination information is extracted, dependence of big data image classification on manually marked samples is reduced, storage cost in the training process is effectively reduced, and the big data image classification method has higher classification accuracy than a representative image classification method based on linear discrimination analysis.

The invention discloses a feature dimension-reduction optimization method for Chinese character recognition. The feature dimension-reduction optimization method comprises a first step of conducting preprocessing and feature extraction on a Chinese character sample and conducting LDA dimension reduction changing on the extracted Chinese character features, a second step of conducting classification and recognition by means of a minimum Euclidean distance classifier, a third step of regarding the sample with wrong classification and recognition as a newly-added sample, adding the newly-added sample to an original sample set, and conducting the dimension reduction changing again by means of a learning method of ILDA incremental linear judgment, a fourth step of conducting the classification and recognition again by means of the minimum Euclidean distance classifier, and a fifth step of repeating the third step and the fourth step, outputting the LDA optimizing parameter after repeated iterative calculation, and enabling the LDA optimized parameter to be used for the classification and recognition of Chinese characters. The feature dimension-reduction optimization method for the Chinese character recognition overcomes the defect that an existing LDA changing method can not effectively optimize the LDA transformation matrix parameters by means of recognition classification and information, and has the advantages that the performance of the LDA feature dimension reduction changing and the accuracy rate of the character recognition can be greatly improved.

The invention relates to a training method and system of a sleep-state classifier, wherein the method comprises the following steps: constructing feature vectors of sample data of a plurality of sleep-state types and a cluster center formed by aggregating the feature vectors, and establishing an objective function according to the feature vectors and the cluster center of the feather vectors; using the objective function to characterize the distance between the minimized sample data of same types and dictionary atoms, and the distance among maximized atoms of different types; respectively selecting a plurality of the feature vectors from the sample data of a plurality of the sleep-state types to be taken as initial values of the atoms, distributing various sample data to the atoms and solving the objective function, thus obtaining a classification dictionary; using the classification dictionary to classify the sample data, comparing the types and the distances of the atoms which are closest to the sample data, if the distance is less than a preset threshold value, judging that the types of the sample data are consistent with the types of the atoms; training the sleep-state classifier according to the classified sample data. By adopting the training method and system of the sleep-state classifier, a more accurate sleep-state classifier can be trained.

The invention discloses a human ear recognition method based on fusion of a gradient direction histogram and a local binary pattern, which solves the problem of low human ear recognition rate in a human ear image. According to the method, firstly, the gradient direction histogram of the image is extracted, dimensionality reduction is carried out through a principal component analysis method, thenthe texture features of the image are extracted through the local binary pattern, then the two features are fused, and finally classification is carried out through a minimum distance classifier. According to the invention, through the multi-feature fusion, the recognition rate of human ear recognition is improved, and the method has good implementability and effectiveness.

The invention provides a base station identification method in a single-frequency network positioning system. The method, based on the positioning system, converts the problem of base station identification to the problem of data classification, and the base station identification is realized through a minimum-distance classifier; a state prediction model is utilized to estimate motion parameters of a mobile station, and thus hardware requirement of the system is lowered; and to get a better base station identification effect in a non-line-of-sight environment, an interaction multi-model method is utilized to reduce non-line-of-sight effect, and thus more accurate mobile station object information is obtained. Two different functions, position estimation and base station identification, of the positioning system are closely associated, so that the method is economic and effective, and is suitable for all digital television standards.

The invention provides a texture classification method on the basis of fractional Fourier transform (FrFT). The texture classification method comprises the following steps: (1) acquiring an image and calculating one-dimensional discrete FrFT of each pixel of the image in four directions; (2) carrying out descending sorting on each obtained one-dimensional discrete FrFT according to amplitudes; (3) calculating a fractional Fourier frequency histogram; and (4) carrying out classification on the texture image by utilizing a chi2- statistical distance classifier. The invention provides the texture classification method which is on the basis of FrFT and comprehensively utilizes the fractional Fourier frequency histogram and the chi2- statistical distance classifier. The texture classification method provided by the invention has the main effect of avoiding the problem of cross terms of Wigner distribution.

The invention discloses an iris recognition method based on wavelet packet decomposition, and the method comprises the steps: collecting a human eye image, carrying out the iris image preprocessing, obtaining a processed iris image, modulating the diagonal high-frequency information of the wavelet packet decomposition of the iris image into an iris feature code through a threshold value, extracting the iris features, and finally, adopting a Hamming distance classifier to perform classification and matching calculation on iris features, and judging whether the two feature codes are from the same iris or not by calculating the similarity between the current iris feature code and iris codes in an iris template library, so that iris images are classified. The iris features extracted through the iris feature recognition method have good classification performance, so that the iris recognition precision is improved.

The invention belongs to the field of computer vision, particularly relates to a compressed domain-oriented video content comparison system, an optimization method and a comparison method, and aims tosolve the problem of low efficiency of completing video content comparison by using full decoding information. The comparison system comprises: a feature learning module which is used for respectively obtaining feature maps of multiple modes based on multiple pieces of compressed domain information of an input video; a multi-modal compressed domain information fusion module which is used for carrying out information fusion on the multi-modal feature maps output by the feature learning module to obtain a fusion feature vector of the input video; a second module which is configured to obtain the L1 distance of the fusion feature vector of the two input videos; and a classifier which is a binary classification network and is configured to perform binary classification of the comparison result based on the L1 distance output by the second module. According to the invention, high-level semantic information of the video content can be effectively extracted, and high comparison speed and high performance of the video content are ensured.

The invention relates to a method for identifying vehicle types of vehicles of the same brand based on spatial position information. The method includes: collecting images of the vehicles; performing histogram equalization on the images of the vehicles; setting the size of the license plate, and adjusting the histogram The image after image equalization is stretched and rotated; the feature area of ​​N vehicle models to be tested is extracted according to the position of the corrected image relative to the license plate; the feature vector of each feature area is calculated; the minimum distance classifier is used to classify each feature vector Classify and compare with the reference feature vector of the brand vehicle in the database to identify the specific model. The method of the invention extracts the feature area of ​​the license plate and then identifies the relevant parameters of the feature area, thereby improving the recognition accuracy of the vehicle type.