175 results about "Virtual sample" patented technology

The invention discloses a method for recognizing faces shaded by eyeglasses, which comprises the following steps of: inputting an image of an eyeglass-free face and a general face model to acquire corresponding points on a face image and the face model, and performing coordinate transformation on all peaks to adjust the peaks under the same coordinate system; processing the model to acquire a facade head model of which the posture is corrected; adding an eyeglass model on the head model and pasting textures; performing fuzzy processing and light reflecting processing on the lenses of the eyeglasses; and generating virtual samples in which the eyeglasses are worn under different conditions by using a projection model. The invention also provides a corresponding face recognition system. The method and the system have the advantages of wide simulation range, high generality, greatly improved recognition effect, real-time property, high practicability, high recognition rate and the like, can comprise more changes of the eyeglasses, are quick and are easy to operate.

The invention relates to a partial-compensation aspherical reflector surface shape detection method, which comprises the following steps of demarcating partial-compensation elements to obtain demarcated phase distribution Wb; using the demarcated phase distribution Wb, sphere theory wavefront Wq and aspherical theory wavefront Wf to obtain phase distribution Wy of digital virtual sample plates; measuring an aspherical reflector surface to be measured to obtain phase distribution Wa; and reducing the phase distribution Wy from the measured phase distribution Wa to achieve surface shape error distribution Wm of the aspherical surface. The partial-compensation aspherical reflector surface shape detection method can accurately measure optical aspherical surface shapes by using a single partial-compensation element, thereby reducing design difficulty of a compensator, simplifying structure of the compensator and being clear in physical definition, simple in data processing and arithmetical operation, simple and practical in experimental operation, low in detection cost and short in test time.

A work state virtual simulation system for an electric haulage shearer based on different geological conditions belongs to automatic control and system simulation of shearers. The work state virtual simulation system comprises devices and a method, wherein the devices consist of a shearer work condition remote monitoring platform, a database, a remote controller, a visual basic system, an industrial Ethernet, a local controller and a machine-mounted detection control system. The method comprises the following steps of: acquiring shearer work condition parameters for the machine-mounted detection control system; carrying out data analysis on the shearer work condition parameters; finally, transmitting the shearer work condition parameters to the shearer monitoring platform for detecting and achieving through the local controller and the industrial Ethernet; on the basis of accessing an achieving database in real time, drawing fully mechanized coal faces under different geological conditions by applying a virtual real technology; and driving a shearer virtual sample machine to realize real reappearance of the work state of the shearer. A shearer virtual sample machine model is driven by using real-time work condition parameters provided by the machine-mounted detection control system, so that a shearer virtual simulation system has the field feeling entering the fully mechanized coal faces and favorable interaction.

For use in acoustic micro imaging, a method (and implementing apparatus) is disclosed for creating a 4D virtual sample data memory. The method comprises employing a pulsed ultrasonic microscope probe to interrogate a sample at three-dimensionally varied locations in the sample. Data developed by the microscope probe includes for each location interrogated a digitized A-scan for that location. The developed data is stored in a data memory.

A graphics system has a mode of operation in which primitive coverage information is generated for real sample locations and virtual sample locations for use in anti-aliasing pixels. An individual pixel has a single real sample with color information and at least one virtual sample. In one implementation each virtual sample within a pixel is a pointer that identifies whether the virtual sample belongs to the single real sample within the pixel or to a proximate neighboring pixel. The virtual sample information permits a blending weight to be determined for blending color values of a partially covered pixel with color values of neighboring pixels.

A computer-implemented graphics system that includes a rasterizer and a shader has a mode of operation in which primitive coverage information is generated for real sample locations and virtual sample locations for use in anti-aliasing. An individual pixel includes a single real sample location and at least one virtual sample location. In some instances, a primitive may cover only virtual sample locations and does not cover a real sample location. These instances can be identified in the coverage information sent from the rasterizer to the shader, so that the shader can determine whether or not it can write color information, depth information and / or stencil information for the real sample location to a framebuffer.

Provided is a multiple-sparse-representation face recognition method for solving the small sample size problem. In the method, two modes are adopted to solve the small sample size problem during face recognition, one mode is that given original training samples produce 'virtual samples' so as to increase the number of the training samples, and the other mode is that three nonlinear feature extraction methods, namely a kernel principle component analysis method, a kernel discriminant analysis method and a kernel locality preserving projection algorithm method are adopted to extract features of the samples on the basis that the virtual samples are produced. Therefore, three feature modes are obtained, sparse-representation models are established for each feature mode. Three sparse-representation models are established for each sample, and finally classification is performed according to representation results. By means of the multiple-sparse-representation face recognition method, virtual faces are produced through mirror symmetry, and then norm L1 based multiple-sparse-representation models are established and classified. Compared with other classification methods, the multiple-sparse-representation face recognition method is good in robustness and classification effect and is especially suitable for a lot of classification occasions with high data dimensionality and few training samples.

An expression recognition method based on AVR and enhanced LBP in the technical field of model recognition comprises the following steps: collecting an original image; expanding virtual samples; carrying out wavelet decomposition on human image; extracting local binary pattern characteristic LBP; calculating an enhanced variance ratio AVR characteristic value and adding with a penalty factor, then extracting a plurality of groups of characteristic values of different dimensionality discriminated by AVR value, carrying out support vector machine classification accuracy test, and taking the characteristic dimensionality with the highest accuracy and corresponding characteristic value as the LBP characteristic. The method of the invention integrates image acquisition, human face test and human eye test, enhances the LBP characteristic by wavelet decomposition, and effectively improves the accuracy by adopting AVR method to extract effective characteristic.

One embodiment of the present invention sets forth a technique for converting alpha values into pixel coverage masks. Geometric coverage is sampled at a number of “real” sample positions within each pixel. Color and depth values are computed for each of these real samples. Fragment alpha values are used to determine an alpha coverage mask for the real samples and additional “virtual” samples, in which the number of bits set in the mask bits is proportional to the alpha value. An alpha-to-coverage mode uses the virtual samples to increase the number of transparency levels for each pixel compared with using only real samples. The alpha-to-coverage mode may be used in conjunction with virtual coverage anti-aliasing to provide higher-quality transparency for rendering anti-aliased images.

The invention provides a virtual sample deep learning-based robot target identification and pose reconstruction method. The method comprises the steps of extracting a region of an operation target in a camera image by adopting a CNN region detector, and preliminarily determining relative positions of the operation target and a robot end camera; estimating an observation angle deviation of a current view angle and an accurate pose solving optimal view angle of the computer end camera by adopting a CNN pose classifier; controlling a robot motion by adopting a multi-observation view angle correction method to enable the end camera to be transferred to the accurate pose solving optimal view angle; and by adopting virtual-real matching of contour features and pose inverse solving at the optimal view angle, realizing accurate calculation of a target pose. According to the method, the problem in massive sample demands of a deep convolutional neural network is solved, and the problems of feature shielding and matching difficulty caused by an excessively large contour matching view angle deviation are solved; and the activeness of robot vision perception and the algorithm flexibility of target pose reconstruction are improved.

One embodiment of the present invention sets forth a technique for converting alpha values into pixel coverage masks. Geometric coverage is sampled at a number of “real” sample positions within each pixel. Color and depth values are computed for each of these real samples. Fragment alpha values are used to determine an alpha coverage mask for the real samples and additional “virtual” samples, in which the number of bits set in the mask bits is proportional to the alpha value. An alpha-to-coverage mode uses the virtual samples to increase the number of transparency levels for each pixel compared with using only real samples. The alpha-to-coverage mode may be used in conjunction with virtual coverage anti-aliasing to provide higher-quality transparency for rendering anti-aliased images.

An image evaluation device includes a storage unit that stores sample image data that represent a virtual sample image simulating a sample image included in a sample printout that is recognized as a non-defective printout; a reading unit that reads an inspection object image included in an inspection object printout obtained by printing the sample image on a recording medium by a printing device using image data representing the sample image; an extraction unit that extracts a line defect including a linear pattern formed in a specific direction from the inspection object image represented by inspection object image data, based on a difference value between the sample image data and the inspection object image data; and an evaluation unit that evaluates a visibility of the line defect extracted by the extraction unit.

The invention relates to a power transmission line defect detection method and system and electronic equipment. The method comprises the steps of constructing a power transmission line defect detection model based on virtual and real sample integration and transfer learning according to image sample data of power transmission line elements, The method specifically comprises the following steps: step a, constructing a virtual and real integrated virtual sample generation and labeling model, integrating rich ground object information in virtual data and real image data, and fusing the virtual data and the real data; B, constructing a deep learning transfer learning model, and completing model optimization based on transfer learning; And step c, training a deep learning model based on a target detection algorithm, and carrying out abnormity diagnosis on the power transmission line element on the basis of target detection of deep learning. Through deep learning distributed training of massimage samples, the power transmission line defect detection model based on multi-image fusion of visible light, infrared light, ultraviolet light and the like is established, the defect recognition accuracy can be improved, and the inspection working efficiency and quality are improved.

The invention discloses an image processing method for expanding a data set under a small sample. In order to solve the problem of poor identification accuracy of a small sample image, the invention adopts a sample expansion technology to overcome the problem caused by insufficient samples, and provides a more general image processing method. The method comprises the steps that the original training samples are translated, rotated, mirrored, scaled and transmitted, the contrast and the brightness are transformed, the noise is added, the virtual training samples are generated, the number of training samples is increased by generating virtual samples, and then the virtual samples are fused with the original training samples. Through a large number of experiments, the method of the inventionhas excellent recognition effect on a small sample training set, and has better recognition performance. If the sample is insufficient, then the image features in the training phase are not enough toeffectively represent the image features changes, so that the difficulty of image recognition is increased, and even the phenomenon that the image can not be recognized appears.

For use in acoustic micro imaging, a method (and implementing apparatus) is disclosed for creating a 4D virtual sample data memory. The method comprises employing a pulsed ultrasonic microscope probe to interrogate a sample at three-dimensionally varied locations in the sample. Data developed by the microscope probe includes for each location interrogated a digitized A-scan for that location. The developed data is stored in a data memory.

The invention discloses a face image virtual sample generating method. The facial image virtual sample generating method comprises the steps of conducting control point calibration on a source posture face image training sample and a target posture face image training sample, then using a source posture face image training sample set and a target posture face image training sample set as input and output of an RBF neural network, and obtaining an RBF neural network fitting model through training; generating a source coordinate matrix according to source posture face images, inputting the source coordinate matrix into the RBF neural network fitting model to obtain a coordinate transformation matrix, conducting textural feature mapping according to the source coordinate matrix and the coordinate transformation matrix, then conducting interpolation on textural feature deficiency points to obtain a target posture face image virtual sample and finally conducting normalization operation and saving on the target posture face image virtual sample. The face image virtual sample generating method adopts the RBF neural network fitting model and enables the generated face image virtual sample to be approximate to a real sample, and further face recognition rate is further improved.

Each of an image coding apparatus and an image decoding apparatus uses a motion compensated prediction using virtual samples so as to detect a motion vector for each of regions of each frame of an input signal. Accuracy of virtual samples is locally determined while the accuracy of virtual samples is associated with the size of each region which is a motion vector detection unit in which a motion vector is detected. Virtual samples having half-pixel accuracy are used for motion vector detection unit regions having a smaller size 8×8 MC, such as blocks of 8×4 size, blocks of 4×8 size, and blocks of 4×4 size, and virtual samples having ¼-pixel accuracy are used for motion vector detection unit regions that are equal to or larger than 8×8 MC in size.

A graphics system has a mode of operation in which real samples and virtual samples are generated for anti-aliasing pixels. Each virtual sample identifies a set of real samples associated with a common primitive that covers a virtual sample location within a pixel. The virtual samples provide additional coverage information that may be used to adjust the weights of real samples.

The invention discloses a mill load parameter soft measuring method based on a virtual sample. The mill load parameter soft measuring method comprises the steps of acquiring a multi-dimension time domain sub-signal of mill cylinder vibration and vibration sound sample signal by means of ensemble empirical mode decomposition (EEMD) technology, and performing further processing for obtaining high-dimension spectral data with different time dimensions; then constructing a feasibility-based planning (FBP) model based on the high-dimension spectral data according to an improved selective integrated kernel partial least squares (IGASEN-KPLS) method, and generating a new virtual sample based on priori knowledge and an FBP model; then obtaining a mixed sampling model after mixing the new virtual sample with a true training sample, performing adaptive selection of a multiple-dimension spectral characteristic by means of a mutual information (MI) based characteristic selecting method, constructing a soft measuring model by means of the selected spectral characteristics and performing soft measurement.

A computer-implemented graphics system has a mode of operation in which primitive coverage information is generated for real sample locations and virtual sample locations for use in anti-aliasing. An individual pixel includes a single real sample location and at least one virtual sample location. A block of real sample locations can be selected to delineate and encompass a region containing a number of virtual sample locations. Pixel attribute values (e.g., z-depth or stencil values) associated with the block of selected real sample locations can be used to associate each virtual sample location within the region with one of the selected real sample locations. The virtual sample location assumes the pixel attribute value of the real sample location with which it is associated.

The invention provides a pedestrian re-identification method and device based on a semi-supervised training mode and a storage medium. The method comprises the following steps: constructing a trainingsample set by using the collected target domain sample set and open source data set based on a semi-supervised training mode; and training a deep residual network by using the training sample set toobtain a trained deep residual network model, identifying the acquired pedestrian images by using the trained deep residual network model to obtain feature values of the pedestrian images, and determining whether the pedestrian images belong to the same person according to cosine distances among the feature values. According to the invention, the virtual sample is generated; a smoothing function is constructed when a virtual sample is generated; and meanwhile, a DBSCAN clustering algorithm is used for adding pseudo tags to the virtual samples, local features and global features are used in thedeep neural network, and a joint loss function of different weight combinations is adopted, so that the trained deep neural network is accurate and more reliable in recognition.

A multi-grade chemical process soft sensor modeling method automatically generating samples comprises the following steps: (1) dividing the data collected from the multi-grade chemical process into atraining set and a test set as raw data; (2) establishing a generative antagonistic network AGAN based on gradient penalty and Wasserstein distance, and inputting the divided training set into the generative antagonistic network to train the network; (3) using the trained AGAN to generate virtual samples, and forming a new training set together with the original training set; (4) using the new training set as the driving data, training the soft sensor modeling, adjusting the parameters of the soft sensor model to adapt to the new training set, and using the trained soft sensor model to predictthe key quality variables of the multi-brand chemical process. The invention utilizes automatic generation of antagonistic network to generate data to make up the deficiency of data quantity, and improves the prediction accuracy of the soft sensor model based on data driving.

A computer-implemented graphics system has a mode of operation in which primitive coverage information is generated by a rasterizer for real sample locations and virtual sample locations for use in anti-aliasing. An individual pixel includes a single real sample location and at least one virtual sample location. If the coverage information cannot be changed by a pixel shader, then the rasterizer can write the coverage information to a framebuffer. If, however, the coverage information can be changed by the shader, then the rasterizer sends the coverage information to the shader.

The invention discloses an SAR image data enhancement method and device and a storage medium, and the method comprises the following steps: carrying out the electromagnetic simulation of an SAR targetimage, and obtaining an SAR electromagnetic simulation image; processing the SAR electromagnetic simulation image and the SAR target image through a generative adversarial network to obtain an SAR target image virtual sample set. According to the method, the SAR target image is subjected to electromagnetic simulation to obtain the multi-azimuth SAR electromagnetic simulation image, so that the defect that the traditional SAR target image sample acquisition is insufficient is overcome, and sufficient data input is provided for subsequently solving the problem of training data scarcity in the deep learning training process; the mapping relation between the SAR electromagnetic simulation image and the SAR target image is learned through the generative adversarial network, the data richness of the SAR electromagnetic simulation image is improved, and therefore the SAR target image with the missing angle is expanded, and powerful support is provided for subsequent SAR target image recognition and detection and other work.

An image evaluation device includes a storage unit that stores sample image data that represent a virtual sample image simulating a sample image included in a sample printout that is recognized as a non-defective printout; a reading unit that reads an inspection object image included in an inspection object printout obtained by printing the sample image on a recording medium by a printing device using image data representing the sample image; an extraction unit that extracts a line defect including a linear pattern formed in a specific direction from the inspection object image represented by inspection object image data, based on a difference value between the sample image data and the inspection object image data; and an evaluation unit that evaluates a visibility of the line defect extracted by the extraction unit.

Each of an image coding apparatus and an image decoding apparatus uses a motion compensated prediction using virtual samples so as to detect a motion vector for each of regions of each frame of an input signal. Accuracy of virtual samples is locally determined while the accuracy of virtual samples is associated with the size of each region which is a motion vector detection unit in which a motion vector is detected. Virtual samples having half-pixel accuracy are used for motion vector detection unit regions having a smaller size 8×8 MC, such as blocks of 8×4 size, blocks of 4×8 size, and blocks of 4×4 size, and virtual samples having ¼-pixel accuracy are used for motion vector detection unit regions that are equal to or larger than 8×8 MC in size.

The present invention discloses a virtual sample generation method comprising the following steps: firstly, acquiring a limited quantity of high dimensional real samples by adopting means of signal acquisition and corresponding devices, then constructing a feasibility based planning (FBP) model by adopting a partial least square (PLS) algorithm, a genetic (GA) algorithm and a back propagation neural network (BPNN) algorithm; secondly, generating an input of a virtual sample on the basis of future knowledge of the known real sample; thirdly, inputting PLS extracted potential features of the virtual sample into FBP, and acquiring an output of the virtual sample on the basis of the future knowledge; and finally, combining input vectors and output vectors of the virtual sample according with a preset rule to acquire a complete virtual sample. Therefore, the virtual sample that can be used for predicting high dimensional data is generated relatively accurately.

The invention discloses a virtual sample based kernel discrimination method for face recognition, which is a virtual sample based quick kernel method. The method comprises the following steps of: constructing a virtual sample set for a training sample set at one time before constructing a kernel matrix for the training sample set; and training / testing through a kernel matrix theory based on the virtual sample set. Since the virtual sample set is an aggregate of a characteristic sample set (MES) and a public vector sample set (MCS) of the training sample set, the virtual sample set has extremely high description capacity for both a known training sample set and an unknown test sample set. Experimental verification of the method on an FERET database shows that the method is quick and effective; the computation speed of the kernel method is greatly increased by using the method; and meanwhile, compared with the conventional kernel method, the recognition rate is also increased.

The invention relates to the technical field of materials, in particular to a GISSMO material failure model parameter optimization method which comprises the following steps: S1, determining a real stress-strain curve of simulation input; s2, comparing the simulation result of the uniaxial stretching virtual sample with the test result to determine the initial range of the WF; s3, under the condition that no keyword * MAT_ADD _ EROSION exists, optimizing the material parameter WF by adopting an interval reduction sequence based on a meta-model; and S4, adding * MATs _ ADD _ EROSION, namely a GISSMO failure model, and optimizing GISSMO failure model parameters by adopting an optimization method consistent with the step S3 and a target function. According to the method, based on a GISSMO failure model provided in commercial finite element software LS-DYNA, GISSMO failure model parameters are reversely solved and calibrated according to material mechanical property test data parameters; by adopting the LS-OPT, material parameters can be quickly identified, so that output engineering stress-strain curves of simulation and test can obtain relatively good consistency, and a reference canbe provided for establishment of a quick, automatic and high-precision failure material library.