151 results about "Domain transformation" patented technology

A steganographic digital watermark signal is decoded from host imagery without requiring a domain transformation for signal synchronization, thereby speeding and simplifying the decoding operation. In time-limited applications, such as in supermarket point-of-sale scanners that attempt watermark decode operations on dozens of video frames every second, the speed improvement allows a greater percentage of each image frame to be analyzed for watermark data. In battery-powered mobile devices, avoidance of repeated domain transformations extends battery life. A great variety of other features and arrangements, including machine learning aspects, are also detailed.

The invention relates to the field of image enhancement and discloses a Retinex-theory-based nonlinear image enhancement method and system. According to the method and the system, a gradient-domain-transformation-based edge-preserving filter manner is used for replacing the conventional Gaussian filter manner to obtain illumination components. On the one hand, the phenomenon of halo artifact near a high-contrast edge is effectively overcome; on the other hand, the edge-preserving filter of color images can be carried out in real time. In addition, a nonlinear compression function is constructed on account of the illumination components, a highlight area is suppressed and a dark area is effectively improved; the phenomenon of overexposure of the images is avoided; the image luminance is mapped to an ideal dynamic range; a self-adaptive local nonlinear enhancement manner is used for enhancing image details, so that the local details of reflection components are enhanced, the quality of the images is well improved; the enhanced images are closer to real natural sites; the details are more clear.

The invention discloses an image encryption method based on quantum chaotic mapping and fractional domain transformation. The method comprises the steps of firstly carrying out iterative scrambling on pixel points by use of Henon mapping; then multiplying a scrambled matrix and a row scrambling matrix to carry out x-direction and alpha-order DFRFT ((Discrete Fractional Fourier Transform); multiplying the transformed matrix and a column scrambling matrix to carry out y-direction and beta-order DFRFT; and finally carrying out a diffusion encryption operation on the transformed matrix by use of quantum Logistic chaotic mapping. According to the method, the defects of few parametric variables, simple system structure, pseudorandom and poor aperiodicity caused by the single use of a certain scheme such as spatial domain, transform domain and chaotic system in conventional methods are overcome, and experiment and simulation results show that the method has the safety higher than that of the conventional encryption method.

This invention relates to a network video encoder multidimensional code rate control method. The image control unit of an input video sequence containing macroblock strips, image frames or image set is divided into different dimensions of set spaces according to the time domain, space domain, time domain transformation, space domain, time domain transformation, space domain transformation and space-time domain transformation, a code rate control unit transmits the time-varying state of band-width, the occupation degree of a buffer zone and resource and process ability of multi-access terminal based on the multiple element isomeric network channels to carry out variable bit rate encode of the divided set space to match with the current video encoder resource, or network transmission code rate, the process ability of the receive end and the decoder, the video decoder reconstructs related multidimensional images to display image sequence based on the grammar ID plugged by received stream.

The invention relates to an image segmentation method having multi-layer segmentation networks based on Bayes framework with edge prior information. The method herein includes the following steps: S1.performing general classification on input data by using a full convolutional neural network, outputting scoring graphs of all types that have the same sizes of the input data, also extracting an implicit edge image from an internal feature layer of the full convolutional neural network; S2. extracting an explicit edge image from the input data by using an edge detection network; S3.performing first restraining on the types that are obtained from S2 by using domain transformation and conditional random field, obtaining an initial segmentation image; S4. transforming the explicit edge image obtained from S2 to an edge distance image; and S5. inputting the edge distance image to defined domain transformation, performing second edge restraining on the initial segmentation image that is obtained from S3, and obtaining a final segmentation result. According to the invention, the method can extract edge prior information through an external edge network and performs edge region segmentationand filtering on the result from the general segmentation by using the edge prior information, so that the method herein can increase the accuracy in segmenting a SAR image.

The invention belong to the technical field of communication and particularly relates to a low complexity three-dimensional (3D) beam forming algorithm based on angle-domain transformation. In allusion to single-cell multi-user-multiple input multiple output (MU-MIMO) downlink broadcast channel scenes, a 3D beam forming problem is modeled to form an optimization model of a minimum base station transmitting power, then the original problem is converted to an equivalence optimization problem on an angle domain through angle-domain transformation, the problem is subjected to dimensionality reduction relaxation by means of angle-domain sparse characteristics of 3D-MIMO channels, thereby, the calculating complexity of solution is reduced, and the algorithm complexity is never increased with increasing of the number of base station antennae. A simulation result shows that the angle-domain dimensionality reduction optimization algorithm can excellently approach original space all-dimension optimization algorithm.

The invention discloses a real-time color image filtering method based on nonlocal domain transformation. The method comprises the following steps: using a two-dimensional color image as a two-dimensional manifold on a five-dimensional vector space, and defining distance-weighted keeping transformation to obtain a multi-channel weighted isometric function; using recursive filtering to realize one-dimensional nonlocal domain transformation filtering of linear time; conducting separated one-dimensional nonlocal domain transformation filtering to the two-dimensional color image. The nonlocal domain transformation method can effectively involve nonlocal information into an image filtering algorithm. Under the premise that a small amount of computation is added, a high-quality filtering image is effectively obtained, and meanwhile the two aspects of requirements that the edge information in the image is not fuzzy are taken into account. A single color (or gray) image is adopted as an input source, the algorithm complexity is lower, the filtering effect is good, and a variety of computer vision application requirements can be met. The method is especially suitable for image detail enhancement, picture stylizing, image colorizing and pencil drawing conversion.

The invention provides an internet of things uniform identification code analysis method and system. The method comprises the steps that an internet of things uniform identification code is analyzed to acquire version information and code system identification information of the internet of things uniform identification code; a regular expression of the domain transformation rule of the main body code of the internet of things uniform identification code is acquired according to the version information and the code system identification information; a network resource identification URI address of an analysis server corresponding to the main body code is acquired according to the regular expression of the domain transformation rule of the main body code; the URI address is sent to ensure that a receiver sends analysis requests of the main body code to the analysis server corresponding to the URI address according to the URI address. The method is used for solving the technical problem that a code analysis method in the prior art is single and poor in compatibility.

The present invention provides an acoustic vector circular array mode-domain robust direction estimation method. The method includes the following steps that: the received signals P(t), Vx (t) and Vy (t) of an acoustic vector circular array acoustic pressure channel, an acoustic vector circular array vibration velocity x channel and an acoustic vector circular array vibration velocity y channel are obtained; the received signals Pe (t), Vex (t) and Vey(t) of a phase mode-domain acoustic pressure channel, a phase mode-domain vibration velocity x channel and a phase mode-domain vibration velocity y channel are obtained; Vex (t) and Vey(t) are rotated electronically, so that a combined vibration velocity Vec (t) can be obtained, a covariance matrix Repv is obtained based on Pe (t) and Vec (t), an unitary matrix Q is introduced to transform the matrix Repv, so that a real-value covariance matrix Rpv of mode-domain acoustic pressure and vibration velocity combined processing can be obtained; a steering vector a<~>r(Phi<~>) which has been subjected to mode-domain transformation and real-value processing is constrained, second-order cone programming is adopted to carry out solving, so that an optimal weight vector Omegarob can be obtained; and an output space spectrogram can be obtained based on the steering vector Omegarob which has been subjected to mode-domain transformation and unitary matrix real-value transformation, and a target direction can be obtained based on a spectrum peak position. With the method of the invention adopted, the problems of performance degradation and high signal-to-noise ratio processing threshold of a minimum variance distortionless response algorithm under a condition that related sound sources are difficult to distinguish and mismatching occurs can be solved. The method has the advantages of high resolution, high robustness, small amount of calculation, strong background noise suppression ability and the like.

The invention discloses a cross-domain target detection method based on a regional full convolutional network and self-adaptation, and belongs to the technical field of computer vision. According to the method, a deep learning target detection technology is used, and the cross-domain robustness of target detection is improved by using an adaptive method in allusion to the problem of different distribution of data of a training domain and a test domain in target detection. The method comprises steps of firstly, constructing a regional full convolutional network model based on deep learning; designing two corresponding domain classifiers on the image level and the target level as adaptive components to reduce the difference of domain transformation, and adding consistency regularization to the domain classifiers; training the network in an end-to-end mode; and finally, removing adaptive components, and applying the network to a target detection task. By adopting the cross-domain target detection method designed by the invention, the average precision of target detection in various domain transformation scenes can be effectively improved.

The invention discloses a data transmission system based on an FPGA and a DSP. The data transmission system comprises an FPGA module, a DSP module and an external memory. The FPGA module comprises a memory read-write control unit, a data processing unit, an asynchronous FIFO unit, a synchronous FIFO unit and a data read-write detection unit. The memory read-write control unit reads the data of predetermined units from the external memory. The data processing unit is used for processing the data of all the units. The asynchronous FIFO unit is used for conducting clock domain transformation on the data of the predetermined units. The synchronous FIFO unit caches the data of the predetermined units, and an interrupt signal is sent to the DSP module when data arrive at the predetermined units. The DSP module reads the data from the synchronous FIFO unit according to the interrupt signal. When the data read-write detection unit detects that the DSP module reads the data of one unit every time, a feedback signal is sent to the memory read-write control unit, so that the memory read-write control unit reads the data of the next unit from the external memory. The data transmission system based on the FPGA and the DSP can guarantee the data transmission continuity between the FPGA and the DSP under an EMIF transmission mode.

The invention provides a single image defogging method based on improved non-local prior and a medium, and belongs to the technical field of computer vision and the technical field of image processing. The single image defogging method comprises the following steps: firstly, carrying out atmospheric light estimation by using an improved dark channel prior method, and then processing the whole image by using non-local prior to estimate the initial transmissivity, wherein global information is considered in non-local prior, and the image does not need to be segmented into different blocks, so that the phenomenon of wheel halo can be avoided; secondly, refining the scene transmissivity by adopting context regularization based on a weighted L1 norm to obtain refined scene transmissivity; and finally, correcting the transmissivity by using domain transformation filtering. Evaluation indexes such as peak signal-to-noise ratio, visible edge number, average gradient, saturated pixel point number and the like are improved.

The invention discloses a hyperspectral image vegetation classification method for spatial feature adaptive optimization. The method comprises the steps that first, normalization processing is performed on a hyperspectral image, PCA dimension reduction is performed on the hyperspectral image obtained after normalization processing, morphological filtering is performed on the hyperspectral image obtained after PCA dimension reduction, and a spatial information set is acquired; second, domain transformation standard convolution filtering is performed on the hyperspectral image obtained after normalization processing for spatial information extraction, and spatial information linear superposition fusion is performed on the acquired spatial information set; and finally the spatial informationset obtained after fusion is classified. Through the method, spatial texture information and relevancy information are effectively utilized, filtering parameters of a best classified structural type is further optimized, a good filtering effect is achieved, and classification precision of the hyperspectral image is effectively improved.

The invention provides a spatial information adaptive fusion-based hyperspectral image classification method. According to the method, spectral information is extracted as spatial information by usingtwo filters and is classified, so that spatial texture information and correlation information are effectively utilized; through domain transformation standard convolution filtering, a certain spatial texture information can be extracted, and relatively good spatial correlation is kept, so that the deficiency that only the spatial texture information can be extracted through bilateral filtering is made up for; and after adaptive fusion of the spatial texture information extracted through the bilateral filtering and the domain transformation standard convolution filtering, optimal classification performance is obtained through SVM classification, and a non-uniform phenomenon is effectively eliminated, so that the method is especially suitable for hyperspectrum with relatively numerous ground objects and complex distribution.

The invention relates to the hydroacoustic engineering technology field, particularly relates to a circular array modal domain orientation estimation method based on space sparse constraint and aims to solve problems that multi-shot sampling data is required to construct a covariance matrix, inversion or characteristic value decomposition is required for the covariance matrix and algorithm execution efficiency is relatively low existing in a modal domain orientation estimation method in the prior art. The method comprises steps that a uniform circular array sound pressure signal is acquired; aphase modal domain transformation matrix is constructed; a sound pressure signal received by a uniform circular array is transformed into a phase modal domain signal through a phase modal domain transformation matrix; the 360-DEG full space orientation is discretized, and a space domain sparse transformation base is further constructed; time domain sampling of a uniform circular array reception signal is carried out to construct multi-shot sampling data in a phase modal domain, and optimization and solution of multi-shot sparse signals are carried out through utilizing the l1 norm; signal energy in different orientation angles is calculated, a spatial spectrum is drawn, and the target orientation is acquired through the maximum peak position. The method is advantaged in that the method issuitable for signal estimation of the hydroacoustic engineering.

The invention provides a depth image optimizing method based on natural scene statistics. The method includes providing a flat image and an initial depth image of the flat image, extracting brightness information of the flat image, conducting domain transformation filtering on the brightness information, conducting wavelet decomposition on the filtered brightness information and the initial depth image, building a Gaussian model for a brightness wavelet pyramid, estimating a shape parameter and an extension parameter, building an energy function of a depth wavelet pyramid, the shape parameter and the extension parameter, minimizing the energy function to obtain the optimized depth wavelet pyramid, and conducting wavelet synthesis on the optimized depth wavelet pyramid to obtain the optimized depth image. By means of the optimizing method, the depth image can be optimized quickly, and depth information accuracy of the depth image is improved, so that the stereo video quality can be improved after the depth image is converted into stereo video, and user viewing experience is improved.

The invention provides an improved frequency dispersion ultrasonic guided wave signal domain transformation method, and belongs to the field of ultrasonic guided wave structure health monitoring. Themethod includes the following steps: (1) obtaining the frequency dispersion wave number curve and the frequency non-dispersion wave number curve of an ultrasonic guided wave signal mode; (2) calculating the ultrasonic guided wave distance domain excitation waveform with a reduced space width; (3) calculating an ultrasonic guided wave distance domain impulse response signal; and (4) calculating a resolution enhanced non-dispersive ultrasonic guided wave distance domain signal. Compared with traditional time-distance domain signal transformation methods, the method provided by the invention recompresses a frequency dispersion-expanded wave packet in the original ultrasonic guided wave time domain signal through dispersion compensation and changes the distance scale of the excitation waveformin order to further reduce the distance domain width of each wave packet in the ultrasonic guided wave signal, so the spatial resolution of the original frequency dispersion ultrasonic guided wave signal is effectively enhanced.

The embodiment of the invention provides a multi-modal remote sensing image change detection method, a model generation method and terminal equipment. A multi-modal remote sensing image change detection model comprises two domain converters, two condition discriminators and two change decision models. The generation method comprises the following steps: constructing two domain converters and two condition discriminators by utilizing a cyclic consistency adversarial network so as to be used for cross-domain conversion among different modal remote sensing images; utilizing a twin neural network to construct respective change decision models of different image domains so as to perform change detection on two images which are subjected to cross-domain conversion to the same image domain; constructing a first target function corresponding to cross-domain conversion and a second target function corresponding to change detection; and performing model training based on the two objective functions. According to the technical scheme, the parallel multi-network integrated model is constructed by combining the cyclic consistency adversarial network and the twin neural network, so that the image domain difference between the multi-modal remote sensing images can be effectively eliminated,.

The invention provides a display method and system for optical fiber detection and an optical time domain reflectometer. According to the method, the system receives and collects reflected signals of tested optical fibers based on testing signals sent to the tested optical fibers to obtain detection data of the corresponding reflected signals, and the dead zone width of events in the tested optical fibers is preset; wavelet domain transformation is carried out on the collected detection data, and an extreme point is determined in high-frequency coefficients; detection data with a peak value and a valley value are determined in the dead zone width behind the detection data corresponding to the extreme point, and the detection data corresponding to the extreme point, the peak value and the valley value are used as key points; inter-partition average calculation is carried out on the other detection data spaced by the determined key points, and detection data are selected from the detection data of partitions as the key points corresponding to the average value of the corresponding partitions; the determined key points are connected and displayed according to the time sequence. Therefore, the detected waveform is displayed accurately and fast.