61results about How to "Good denoising effect" patented technology

The invention discloses an image de-noising method based on a compression convolutional neural network, including the following steps: constructing a training data set; constructing a compression de-noising convolutional neural network model; using the training data set to train the network model; and inputting a noisy image to the trained network, and subtracting the output image of the network from the noisy image to get a clear de-noised image. The de-noising convolutional neural network in the invention is mainly characterized in that the convolution layer of the original de-noising convolutional neural network is replaced with a convolution layer decomposed and compressed by a low-rank matrix. By improving the existing de-noising convolutional neural network DnCNN, the network parameters are reduced by at least 75%, the network is simplified, and an excellent de-noising effect is maintained.

The invention discloses a method and device for performing adaptive denoising on a video image. The device comprises a video image reading module, an edge detection module, a filtering mode selection module and a filtering module; the method comprises the following steps of: reading in a frame of video image from a buffer zone, respectively calculating motion vectors between each pixel in the frame of image and the pixel at the same location in a neighboring frame, and respectively executing the edge detection on each pixel in the frame of image; respectively selecting the filtering mode having corresponding intensity for each pixel according to the edge detection result and the calculation result of the motion vector and of each pixel, and then, correspondingly filtering each pixel by a Gaussian similarity filter. The method and the device have obvious denoising effect on various noises of different scene sequences, and can effectively remove noises by combination of a time domain and a space domain.

The invention discloses a far-field speech recognition processing method and device. The method comprises the steps that a far-field speech is received; the far-field speech is input into a neural network-based speech training model obtained through pre-training; interference removing processing is conducted on audio features of the received far-field speech through audio features of far-field speeches and near-field speeches in the speech training model, and the processed far-field speech is obtained; the processed far-field speech is recognized. According to the method, optimization processing on the far-field speech can be achieved, a better processing result can be acquired, and equipment cost input is reduced.

The invention discloses a method for removing ocular artifacts in EEG signals. The two methods of ICA and MEMD are combined, the MEMD is used for extracting ocular artifacts from the EEG signals completely and retaining part of EEG information, so as to avoid the loss of the part of EEG information when the ocular artifacts are removed; the ICA is used for separating the ocular artifacts from the EEG signals, judging the artifacts through calculating a fourth-order cumulant of independent components, and then removing the electro-oculogram. The method for removing ocular artifacts in EEG signals can not only identify and remove the ocular artifacts automatically, but also retains a large number of EEG information, thereby providing a new idea for EEG de-noising; the method can not only remove the ocular artifacts effectively, but also retains a large number of useful EEG information; compared with the independent ICA method, the IMEMD method has better de-noising effect, can further improve the signal to noise ratio and reduces the mean-square error of signals.

The invention discloses a low-dose CT image denoising method, and mainly solves the problems of image edge blurring and detail loss caused by the fact of only considering restoring image grayscale information in the prior art. The implementation scheme is that 1) multiple low-dose and full-dose CT images are acquired from the same part of the body; 2) the acquired CT image data set is extended andnormalized and then block taking of each pair of CT images is performed so as to obtain a CT image block data set; 3) a twelve-layer full convolutional denoising neural network is established, the CTimage block data set is used as the network training data and the network is optimized by using a small batch gradient descent algorithm with the momentum so as to obtain the trained network; and 4)a complete low-dose CT image is inputted to the network so as to output the corresponding denoised CT image. The edge and the details of the image can be greatly kept in case of image denoising so that the method can be used for enhancement of the low-dose CT image and is convenient for the doctor to recognize the CT image case.

The invention provides an underwater image denoising method based on a generative adversarial network, and the method comprises the steps: inputting an underwater image into a generation network composed of a plurality of residual blocks, and obtaining a feature mapping graph; mapping output feature mapping graph and clear water noiseless label image through VGG-19 network to obtain depth featurespace, calculating the perception cost of the feature mapping graph and a clear water noise-free label image in a depth feature space, calculating the perception cost, inputting the feature mapping graph output by a generation network into an adversarial network, and finally inputting a noise-containing underwater image into the generation network after training is completed, and outputting the noise-free image as a processed noise-free image. According to the method, the confrontation mechanism is introduced, the denoising effect is obvious, particularly, the edge texture information in the image can be effectively reserved and even enhanced through the method, and the better visual effect and imaging quality are achieved.

The invention discloses a video de-noising method for Poisson, Gaussian and impulse mixed noises and belongs to the image processing field. The method includes the following steps that: each frame of image of a noise video is divided into image blocks, similar image blocks are searched in the frame sequence of the noise video for each image block through a matching method; Poisson-Gaussian mixed noise priors are constructed, a video de-noising model under a Poisson, Gaussian and impulse mixed noise environment is built; and the de-noising model is solved through using an optimization algorithm, and de-noised clear image blocks are obtained, and de-noised clear video images are generated based on the de-noised clear image blocks. A traditional video de-noising method is only used for a certain type of noises. The method of the invention can de-noise Poisson, Gaussian and impulse mixed noise videos, and therefore, with the method of the invention adopted, a video de-noising problem under a practical noise environment can be better solved. Compared with the traditional video de-noising method, the method of the invention can obtain higher-quality video de-noised images.

The invention relates to a bearing signal denoising method based on an improved wavelet algorithm, and belongs to the technical field of fault detection of mechanical and electrical equipment bearings. The technical scheme comprises the following step of firstly performing wavelet transform on a noisy signal to obtain a wavelet decomposition coefficient; obtaining an estimated wavelet decomposition coefficient by selecting appropriate threshold and threshold function for appropriate threshold and threshold function for processing; and reconstructing the estimated wavelet decomposition coefficient to obtain a denoising signal. The bearing signal denoising method based on the improved wavelet algorithm provided by the invention uses an improved progressive method to carry out denoising research on the wavelet threshold and the wavelet threshold function. From the selection of the traditional threshold function, the improved threshold function, and the traditional threshold and the improved threshold, the combination of the traditional threshold function with the improved threshold has the best denoising effect, the combination of the traditional threshold function with the improved threshold has the secondary denoising effect, and the combination of the traditional threshold function with the traditional threshold has the worst denoising effect. In the wavelet denoising algorithm, the improved wavelet threshold method can obtain good denoising effect.

A hyperspectral image denoising method combining L0 gradient constraint and local low-rank matrix recovery comprises the following steps: 1) obtaining hyperspectral image data to be denoised, and defining a hyperspectral image denoising model; The method comprises the steps of (1) obtaining a to-be-processed hyperspectral image, (2) calculating an L0 gradient matrix of the to-be-processed hyperspectral image, (3) establishing a local low-rank constraint-based model, and (4) establishing a denoising model in combination with L0 gradient constraint and local high-rank property of the local high-spectral image, and recovering the three-dimensional noise-free high-spectral image.

The invention relates to a dictionary learning algorithm based on sparse model analysis. According to the algorithm, a cost function is established directly through noise signals, the cost function is solved through a gradient descent method, and it is guaranteed that the value obtained through iteration conducted each time is optimal within the local range. A dictionary is adaptively updated through the algorithm, so that a super-complete dictionary capable of better expressing the image structure is formed. The dictionary learning algorithm can be used for image noise removal, image quality can be improved, more accurate target and background information can be provided, and the ideal noise removal effect can be achieved. The algorithm is widely applied to the military field and non-military field of target detection, optical imaging, safety monitoring systems and the like.

The invention relates to an image denoising method based on a pixel-level global noise estimation coding and decoding network, and the method comprises the steps: inputting an original noisy image into an input module of the coding network, carrying out the preliminary feature extraction of the original noisy image through convolution, and outputting an original feature map; processing the original feature map through a plurality of cascaded coding modules in the coding network, and outputting a denoised high-level feature map with a relatively small spatial size and a relatively high semanticlevel; processing the high-level feature map through a plurality of decoding modules with skip connection structures in a decoding network symmetrical to the encoding network, and obtaining an outputfeature map after noise removal considering space information and semantic information; and mapping the output feature map of the decoding network to an output feature dimension by using convolutionprocessing through an output module of the decoding network, and outputting a final clear image. According to the method, real image noise, global information and pixel value related characteristics are fully considered, and the denoising effect and the operation speed are both considered.

The invention discloses an image processing method of electrophoretic electronic paper detection equipment, which comprises the following steps of: 1, shooting electronic paper for multiple times to obtain a sample image set consisting of a plurality of electronic paper photos polluted by light; 2, randomly selecting one sample image in the sample image set, converting the sample image into a grayscale image, and drawing a three-dimensional image of the sample image; 3, carrying out optical noise removal on the sample image by adopting Gabor wavelets to obtain a filtered sample image, and drawing a three-dimensional image of the denoised sample image; 4, calculating a variance value of the filtered sample image and the original sample image by using a variance formula S, if the variance value of the filtered sample image is greater than 60% of the variance value of the original sample image, entering a step 3, otherwise, entering a step 5; 5, acquiring a filtered image, and optimizingthe filtered image; The image content of the shot electronic paper is consistent with the display content of the electronic paper, and the reliability is high.

The invention relates to a partitioned bilateral total-variation regularization image noise elimination method. The method comprises the steps that (1) a pollution image X<0> is acquired and is used to initialize a denoised image subjected to the first iteration, and then the step (2) is entered; (2) a partitioned bilateral structure similar distance matrix DW<t> of a denoised image subjected to the t(th) iteration is calculated, and then the step (3) is entered; (3) partitioned bilateral total-variation regular terms of the denoised image subjected to the t(th) iteration are constructed, and then the step (4) is entered; (4) an energy functional E<t> composed of fidelity terms and the partitioned bilateral total-variation regular terms is constructed, and the step (5) is entered; (5) a steepest descent method is adopted to solve a minimization problem of the energy functional E<t>, a denoised image subjected to the (t+1)(th) iteration is obtained, and the step (6) is entered; and (6) whether the number of the iterations is smaller than the maximum number N of iterations is judged, if the number of the iterations is smaller than the maximum number N of iterations, t is made to be equal to t+1, and the step (2) is entered, and otherwise the denoised image subjected to the (t+1)(th) iteration is output to end the operation.

The invention provides a clustering-based pressure vessel self-adaptive leakage detection method. The method comprises the following steps of firstly, receiving an acoustic emission wave of a pressurevessel by using an acoustic emission sensor; then carrying out local mean value decomposition on a mixed observation signal, and combining a wavelet threshold to denoise the signal so as to obtain anoriginal signal of an acoustic emission signal; carrying out characteristic extraction on the signal to construct a characteristic vector set of the signal, using a weighted fuzzy C-mean value algorithm to carry out clustering analysis, and combining priori knowledge to identify a leakage mode; and finally, training a classifier by using existing various leakage mode label data and an initial clustering result, and then dynamically updating a diagnosis model by using sample information acquired in real time to construct an online leakage mode adaptive identification model. An unknown sample can be identified in real time in an online mode, and adaptive adjustment can be realized.

The invention relates to a method for extracting bearing fault feature frequency based on singular value decomposition and optimized frequency band entropy and the application thereof, and belongs tothe field of mechanical fault diagnosis and signal processing. The concept of the singularity kurtosis value relative change rate is provided based on kurtosis indexes, and the SVD reconstruction order is determined by adopting a singular kurtosis value relative change rate. The method is simple in principle and takes kurtosis value as a theoretical basis and has a solid theoretical basis when compared with other methods. The denoising effect can be better than that of other methods. After the SVD reconstruction order is determined, a reconstructed signal is obtained, and an optimized band-pass filter is designed by utilizing the optimized frequency band entropy, further noise reduction processing is carried out on the reconstructed signals, and an analysis result is good in effect. According to the method, the bearing fault feature frequency can be effectively extracted. The method is applied to bearing simulation signals and actual bearing signal analysis, and has wide practicability.

The invention discloses an improved threshold denoising algorithm and provides an improved adaptive threshold denoising algorithm based on a LoG operator. The improved adaptive threshold denoising algorithm based on the LoG operator aims to solve the problem that in the denoising process, the denoising effect of the edge part and the non-edge part of an image is poor. The improved adaptive threshold denoising algorithm based on the LoG operator comprises the steps that firstly, edge contour information of the image is extracted through the LoG operator; secondly, the non-edge part of the image is denoised, a threshold correction coefficient is added to a soft threshold function, and then a new threshold function is established; thirdly, the edge part of the image is denoised, energy nearby a wavelet coefficient and a threshold are combined, and then a new threshold function is established; fourthly, an R channel, a G channel and a B channel of the image are processed, and all detail information of the image is reserved. Experimental results show that by means of the improved adaptive threshold denoising algorithm based on the LoG operator, edge information of the image is effectively stored, and the comprehensive denoising effect is improved remarkably.

The invention discloses a quick ISO (international standardization organization) denoising method and a quick ISO denoising system based on lifting wavelet transform, relates to the image processing technology, and aims to disclose a quick ISO denoising algorithm which can be used for simultaneously retaining low-contrast information and high-contrast edge information of images. The technical key point is that the quick ISO denoising method comprises the following steps: performing single-channel decomposition on images, thereby obtaining a Y-channel image, a U-channel image and a V-channel image; performing n-layer linear lifting wavelet transform on the Y-channel image; one by one denoising wavelet detail images in results obtained by transforming wavelets of the third layer to n-th layer; performing inverse wavelet transform to obtain denoised similar images in first-layer wavelet transform results; denoising the similar images in the first-layer wavelet transform results again; performing inverse wavelet transform to obtain the Y-channel image of original space; denoising the Y-channel image of the original space; respectively performing wavelet transform on the U-channel image and the V-channel image and denoising the U-channel image and the V-channel image; and obtaining denoised results of the Y-channel image, the U-channel image and the V-channel image.

The invention discloses a method for processing an infrared digital video signal at night, which comprises the following steps of: firstly, converting a monitoring video image represented by RGB color codes into an image represented by YUV color codes to acquire the component Y of a gray value in the monitoring image represented by the YUV color codes; secondly, in a loop filtering link, performing filtering on the component Y of the gray value of an image pixel represented by the YUV color codes through a mathematic morphological algorithm, performing an erosion operation on the image converted into the format of YUV color codes by the mathematic morphological algorithm and by using open operation according to the property of the mathematic morphological algorithm, and then performing a dilation operation on the image; and finally, performing software filtering on the component Y subjected to the software filtering of the mathematic morphological algorithm by a median filtering algorithm. In the method, the monitoring image at night is subjected to denoising enhancement by combining the mathematic morphological algorithm and the median filtering algorithm, and practice tests prove that the method has the excellent denoising effect.

The invention discloses a method for processing spots of a three-dimensional ultrasonic image, belonging to the technical field of the digital image processing. The method comprises the following steps: (1) decomposing the three-dimensional ultrasonic image into seven high-frequency parts and one low-frequency part by using the time and frequency separation characteristic of the multi-resolution of a three-dimensional wavelet; and (2) processing the high-frequency parts of the three-dimensional ultrasonic image subjected to wavelet decomposition by adopting an normalized anisotropic diffusion method according to the multiplicative properties of the spots of the three-dimensional ultrasonic image so as to remove the spots; and (3) reconstructing the three-dimensional ultrasonic image by carrying out three-dimensional wavelet inverse transformation to the processed high-frequency parts and the original low-frequency part. The method disclosed by the invention can be used for rapidly and effectively inhibiting the ultrasonic spots to improve the quality of the image, is very beneficial to the cutting, segmentation, feature analysis and recognition of the ultrasonic image and can be also used to process the image with multiplicative noise.

The invention discloses an elimination method for eliminating the stripe noise of an infrared image. The method comprises the following steps of 1, inputting an original infrared image; 2, setting an iteration times K, an iteration sequence number K and conducting the iteration initialization; 3, calculating an infrared image X k (i, j) after the k correction process; 4, calculating the module (img file = 'DDA 0001218487710000011. TIF 'wi = '172 'he = '79' / ) of the infrared image gradient after the k correction process; 5, updating correction parameters Gk(j) and Ok(j); 6, judging whether the condition of k < K is satisfied or not; 7, outputting a final infrared image XK (i, j). The method of the invention is novel in design and simple in step. For the stripe noise generation mechanism of the infrared image, correction parameters are estimated through minimizing the energy function of a correction image. Meanwhile, the stripe noise of the image is corrected through linear transformation. Therefore, the stripe noise of the image can be eliminated only by utilizing the infrared image itself. In this way, the periodic calibration of a focal plane array by means of a black body is not required. On the premise that the stripe noise of the infrared image is effectively eliminated, the blurring phenomenon of image details is avoided. The method is small in calculation amount, good in real-time performance and convenient to popularize and use.

The invention relates to a bilateral regression filter method for grayscale and colored images. As an effective information carrier, images can be used as an important manner for obtaining and exchanging effective information by users; but the images systematically import certain noises due to various interference factors in the forming, transmission and reception processes, some detail features are hidden in the noises, and great difficulties are brought to the image processing processes such as image observation as well as feature information extraction and analysis, so that adopting a proper method to remove pollution noises in the images is a very important pre-processing step. The bilateral regression filter method provided by the invention can be used for the filtering of grayscale and colored images.

The invention discloses a spinal CT image bilateral filtering denoising method based on feature selection, belonging to the medical image processing field, the method aims to overcome the defect thatimage quality cannot completely meet actual clinical operation and teaching research requirements after being processed through a conventional denoising method. The region of interest is enhanced while the irrelevant region in the spinal medical image is removed; the foundation is laid for later image segmentation, positioning, three-dimensional reconstruction and other operations. The method comprises the following steps: firstly, spine point cloud data in spatial transverse, longitudinal and vertical directions is gridded; the number of the point clouds in all the grids is judged, then the noise point clouds are divided into point clouds far away from the spine interest area and sparse, point clouds far away from the spine interest area and dense and noise point clouds mixed with the spine interest area, and the noise point clouds are removed in sequence to achieve the better denoising effect.

The invention discloses a denoising method for a remote sensing satellite image, and the method comprises the steps: loading an original remote sensing satellite image, and enhancing the image contrast by adopting an adaptive contrast enhancement method to obtain a to-be-denoised image; Dividing the to-be-denoised image into M * N bounded discretization grids to obtain a discretized to-be-denoisedimage; Detecting pulse noise in the discretized image to be denoised by adopting a median absolute difference judgment criterion to obtain a two-dimensional identification matrix F, wherein pixels inthe two-dimensional identification matrix F are in one-to-one correspondence with pixels of the image to be denoised; Setting a search window and a neighborhood window in the discretized image to bedenoised; And enabling the domain window to slide in the search window and denoising the domain window and by combining a similarity calculation method and a Gaussian weighted average calculation method, obtaining a denoised image. According to the method, excellent texture detail maintenance and a good image denoising effect can be realized, and meanwhile, the algorithm complexity is relatively low. Treatment efficiency is high.

The invention provides a GAN image denoising algorithm fused with an improved residual network. The GAN image denoising algorithm comprises: S1, preprocessing a data set image; S2, extracting features of the noisy image by a generator to generate a de-noised image; S3, judging the input image by the discriminator, and outputting a judgment result; and S4, performing alternate iteration training on the processes according to a loss function. Compared with a traditional residual network, the multi-layer residual feature extraction network creatively used in the invention retains the advantages of the original residual network, that is, the problem of gradient disappearance or explosion caused by a single-stack convolutional neural network is solved, meanwhile, the multi-layer residual feature extraction network also realizes the extraction of deep-level features and shallow-level detail feature information of the input picture, and the residual network used in the invention also reduces model network parameters. According to the invention, the dual-channel discriminator construction model is used, so that the discrimination capability of the discriminator can be well improved, the generator G can be well trained, and a picture with a better denoising effect can be generated.

The invention relates to a self-adaptive noise reduction method based on empirical mode decomposition and a wavelet threshold. Firstly, EMD (empirical mode decomposition) is carried out on a noisy bearing vibration signal to obtain limited intrinsic mode components imf(t), the step replaces a traditional wavelet decomposition part, and the problem that wavelet basis function types and decomposition layers need to be preset in traditional wavelet decomposition is solved; then a proper threshold value and a threshold value function for the noisy imf(t) are respectively selected for processing, and an estimated f(imf<j,k>(t)) is obtained; and finally, loop iteration is carried out on the estimated f(imf<j,k>(t)) until the RMSE denoising condition is optimal. According to the method, theproblem that the optimal denoising effect is achieved by manually and repeatedly setting the number of decomposition layers during traditional wavelet reconstruction is solved.

The invention discloses an unsupervised video segmentation method based on non-local space-time characteristic learning. The method comprises the following steps of acquiring a video sequence which needs to be segmented; using a superpixel to segment and process the video sequence; using a light stream to carry out previous and next frame information matching; according to adjacent frame information of the video sequence, acquiring a range of a motion target and taking as model initialization input; using global information to optimize a matching result; and establishing a graph model and using a graph segmentation algorithm to solve a segmentation result, and through video segmentation, acquiring output of the motion target. Through carrying out superpixel segmentation on each frame of image in an input video, an operation complexity can be greatly reduced; and non-local space-time information is used to optimize matching information acquired through the light stream so that segmentation robustness can be increased and a noise influence is reduced. Any manual intervention is not needed, and an accurate segmentation result can be acquired completely based on video image information.

The invention discloses a method for removing small area noises in a binary image based on a stack theory, and the method can reduce operation complexity, decrease operation time and has relatively high robustness. The method comprises the following steps: a, when a stack is empty, each pixel which does not enter the stack in the binary image is scanned sequentially; b, if a present pixel is black, the present pixel is put into the stack and it is marked that the pixel which is already put into the stack; c, in the moment, the stack is not empty; a stack top is enabled to be out of the stack and areas of communicated regions are recorded in a progressive increasing mode, and eight points around the pixel which is marked to be in the stack are sequentially scanned; d, the step b and the step c are cycled to complete scanning of the whole communicated region; whether the communicated region is smaller than a set threshold is determined; if the communicated region is smaller than the set threshold, the communicated region is removed; and if the communicated region is not smaller than the set threshold, the communicated region is saved.

The embodiment of the invention discloses an image segmentation method and device, computer equipment and a storage medium, and the method comprises the steps: setting the pixel value of a first pixelpoint in a thermodynamic image as a first preset temperature value according to a background thermodynamic image corresponding to a to-be-segmented image, so as to obtain an infrared preliminary denoised image; determining a temperature value weight corresponding to each second pixel point in the infrared preliminary denoised image according to the temperature values of the second pixel points inthe infrared preliminary denoised image; setting a pixel value of an abnormal pixel point of the infrared preliminary denoised image as a second preset temperature value to obtain a target denoised thermal image, the abnormal pixel point being a second pixel point whose temperature value weight is less than a first noise point threshold; and processing the to-be-segmented image according to the target denoising thermal image to obtain a target segmented image. In this way, segmentation of multiple targets in the image can be achieved.

The invention relates to a dark light image denoising method based on dense connection convolution, and the method comprises the following steps: constructing a dense connection denoising convolutional neural network model, and carrying out the training of the dense connection denoising convolutional neural network model; acquiring a to-be-processed image, and preprocessing the to-be-processed image to obtain a target to-be-processed image; and processing the target to-be-processed image through the trained dense connection denoising convolutional neural network model to obtain a denoised image. According to the method, a large amount of image detail information can be retained even when an image in a dark light or backlight scene is processed, a high-quality denoised image is obtained, and a good denoising effect is achieved.