48 results about "Exponential transformation" patented technology

A medical ultrasound image denoising method based on wavelet transform and quick bilateral filtering comprises the following steps of the first step of establishing a medical ultrasound image model, the second step of carrying out wavelet transform on an image obtained in the first step after logarithm transformation to obtain four frequency domains (LL1, LH1, HL1and HH1), continuously carrying out wavelet transform on the low frequency domain LL1 to obtain four frequency domains (LL2, LH2, HL2 and HH2) again and then repeating the step until resolving the maximum layer number, the third step of carrying out quick bilateral filtering on the low frequency portion (LLJ) at the last layer, the fourth step of carrying out threshold value method shrinkage on wavelet coefficients of the high frequency portions (LHj, HLj and HHj) of all layers and the fifth step of carrying out wavelet inverse transformation to obtain the denoised medical ultrasound image. In addition, the J is 1, 2, ..., J. If denoised ultrasound envelope signals are wanted, exponential transformation is carried out on the ultrasound image obtained in the fifth step.

The invention discloses a self-adaptive de-noising and characteristic enhancing method of an SAR (Synthetic Aperture Radar) image, which mainly aims at overcoming the defects that the traditional method has poor de-noising property and can not self-adaptively select parameters in the experiment. The method comprises the following steps: firstly, carrying out logaritmim and then ME-curvelet transformation on an original SAR image; secondly, self-adaptively selecting and optimizing parameters in improved gain functions by adopting an improved PSO (Particle Swarm Optimization) algorithm according to the provided evaluation criterion; and finally, carrying out the nonlinear transformation on ME-curvelet coefficients, ME-curvelet inverse transformation and exponential transformation by adopting the improved gain functions to obtain the final SAR image subjected to the de-noising and characteristic enhancement. By using the method, the noise can be removed while the characteristics are enhanced, the complexity for processing can be reduced and the better de-noising and characteristic enhancing effects of the SAR image can be achieved.

The invention discloses a method for de-noising a medical ultrasonic image based on wavelet transformation and a guide filter, and the method comprises the following steps of: step 1) creating a medical ultrasonic image model; step 2) performing wavelet decomposition to the logarithm transformation image in the step 1) to obtain four frequency domains (LL1, LH1, HL1 and HH1); performing wavelet decomposition to LL1 in the low frequency domain so as to obtain four frequency domains (LL2, LH2, HL2 and HH2); repeating the step until to decompose the maximum layer J; step 3) performing filter processing to a wavelet coefficient in the final-layer low-frequency part (LLJ) by the guide filter; step 4) performing threshold valve method contraction treatment to the wavelet coefficient in each-layer high-frequency part (LHj, HLj and HHj, j=1, 2, ......, J); step 5) performing wavelet inverse transformation treatment to gain the de-noised medical ultrasonic image; performing exponential transformation to the medical ultrasonic image obtained in the step 5) to obtain de-noised ultrasonic envelope signals.

The invention discloses a PCNN-based method for de-noising a wavelet domain ultrasonic medical image, which comprises the following steps: firstly, performing logarithmic transformation and wavelet transformation on a noise image and the corresponding pretreatment on a wavelet coefficient; secondly, processing the wavelet coefficient by using a PCNN method and performing the corresponding post-treatment on the wavelet coefficient; and finally, performing the inverse wavelet transformation and exponential transformation on the wavelet coefficient subjected to the post-treatment to obtain a de-noised image. Compared with wavelet de-noise, the method makes the edge of the de-noised image clearer, well retains image details and improves a signal to mean square error ratio; and compared with PCNN, the method overcomes the drawback that during the removal of speckle noises, the PCNN has difficulty in determining model parameters and step length and realizes a higher signal to the mean square error ratio after image de-noise.

A method for determining an individual's intensity of focused attention, comprising the steps of: obtaining a representative frontal lobe brainwave signal from at least one first sensor in an electrically connective relation to the individual's frontal lobe; obtaining a representative reference signal from at least one second sensor in an electrically connective relation to a more electrically-neutral location; subtracting the representative reference signal from the representative frontal lobe brainwave signal to produce a difference frontal lobe brainwave signal, and processing the difference frontal lobe brainwave signal to produce an Attention Indicator signal indicative of the individual's intensity of focused attention, where the Attention Indicator signal is inversely proportional to any mathematical transformation of an amplitude measure of the difference frontal lobe brainwave signal; inputting the Attention Indicator signal to a device; and, repeating these steps, as desired. The mathematical transformation can relate to amplitude, power or any linear, logarithmic or exponential transformation thereof.

The invention discloses an illumination normalization method for processing face images. The illumination normalization method includes steps of reading face images to be processed and taking logarithm of the face images to be processed; computing images of a shadow layer; computing images of a reflecting layer and performing exponential transformation; selecting sample images and computing a histogram of the sample images; normalizing the images of the reflecting layer by a histogram matching method to obtain the images of the to-be-processed face images corrected by the illumination normalization method. By an edge-preserving filter, feature information on a large-scale layer can be kept. By adopting the sample images in the optimum vision area, mean value and variance of the corrected images of the reflecting layer can be kept in the optimum vision area, a subsequent face identification system is facilitated, accuracy in face identification can be improved, and the problem that the feature information is greatly lost to cause disadvantages to subsequent face identification during face identification in existing illumination invariant feature extracting method is solved.

The invention discloses a speech-emotion recognition method based on improved quadratic discriminant. By utilizing the method, false recognition rate caused by probability distribution diversity of emotional characteristic parameter statistics can be effectively reduced. The method comprises the steps of performing exponential transformation on characteristic parameters to enable parameter distribution after transformation to be near normalized, estimating probability distribution functions of original characteristics on the premise of subjecting transformed parameters to normal distribution, and using a logarithmic form to obtain improved quadratic discriminant. Compared with the prior other characteristic normalization transformation, the exponential transformation adopted by the method can more effectively normalize the characteristic parameters, and recognition rate can be effectively improved by adopting the improved quadratic discriminant.

The invention discloses a medical ultrasound image speckle removing method through quantum inspiration. The method comprises the steps that a medical ultrasound image containing speckle noise is input; logarithm transformation is carried out to convert the multiplicative noise image into the additive noise image; complex wavelet transformation is carried out to convert a grey value of the image into a wavelet coefficient; noise variance, and variance and smoothing parameters of a probability density function of an ideal image signal are estimated to obtain a noise statistic model and a statistic model of the ideal image signal; an adaptive adjustment threshold value is calculated according to the theory of quantum inspiration, and soft threshold processing is carried out on the wavelet coefficient to obtain a wavelet coefficient estimation value of the ideal image signal; the wavelet coefficient estimation value of the ideal image signal is used for wavelet reconstruction to obtain the image; exponential transformation is carried out on the image to compensate the logarithm transformation in the first step, and the image with speckles removed is obtained. The method can keep organization details in the image well on the basis of effectively removing the speckle noise of the medical ultrasound image, and play a good role in assisting in medical treatment.

The invention relates to the technical field of website data statistic analysis and particularly discloses a website page view prediction method based on historical tendency weights. The method comprises the steps of data preprocessing and prediction result processing, wherein in the data preprocessing step, the logarithm of page views is taken, the variance of all time points in the historical tendency is calculated, the regression coefficient of the ith historical tendency to the current tendency is calculated, the variance of correlation coefficients of a current value estimated according to the ith historical tendency is calculated, deviation is estimated through an optimization minimization method to solve the weights, and the value appearing after the current tendency is predicted according to the weights; in the prediction result processing step, a prediction result is subjected to exponential transformation. According to the website page view prediction method, the known current tendency is compared with the historical tendencies for calculation of the correlation coefficients, the deviation of the current tendency is estimated according to all the historical tendencies, the weight of each historical tendency is selected through the optimization method, estimations at all dates are superposed according to the weights, and the current tendency and a follow-up tendency can be predicted according to a superposition result, so that the deviation of the estimations of the current tendency by synthesizing the historical tendencies is minimum, and reliable prediction is achieved.

The invention discloses a non-local mean image de-noising method with a noise intensity self-adaptation function. The non-local mean image de-noising method with the noise intensity self-adaptation function comprises the steps that firstly, a gray scale strip image is collected, and de-noising processing is conducted on the gray scale strip image by using different de-noising intensity parameters through a non-local mean method, so that optimal de-noising intensity parameters under the condition of different degrees of brightness are obtained; then, optimal de-noising intensity parameters corresponding to other degrees of brightness are calculated through a linear interpolation method; finally, de-noising processing is conducted on the image in a logarithm domain through the optimal de-noising intensity parameters corresponding to different degrees of brightness, and exponential transformation is conducted on the de-noised image in the logarithm domain, so that a final de-noised image is obtained. The non-local mean image de-noising method overcomes the defect that in an existing method, de-noising intensity parameters are fixed, and the de-noising effect of the image is improved; due to processing in the logarithm domain, the difference of brightness of pixels in a dark region can be increased, the difference of brightness of pixels in a bright region can be reduced, and the de-noising effect of the image can be improved.

The invention relates to a small aircraft attack angle sliding mode tracking method based on an inverse solution transfer function, and belongs to the field of rehabilitation robots. The method comprises the following steps: measuring an attack angle of an aircraft, forming an attack angle error signal with an instruction signal, and performing softening and exponential transformation to obtain asoftening and exponential error signal; carrying out integration to obtain a corresponding integration signal; then carrying out filtering differential to obtain an attack angle differential signal, and measuring a pitch angle rate signal to carry out linear combination so as to obtain a sliding mode surface signal; forming an equivalent nominal control item according to nominal model information,and forming a combined control quantity together with nonlinear transformation of a sliding mode surface and the attack angle error signal; and then through a nonlinear inverse solution transfer function method, weakening flutter brought by sliding mode control, generating a pitch channel control law, and realizing attack angle tracking control of the aircraft. The method has advantages of high attack angle tracking precision, a high anti-interference capability, small flutter and a good attack angle tracking dynamic effect.

The invention provides a registration method and system for a medical image with incompressible organs. The method comprises the steps that a fixed image and a to-be-registered image are received, andafter pre-registration is performed on the fixed image and the to-be-registered image, a first velocity field of current iteration after the to-be-registered image is registered to the fixed image isobtained; according to the first velocity field of the current iteration and a velocity field obtained in last iteration, a second velocity field of the current iteration is obtained through calculation based on a BCH formula; the second velocity field of the current iteration is decomposed through Hodges-Helmholtz decomposition, a blending field is magnified through an adaptive weight of the blending field, and the magnified blending field is superposed with a passive field to obtain a velocity field of the current iteration; and the velocity field of the current iteration is subjected to inverse transformation through exponential transformation, and a deformation field of the current iteration is obtained. Through the method, the solved deformation field has incompressible performance and can track the incompressible organs with sliding displacement at high precision, and registration precision and speed are improved.

The invention discloses an image denoising method through combination of adaptive nonlocal samples and low rank. Firstly an image is transformed to a logarithmic domain by using logarithmic transformation, and a multiplicative noise model is transformed into an additive noise model; the image is partitioned and grouping is performed according to the similarity so as to obtain image groups having similar blocks; then low-tank approximation processing is performed on the image groups so as to obtain an initial estimation value; then adaptive nonlocal sample model processing is performed on the initial estimation value so as to obtain a logarithmic domain recovery result; and finally the logarithmic domain image is restored to a real number domain by using exponential transformation and corrected so as to obtain a final denoised image. The experiment result indicates that the method has great robustness for the multiplicative noise, and the great peak signal-to-noise ratio and the structural similarity can be obtained and the visual quality of the image can be greatly improved for the image having the multiplicative noise.

The invention discloses an SAR (Synthetic Aperture Radar) image speckle reduction method based on sparse representation, which mainly solves the problems of incomplete detail information and unsmooth homogeneous regions for SAR image speckle reduction in the prior art. The method comprises the following steps of: carrying out logarithmic transformation on an original SAR image Y; carrying out overlapping blocking on an image Y' subjected to logarithmic transformation; taking the own information of the image blocks as control factors and carrying out sparse representation on the control factors; carrying out dictionary learning by adopting an approximate KSVD (K Singular Value Decomposition) algorithm to obtain a self-adaptive dictionary and updated sparse representation coefficients; obtaining an image W by adopting the self-adaptive dictionary and the updated sparse representation coefficients, and carrying out exponential transformation on the image W to obtain an image R; and carrying out nonlinear anisotropic diffusion on a difference image V obtained from the original SAR image Y and the image R subjected to the exponential transformation, thereby obtaining a final speckle-reduced image. The method disclosed by the invention has the advantages that the homogeneous regions of speckle reduction results are smooth, point targets are clear and visible, and edge information iskept complete and the like, besides, the method is suitable for a preprocessing process of SAR image understanding.

The invention provides an ultrasonic image noise suppression method which comprises the steps of: 1, carrying out logarithmic transformation on an ultrasonic image to obtain a logarithmic image; 2, extracting an edge signal area of the logarithmic image; 3, carrying out non-downsampling contourlet transformation to obtain a plurality of high-frequency sub-bands; 4, extracting an edge signal area and a signal stable area of each high-frequency sub-band; 5, carrying out noise reduction on each high-frequency sub-band in the edge signal area and the signal stable area; 6, carrying out non-downsampling contourlet inverse transformation on the high-frequency sub-bands subjected to noise reduction; and 7, carrying out exponential transformation on the image obtained in the step 6 to obtain a final noise reduction image. By adopting the noise reduction method provided by the invention, speckle noise can be effectively suppressed, detail information in the ultrasonic image is reserved largely, and a better noise reduction effect is achieved.

The invention discloses a non-subsampled shearlet transformation-based high-resolution three-SAR image speckle reduction method. The method comprises the steps of S1, converting an image noise model;S2, performing shear wave conversion; S3, processing a high-frequency coefficient by a wavelet hard threshold value; S4, shearing and reconstructing; S5, performing exponential transformation. According to the method disclosed by the invention, high-resolution three-SAR images with the good sparsity can be processed by adopting shear waves with the optimal sparse representation. Therefore, the non-subsampled shear wave transformation is adopted to replace the traditional shear wave transformation. The frequency spectrum aliasing phenomenon caused by the sampling operation is avoided. The gooddirectionality of shear waves is effectively utilized, so that the edge detail part of an image can be effectively reserved. Meanwhile, for the potential noise characteristic of a high-frequency wavelet coefficient, a wavelet hard threshold algorithm is adopted to process the noise. Therefore, the effect of filtering and removing spots is effectively achieved.

The invention belongs to the technical field of digital image copyright protection, and particularly relates to a color image zero watermarking method based on rapid quaternion generalized polar complex exponential transformation. The invention provides a color image zero watermarking method based on fast quaternion generalized polar complex exponential transformation, and the method comprises thesteps: firstly, calculating the fast quaternion generalized polar complex exponential transformation FQGPCET of an original carrier image, carrying out the moment selection, and constructing a hybridlow-order moment feature with the good robustness and discrimination; secondly, utilizing asymmetric Tent mapping with good security to encrypt and convert the feature vector into a feature map; then,, scrambling the feature map and the watermark image by using generalized Arnold transformation, and taking the XOR of the scrambling results of the feature map and the watermark image to obtain a zero watermark image which can be used for copyright authentication; and finally, in order to ensure that the copyright can be publicly verified, enabling the credible mechanism to authenticate the zerowatermark, the secret key and the copyright owner identity information and adds a digital timestamp.

The invention discloses a low frequency logarithmic spectrum based robust feature extraction method. Feature parameters can be extracted by using a logarithmic spectrum profile of a speech signal. Firstly, logarithmic transformation is performed on an amplitude spectrum of the speech signal to obtain a logarithmic spectrum. Then, the logarithmic spectrum is regarded as a time domain signal, and low-pass filtering is performed on the logarithmic spectrum by using a digital filter to obtain a low-frequency logarithmic spectrum. Finally, exponential transformation, Mel filtering, logarithmic transformation and discrete cosine transformation are performed on the low frequency logarithmic spectrum of the speech signal, and time domain difference is performed to obtain feature parameters of thespeech signal. The method can improve the environmental robustness of the feature parameters of the speech signal, reduce the influence of the speaker change on a speech recognition system, and has the advantages of small calculation amount and easy real-time realization.

The invention belongs to the technical field of machine vision, and particularly relates to an infrared target detection method based on target boundary positioning. In the algorithm, firstly, the contrast of an infrared image is enhanced by using adaptive exponential transformation, and semantic details of the image are enhanced; the improved composite Resnet-50 is adopted to extract characteristic spectrums under different scales in the image; feature fusion is carried out by adopting a bidirectional feature pyramid, and the detection capability of the model on weak and small targets is enhanced; and Focal Loss is adopted as a classification loss function of the model, a loss function based on target boundary positioning is adopted as a regression loss function of the model, and the detection model is trained. According to the invention, effective detection of the infrared target is realized.

The invention discloses a robust video zero-watermarking method based on polar complex exponential transformation and a residual neural network, and the method comprises the steps: carrying out the preprocessing of a video, and selecting a key frame of each group of shots; encrypting the original watermark of the video; obtaining an invariant moment of the key frame by means of polar complex exponential transformation; sending the invariant moment into a pre-trained deep residual neural network model; extracting robust content characteristics of the key frame; carrying out exclusive-OR operation on the robust content characteristics and the encrypted watermarks; generating a unique robust zero-watermark signal of the video, selecting a key frame from the to-be-verified video, extracting robust content characteristics of the key frame, and performing exclusive-OR operation on the robust content characteristics and the robust zero-watermark signal corresponding to the to-be-verified video to obtain an original watermark, thereby realizing copyright verification of the video. According to the robust video zero-watermark method, the balance problem between robustness and imperceptibility is solved, the desynchronization attack resistance is improved, and the copyright of video media can be effectively protected.

The invention relates to a mass spectrum data missing value filling method and system based on non-negative matrix factorization, and the method comprises the steps: carrying out the pre-filling of missing values of a data set matrix, and obtaining a missing-free initial data matrix; carrying out logarithm transformation on all elements in the initial data matrix without loss; taking a group of dimension parameters of non-negative matrix factorization, and respectively carrying out non-negative matrix factorization to obtain a group of corresponding reconstruction matrixes; performing exponential transformation on the element values of the reconstructed matrix; calculating reconstruction errors between all reconstruction matrixes after exponential transformation and the missing-free initial data matrix; calculating corresponding weights under different reconstruction matrixes according to the reconstruction errors; performing weighted average on the reconstruction matrix to obtain a weighted reconstruction matrix; filling the missing positions in the data set matrix with the element values at the corresponding positions in the weighted reconstruction matrix; and carrying out characteristic metabolite identification and pathway analysis based on the missing-free final data matrix. According to the method, the data filling precision can be improved.

The invention relates to a color image zero-watermarking algorithm based on ternary number decimal order polar complex exponential transformation, which comprises the following steps: a, calculating ternary number decimal order polar complex exponential transformation of an original color image; b, constructing an amplitude sequence through copying and expanding; c, carrying out binarization processing on the amplitude sequence to obtain a binarized amplitude sequence; d, changing the binary amplitude sequence into a binary feature image; e, embedding the LOGO image into the binary feature image by using an XOR operation to obtain a zero-watermark image; and f, finally carrying out zero watermark verification. According to the method, a ternary number theory is combined, ternary number decimal order polar complex exponential transformation is constructed, the relationship among three channels of the color image is utilized, information redundancy is avoided, and the expression and calculation efficiency of the color image is improved. And the constructed ternary number decimal order polar complex exponential transformation is applied to the color image reconstruction and zero watermark algorithm, so that the good stability of the ternary number decimal order polar complex exponential transformation is verified.

The invention discloses an image processing method and device, electronic equipment, a storage medium and a program product, and belongs to the technical field of image processing. The method comprises the following steps: determining an oil light area in a face area from a target image to be processed; performing color deepening processing on the oil gloss area in the target image to obtain a first image; superposing and mixing the oil light area in the first image to obtain a second image; performing power exponential transformation processing on the oil light area in the second image to obtain a third image; and according to the third image, determining a target image from which the oil light is removed. According to the invention, the oily light of the face area can be removed, so thatthe face area looks natural and not greasy.

The invention discloses a hybrid high-order variation ultrasonic image denoising method based on Weibull distribution, and the method comprises the steps of carrying out the noise distribution fittingof an ultrasonic image, carrying out the gray histogram fitting of a selected image local uniform region through employing a Weibull probability density function, and carrying out the estimation of afitting distribution parameter through employing a maximum likelihood method; deducing a likelihood function of the logarithm real noiseless image according to the fitted noise distribution, and performing negative logarithm transformation on the likelihood function to obtain a data fidelity term of the minimized energy function; performing weighted fusion on the first-order regularization expression and the second-order regularization expression of the logarithmic real noiseless image, and constructing a mixed high-order regularization term of a minimized energy function; and enabling the data fidelity term and the mixed high-order regularization term to form a minimized energy function model, using a Spit-Bregman iterative method to perform rapid solution, and after iterative convergence, exponential transformation is utilized to obtain a denoised ultrasonic image. According to the method, speckle noise can be effectively reduced, and details and edge information of the image can bebetter reserved, so that the visual interpretation of the ultrasonic image is enhanced.

The invention discloses an image contour extraction method with high anti-interference capability, and relates to an image contour extraction method. The method comprises a chain code tracking storagecontour information algorithm, and further comprises the following steps: S1, converting a color image into a gray level image by utilizing gray level conversion, and performing nonlinear conversion;S2, performing morphological processing on the image, and performing expansion and corrosion in sequence; S3, performing smoothing processing on the image; and S4, binarizing the image, and extracting the image contour in combination with the chain code tracking storage contour information algorithm. Exponential transformation is adopted in the nonlinear transformation. The method can effectivelyimprove the contour detection precision.

The invention discloses an HDR image tone mapping method based on multi-scale morphology. The method comprises the following steps: S1, inputting an HDR image to be compressed; S2, constructing a logarithm image of the HDR image to be compressed; S3, performing dynamic range compression multi-scale decomposition on the logarithm image to obtain a first Gaussian pyramid and a first Laplace pyramid;S4, extracting an enhanced detail layer of the first Gaussian pyramid by adopting a one-dimensional boundary polishing operator, and adding the detail layer of the first Gaussian pyramid to a layer corresponding to the first Laplacian pyramid layer by layer to obtain a second Laplacian pyramid; S5, performing dynamic range compression reconstruction on the second Laplace pyramid to obtain a detail enhanced image after dynamic range compression; S6, performing exponential transformation on the detail-enhanced image to obtain a first LDR image; and S7, performing color correction on the first LDR image to obtain a to-be-output LDR image. Effective compression of an HDR image dynamic range can be realized, and certain real-time application requirements can be met.

The invention discloses an illumination normalization method for processing human face images, which is specifically implemented according to the following steps: read in the human face image to be processed and take logarithm of it; calculate the shadow layer image; calculate the reflective layer image and perform exponential transformation ; select the sample image and calculate its histogram; normalize the reflective layer image by using the histogram matching method to obtain the corrected image of the face image to be processed by the illumination normalization method. The feature information on the large-scale layer can be preserved by using the edge-preserving filter; the sample image in the visually optimal area is used, so that the mean and variance of the corrected reflection layer image can be in the visually optimal area, which is beneficial to the subsequent face The recognition system can improve the correct rate of face recognition, and solve the problem that the existing feature extraction method of invariant illumination causes a large loss of feature information during face recognition, which is not conducive to the subsequent face recognition.

The invention discloses a synthetic image tampering detection method based on discrete polarity complex exponential transformation. The synthetic image tampering detection method comprises the following steps: constructing a polarity complex exponential transformation framework with a rotation invariant moment; carrying out discretization under the framework of polarity composite exponential transformation with the rotation invariant moment, and conducting the discrete rotation invariant moment of the polarity composite exponential transformation; converting the detected image into a grayscaleimage, and defining the detected image in a discrete spatial domain; constructing a 9 * 9 pixel template, and realizing approximate mapping of the discrete rotation invariant moment of polarity composite exponential transformation from a polar coordinate space to a Cartesian space in the pixel template; extracting features of the detected image by using a discrete rotation invariant moment of polarity complex exponential transformation to obtain effective image features; and obtaining matching feature pairs through consistency sensitive hash operation, and carrying out image matting display.The synthetic image tampering detection method aims at copying-pasting tampered images of translation and rotation deformation, is higher in the detection success rate, and eliminates background interference.

The invention discloses a noise suppression method used for a synthetic aperture radar (SAR) image of a city region. The method comprises the following steps of estimating Alpha-Stable distribution parameters for sample data in sliding windows on real part / imaginary part data of the SAR image, and recording the parameters of all the sliding windows; performing logarithmic transformation on amplitude data of the SAR image, defining sliding windows corresponding to a real part / an imagery part on the image subjected to the logarithmic transformation, and calculating an Alpha-Stable distribution expression and a noise distribution expression in each corresponding window; calculating a posterior probability of a product of the Alpha-Stable distribution expression and the noise distribution expression, and taking the value of a maximum posterior probability in a dynamic range as the value of a central pixel of the sliding window; and performing exponential transformation on the values of the central pixels of all the sliding windows subjected to maximum posterior estimation, thereby obtaining a filtered amplitude image. According to the technical scheme, speckle noises of the SAR image of the city region are suppressed more effectively, so that the quality of the SAR image is improved.