A method for fog detection imaging of active polarization fog removal and noise removal processing cooperation

Abstract

The application provides a fog detection imaging method of active polarization fog removal and noise removal processing cooperation, laser active illumination is used to increase target surface illumination, a space gating imaging mechanism of distance gating is used to shield scattered light from a non-gating area, and the signal-to-noise ratio of target detection in a scattering medium scene is improved; for the superimposed signal of target light and scattered light collected from the gating area, a scattering imaging model based on active detection is established, a polarization fog removal algorithm under active detection is developed based on the polarization difference between the target light and the backscattering light, and the influence of the backscattering light is removed; based on the additive noise property of the forward scattering light on the image, the BM3D algorithm is used to remove the influence of the additive noise such as the forward scattering light to optimize the image, and clear imaging of the target in the thick fog scene is realized; the application uses the fusion means of the active detection mode and the polarization fog removal algorithm, and can realize the fog-transparent detection imaging with the consideration of resolution and speed in the thick fog scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of imaging systems and image processing technology, and in particular to a fog detection imaging method with active polarization fog removal and noise removal processing cooperation. BACKGROUND

[0002] In modern imaging detection, scattering medium environments such as clouds, smoke, fog, haze and water vapor seriously affect the imaging performance of optical equipment and reduce its target detection and recognition capability. Traditional optical detection and imaging systems are mostly designed and imaged in a fog-free scene. When there is a scattering medium in the imaging light path, the scattering medium and the target signal are mixed and received by the detection system, which reduces the imaging quality and even makes the detection completely invalid. Therefore, in practical applications, there is a great demand for imaging through scattering media.

[0003] At present, the existing imaging methods in the scattering medium environment mainly include two categories: passive image processing algorithm enhancement and active light illumination to improve the signal-to-noise ratio. Among them, the passive image processing algorithm is based on the pictures taken by ordinary imaging systems, and uses signal processing to improve the image clarity and other characteristics, which can only work for some weak scattering or specific scenes. In thick fog conditions, most of the photons from the target are scattered by the medium particles before reaching the detector, resulting in the target signal being completely submerged in noise, and the extremely low image signal-to-noise ratio cannot meet the basic image quality requirements that can be processed by passive methods. Active fog imaging technology has its own light source, which can rely on the light source and the detection end hardware system to realize scattering interference suppression and effective target signal enhancement, and is considered to be a potential target imaging method in complex environments and severe weather conditions.

[0004] In prior art (Sun Jianfeng et al. patent, application number CN202110607936.0, application date June 1, 2021 "Single-photon laser fog penetration method based on double-quantity estimation method"), a depth image detection and imaging method based on active single-photon detection and target and background signal separation algorithm is proposed, which improves the estimation accuracy of scattering medium signal and the extraction ability of target signal, and can extract weak target signal in high scattering environment. However, the fog penetration imaging detection based on this method needs to be scanned point by point, and the imaging resolution and imaging speed cannot be compatible, and environmental light and other noise have a great impact on the image detection quality.

[0005] Therefore, it is necessary to develop a new active detection imaging method to meet the ZCha detection requirements in thick fog scenes. SUMMARY

[0006] The present application proposes a fog detection imaging method with active polarization fog removal and noise removal processing cooperation, which uses active detection method and polarization fog removal algorithm fusion means to realize fog penetration detection imaging with consideration of resolution and speed in thick fog scenes.

[0007] The application adopts the following technical solutions.

[0008] A fog detection imaging method of active polarization fog removal and noise removal processing cooperation, the method adopts laser active illumination to increase the target surface illumination, uses the distance gating spatial gating imaging mechanism to shield the scattered light from the non-gating area, and improves the signal-to-noise ratio of the target detection in the scattering medium scene; for the collected target light and scattered light superposition signal from the gating area, a scattering imaging model based on active detection is established, a polarization fog removal algorithm under active detection is developed based on the polarization difference of the target light and the backscattering light, and the influence of the backscattering light is removed; based on the additive noise property of the forward scattering light on the image, the BM3D algorithm is used to remove the influence of the additive noise such as forward scattering light to optimize the image, realize the clear imaging of the target in the thick fog scene, and the method comprises the following steps.

[0009] Step S1: build a distance-gated active detection imaging system;

[0010] Step S2: determine the gating detection delay time required by the target to be measured;

[0011] Step S3: use a polarizer and an analyzer to collect target images of different polarization directions;

[0012] Step S4: based on the developed active detection mechanism scattering medium imaging model, the collected target image is subjected to polarization fog removal processing;

[0013] Step S5: based on the improved BM3D algorithm, the image after fog removal processing is subjected to denoising optimization processing.

[0014] The distance-gated active detection imaging device in step S1 comprises a laser emitting module, a synchronous control module and a detection collection module;

[0015] The laser emitting module in the distance-gated active detection imaging device adopts a high-repetition-frequency pulsed laser as a light source, a polarizer is placed in front of the light source to ensure that completely linearly polarized light is emitted, and the emitted laser is divided into trigger light and illumination light by a prism; the illumination light is homogenized by a homogenizing element and the beam divergence angle is adjusted by a beam expander element, which is used to provide uniform illumination for the target in the far scene;

[0016] The synchronous control module in the distance-gated active detection imaging device adopts a light-triggered synchronous control mode, the trigger light for light triggering is converted into an electrical signal by an avalanche photodiode (APD) and transmitted to the detection collection module, which is used to realize the time synchronization between the laser light emission and the detector;

[0017] The detection collection module in the range-gated active detection imaging device adopts an intensified CCD camera (ICCD), a laser same-wavelength filter is placed in front of an ICCD camera lens to filter out influences of other wavebands on imaging, and a linear polarization analyzer is configured to collect images in four polarization directions of 0°, 45°, 90° and 135°.

[0018] The working process of the range-gated imaging system is as follows: a high-repetition-frequency pulsed laser beam becomes completely polarized light through a polarizer, is divided into trigger light and illumination light through a prism, the trigger light is irradiated on an APD to generate an electrical signal, and the electrical signal is transmitted to an ICCD to synchronize the time of the ICCD with the laser light emission; the illumination light is transmitted to a target after passing through a light homogenizing element and a beam expanding element; the gating of the ICCD is started at the moment of synchronous laser emission, and is opened and closed according to the set delay and opening time length, so that the echo is selectively collected.

[0019] In step S3, the range-gated active imaging device is used to perform gated imaging on the target in the fog scene, the influence of the scattered light of the water mist and the stray light of the ambient light in the non-gated area is shielded, and the signal-to-noise ratio of the echo is improved.

[0020] In step S4, a scattering medium imaging model based on active illumination and gated imaging mechanism is established to deduce the echo intensities of the forward scattering light, the backward scattering light and the target light in detail, and the total light intensity collected by the detection camera is obtained as follows:

[0021] I = I1 + I2 + I3 = (A + C) · t + B · (1 - δ · t) Formula 1

[0022] In the formula, I represents the total intensity collected by the detection system, I1, I2 and I3 respectively represent the target light intensity, the forward scattering light intensity and the backward scattering light intensity; A, B and C are parameters introduced for deducing to describe the target light intensity, the forward scattering light intensity and the backward scattering light intensity, and are related to the active laser and the detection system; t represents the light transmittance of the water mist between the detection system and the target; and δ represents an adjusting parameter introduced due to the gated imaging.

[0023] The polarization properties of the forward scattering light, the backward scattering light and the target light in the water mist scene are as follows: the active polarized light beam used for illumination forms backward scattering light in the water mist scene, and the backward scattering light still has obvious polarization characteristics, while the polarization characteristics of the forward scattering light and the reflection light of the non-specular target are seriously degraded, and basically have no polarization information.

[0024] In step S4, based on the polarization difference between the backward scattering light and the forward scattering light and the target light, a Stokes polarization defogging method based on the active detection scattering imaging model is formed.

[0025] The removal calculation formula of the backward scattering light is derived from Formula 1 as follows:

[0026]

[0027] wherein E is the target signal intensity without backscattering light, the backscattering light intensity I2 varies with the pixel; the backscattering light intensity parameter B represents the backscattering light intensity collected by the system when the target is at infinity; δ is a parameter related to the distance between the gated imaging area and the detection system, which is a constant and can be ignored in actual image calculation and reconstruction.

[0028] The method comprises the following steps:

[0029] Step A1: acquiring the distance gated images with polarization directions of 0°, 45°, 90° and 135° under active laser illumination;

[0030] Step A2: calculating the Stokes parameters and the polarization angle of the acquired images, the formula is:

[0031]

[0032] S1=I(0°)-I(90°) Formula 4;

[0033] S2=I(45°)-I(135°) Formula 5;

[0034]

[0035] In the formula, S0 represents the total light intensity of the echo beam, S1 represents the difference between the horizontal polarization light intensity and the vertical polarization light intensity, S2 represents the difference between the 45° polarization light intensity and the 135° polarization light intensity, and θ represents the polarization angle;

[0036] Step A3: statistically processing the calculated polarization angle, selecting the polarization angle with the highest occurrence probability as the global polarization angle, and taking the global polarization angle as the polarization angle of the backscattering light;

[0037] Step A4: estimating the backscattering light intensity based on the global polarization angle, the formula is:

[0038]

[0039]

[0040] wherein p z is the degree of polarization calculated by using the global polarization angle, and ξ is an introduced parameter, which is used to avoid the excessive error caused by the estimated intensity and the acquired intensity being too close to each other.

[0041] Step A5: selecting the maximum light intensity value in the target-free area of the image, which is used to estimate the backscattering light intensity B collected by the system when the target is at infinity;

[0042] Step A6: Substitute into formula 2 to obtain the image after removing backscattering light.

[0043] In step S5, the polar fog-removed image is optimized and denoised by a three-dimensional block matching algorithm BM3D to remove the influence of forward scattering light, dark noise and other additive Gaussian noise, and to improve the image quality.

[0044] The optimized denoising processing based on the three-dimensional block matching algorithm includes the following steps:

[0045] Step B1: In the obtained image, set a fixed size reference block with a certain pixel step, find the matching block with a difference below the set upper limit in the image, set the upper limit as a multiple of noise variance, 3 times or 5 times noise variance, measure the similarity degree by Euclidean distance, select 8-12 matching blocks with higher similarity degree, and stack the reference block and the matching block into a three-dimensional matrix in any order;

[0046] Step B2: Perform three-dimensional collaborative transformation and filtering on the obtained three-dimensional matrix; use three-dimensional discrete cosine transformation 3DDCT to perform transformation and filtering, and calculate the transformed coefficient matrix C of the input three-dimensional matrix by the following formula:

[0047]

[0048]

[0049] Formula 9;

[0050] Wherein, C(k1, k1, k1) represents an element in the transformed coefficient matrix, N1, N2, N3 represent the size of the three dimensions of the input three-dimensional matrix;

[0051] Step B3: Perform hard threshold filtering on the transformed three-dimensional matrix, if the absolute value of the transformed coefficient exceeds the threshold, the value of the coefficient is kept unchanged; if the absolute value of the coefficient is less than or equal to the threshold, the value of the coefficient is set to 0; through hard threshold filtering, the smaller amplitude noise component in the coefficient matrix is removed, and the more significant signal features are retained;

[0052] Step B4: Perform inverse transformation on the processed three-dimensional matrix to obtain a denoised two-dimensional image block; the formula is:

[0053] X(i, j) = DCT -1 T(Ct)(i, j) Formula 10;

[0054] Wherein, X(i, j) represents the pixel value in the denoised two-dimensional image block, Ct represents the three-dimensional coefficient matrix after hard threshold filtering, DCT -1denotes inverse discrete cosine transform, T denotes three-dimensional collaborative transform. The formula indicates that the denoised two-dimensional image block is restored by performing inverse transform on the three-dimensional coefficient matrix filtered by the hard threshold value; the inverse transform includes the inverse discrete cosine transform;

[0055] Step B5: the obtained denoised image blocks are aggregated to the original positions respectively, and the intensity of each pixel is obtained by weighted average of the coefficients of the corresponding two-dimensional block, to obtain a basic estimation image; assuming that the basic estimation image is M, the denoised image block at the corresponding position is J, and the intensity of each pixel is obtained by weighted average of the coefficients of the corresponding two-dimensional block;

[0056]

[0057] wherein M(x, y) denotes the intensity of the pixel at the coordinate (x, y) in the basic estimation image; J(x, y) denotes the intensity of the pixel at the coordinate (x, y) in the denoised image block; C(x, y) denotes the coefficient of the corresponding two-dimensional block; Z(x, y) denotes a normalization factor, used to ensure that the pixel value after the weighted average is within a reasonable range; and it is the sum of the coefficients of the corresponding two-dimensional block;

[0058] Step B6: based on the basic estimation image, reference blocks are selected on the image at the same step length, and matching blocks are searched below the lower difference upper limit, the reference blocks and the matching blocks are combined to form a basic estimation image block three-dimensional matrix, and the same reference blocks and matching blocks are combined to form a noisy original image block three-dimensional matrix in the noisy original image;

[0059] Step B7: three-dimensional collaborative transform and filtering are performed on the two three-dimensional matrices;

[0060] Step B8: according to the power spectrum of the transform coefficients of the basic estimation image block three-dimensional matrix and the noise intensity, the transform result of the noisy original image block three-dimensional matrix is scaled by the coefficients of the Wiener filter; for the transform coefficient matrix Cn(k1, k2, k3) of the noisy original image block, wherein k1, k2, k3 respectively denote three-dimensional indexes of the transform domain, the coefficient scaling formula of the Wiener filter is:

[0061] Cw(k1, k2, k3) = Cn(k1, k2, k3)(Pw(k1, k2, k3) / (Pw(k1, k2, k3)+σ 2 / Pn(k1, k2, k3))

[0062] Formula 12;

[0063] wherein Cw(k1, k2, k3) denotes the transform coefficient after the Wiener filter; Pw(k1, k2, k3) denotes the power spectrum of the transform coefficient matrix of the basic estimation image block; Pn(k1, k2, k3) denotes the noise power spectrum, which can be obtained by estimating the variance of the noise; σ2 Represents the variance of the noise;

[0064] Step B9: Inversely transform the 3D matrix of the processed noisy original image patch to obtain a 2D image patch, and aggregate the 2D image patches according to their corresponding positions using a weighted average method; for each pixel position, combine the pixel values ​​of multiple 2D image patches using a weighted average method; the weighted average uses a fixed weight or is dynamically adjusted according to the similarity of the image patches to obtain the optimized denoised image.

[0065] Compared with the prior art, the present invention has the following advantages:

[0066] 1. Range-gated imaging technology is used to detect targets in dense fog scenarios. By shielding the scattered light from other non-gated areas, the signal-to-noise ratio is greatly improved, overcoming the problem that existing passive defogging imaging methods cannot be applied to dense fog scenarios. Based on the focal plane array, single-exposure imaging breaks away from the limitations of existing active technology point-by-point scanning detection in terms of imaging resolution and imaging speed, and develops an effective and reliable photoelectric detection method for scenarios such as dense fog, clouds, and ZC smoke.

[0067] 2. This invention combines active detection imaging technology with image enhancement algorithms, develops an active polarization dehazing algorithm based on a range-gated detection mechanism, and superimposes a denoising optimization algorithm based on the representation of scattered light in the image. It can achieve clear imaging of targets in high-concentration scattering media with an optical thickness of up to 5.87, and its effect is better than most existing active and passive dehazing detection methods. Attached Figure Description

[0068] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0069] Appendix Figure 1 This is a schematic diagram of the distance gating imaging system provided in an embodiment of the present invention;

[0070] Appendix Figure 2 This is a schematic diagram of the target imaging method in dense fog scene based on the fusion of range-gated active detection and defogging algorithm provided in an embodiment of the present invention;

[0071] Appendix Figure 3 This is a schematic diagram of target images acquired by a distance gating system at four polarization directions: 0°, 45°, 90°, and 135°, according to an embodiment of the present invention.

[0072] Appendix Figure 4 This is a schematic diagram of the target image after polarization dehazing processing provided in an embodiment of the present invention;

[0073] Appendix Figure 5 This is a schematic diagram of the denoised and optimized target image provided in an embodiment of the present invention;

[0074] Figure 2 is a schematic diagram of imaging effects in different concentration water mist scenarios provided by embodiments of the present application; Figure 6 Figure 3 is a schematic diagram of imaging effects in different concentration water mist scenarios provided by embodiments of the present application;

[0075] Figure 4 is a schematic diagram of imaging effects in different concentration water mist scenarios provided by embodiments of the present application; Figure 7 Figure 5 is a schematic diagram of imaging effects in different concentration water mist scenarios provided by embodiments of the present application. DETAILED DESCRIPTION

[0076] As shown in the figure, an active polarization dehazing and de-noising processing cooperative dehazing detection imaging method, the method uses laser active illumination to increase target surface illumination, uses distance gated spatial gating imaging mechanism to shield scattered light from non-gated regions, and improves the signal-to-noise ratio of target detection in a scattering medium scene; for the superimposed signal of target light and scattered light collected from the gated region, a scattering imaging model based on active detection is established, a polarization dehazing algorithm under active detection is developed based on the polarization difference between target light and backscattered light, and the influence of backscattered light is removed; based on the additive noise properties of forward scattered light shown in the image, the BM3D algorithm is used to remove the influence of additive noise such as forward scattered light to optimize the image, and clear imaging of the target in the thick fog scene is realized, including the following steps.

[0077] Step S1: build a distance-gated active detection imaging system;

[0078] Step S2: determine the gated detection delay time required for the target to be measured;

[0079] Step S3: use a polarizer and an analyzer to collect target images in different polarization directions;

[0080] Step S4: based on the developed active detection mechanism scattering medium imaging model, perform polarization dehazing processing on the collected target image;

[0081] Step S5: based on the improved BM3D algorithm, perform de-noising optimization processing on the image after dehazing processing.

[0082] The distance-gated active detection imaging device in step S1 includes a laser emission module, a synchronous control module, and a detection collection module;

[0083] The laser emission module in the distance-gated active detection imaging device uses a high-repetition-rate pulsed laser as a light source, a polarizer is placed in front of the light source to ensure that completely linearly polarized light is emitted, and the emitted laser is divided into trigger light and illumination light by a prism; the illumination light is homogenized by a homogenizing element and the beam divergence angle is adjusted by a beam expander element, which is used to provide uniform illumination for targets in a distant scene;

[0084] The synchronization control module in the range-gated active detection imaging device adopts a light-triggered synchronization control mode, and is used for converting trigger light into an electrical signal through an avalanche photodiode (APD) and transmitting the electrical signal to a detection collection module, so as to realize time synchronization between laser light emission and a detector.

[0085] The detection collection module in the range-gated active detection imaging device adopts an intensified charge-coupled device (ICCD) camera, a laser same-wavelength filter is placed in front of an ICCD camera lens to filter out influences of other wavebands on imaging, and a linear polarization analyzer is configured to collect images in four polarization directions of 0°, 45°, 90° and 135°.

[0086] The working process of the range-gated imaging system is as follows: a high-repetition-frequency pulsed laser beam is converted into completely polarized light through a polarizer, is divided into trigger light and illumination light through a prism, the trigger light is irradiated on an APD to generate an electrical signal, the electrical signal is transmitted to an ICCD to synchronize the time of the ICCD with laser light emission; the illumination light is transmitted to a target after passing through a light homogenizing element and a beam expanding element; and a gate of the ICCD is opened and closed according to a set delay and opening time length with the laser light emission time as a starting point, so that selective collection of a return wave is realized.

[0087] In step S3, a range-gated active imaging device is used to perform gated imaging on a target in a fog scene, so that the signal-to-noise ratio of a return wave is improved by shielding water mist scattering light and environmental light in a non-gated area.

[0088] In step S4, a scattering medium imaging model based on active illumination and gated imaging is established to deduce return intensities of three light beams of target light, forward scattering light and backward scattering light, and total light intensity collected by a detection camera is obtained as follows:

[0089] I = I1 + I2 + I3 = (A + C) · t + B · (1 - δ · t) Formula 1

[0090] In the formula, I represents total intensity collected by a detection system, I1, I2 and I3 respectively represent target light intensity, forward scattering light intensity and backward scattering light intensity; A, B and C are parameters introduced for deducing target light, forward scattering light and backward scattering light intensities, and are related to active laser and a detection system; t represents light transmittance of water mist between a detection system and a target; and δ represents an adjusting parameter introduced due to gated imaging.

[0091] Polarization properties of forward scattering light, backward scattering light and target light in a water mist scene are as follows: active polarized light beams used for illumination form backward scattering light in a water mist scene, and the backward scattering light still has obvious polarization characteristics, while polarization characteristics of formed forward scattering light and reflection light of a non-specular target are seriously degraded and basically have no polarization information.

[0092] In step S4, based on the polarization difference of backscattering light, forward scattering light and target light, a Stokes polarization defogging method based on an active detection scattering imaging model is formed.

[0093] The removal calculation formula of backscattering light derived from formula 1 is:

[0094]

[0095] Wherein E is the target signal intensity without backscattering light, the backscattering light intensity I2 changes with the pixel; the backscattering light intensity parameter B represents the backscattering light intensity collected by the system when the target is at infinity; δ is a parameter related to the distance between the gated imaging area and the detection system, which is a constant and can be ignored in actual image calculation and reconstruction.

[0096] The method comprises the following steps:

[0097] Step A1: acquiring the distance gated images with polarization directions of 0°, 45°, 90° and 135° under active laser illumination;

[0098] Step A2: calculating the Stokes parameters and polarization angle of the acquired images, the formula is:

[0099]

[0100] S1=I(0°)-I(90°) Formula 4;

[0101] S2=I(45°)-I(135°) Formula 5;

[0102]

[0103] In the formula, S0 represents the total light intensity of the echo beam, S1 represents the difference between the horizontal polarization light intensity and the vertical polarization light intensity, S2 represents the difference between the 45° polarization light intensity and the 135° polarization light intensity, and θ represents the polarization angle.

[0104] Step A3: statistically analyzing the calculated polarization angle, selecting the polarization angle with the highest occurrence probability as the global polarization angle, and taking it as the polarization angle of the backscattering light;

[0105] Step A4: estimating the backscattering light intensity based on the global polarization angle, the formula is:

[0106]

[0107]

[0108] Wherein, p zis the polarization degree calculated by global polarization angle, and ξ is a parameter introduced to avoid the large error caused by the too close estimated intensity and collected intensity of individual pixels.

[0109] Step A5: Select the maximum light intensity value in the non-target region of the image to estimate the backscattering light intensity B collected by the system when the target is at infinity;

[0110] Step A6: Substitute into formula 2 to obtain the image after removing backscattering light.

[0111] In step S5, the polarized fog-removed image is optimized and denoised by the three-dimensional block matching algorithm BM3D to remove the influence of various additive Gaussian noises such as forward scattering light and dark noise, and to improve the image quality.

[0112] The optimized denoising processing based on the three-dimensional block matching algorithm includes the following steps:

[0113] Step B1: Set a fixed size reference block in the obtained image at a certain pixel step, search for a matching block with a difference below the set upper limit in each reference block, set the upper limit as a multiple of the noise variance, 3 times or 5 times the noise variance, measure the similarity degree by the Euclidean distance, select 8-12 matching blocks with higher similarity degree, and stack the reference block and the matching block into a three-dimensional matrix in any order;

[0114] Step B2: Perform three-dimensional collaborative transformation and filtering on the obtained three-dimensional matrix; use three-dimensional discrete cosine transformation 3DDCT to perform transformation and filtering, and calculate the transformed coefficient matrix C of the input three-dimensional matrix by the following formula:

[0115]

[0116] Formula 9;

[0117] Wherein, C(k1, k1, k1) represents an element in the transformed coefficient matrix, N1, N2, N3 represent the size of the three dimensions of the input three-dimensional matrix;

[0118] Step B3: Perform hard threshold filtering on the transformed three-dimensional matrix, if the absolute value of the transformed coefficient exceeds the threshold, the value of the coefficient is kept unchanged; if the absolute value of the coefficient is less than or equal to the threshold, the value of the coefficient is set to 0; through hard threshold filtering, the smaller amplitude noise components in the coefficient matrix are removed, and the more significant signal features are retained;

[0119] Step B4: Perform inverse transformation on the processed three-dimensional matrix to obtain the denoised two-dimensional image block; the formula is:

[0120] X(i, j) = DCT -1 T(Ct)(i, j) Formula 10;

[0121] wherein X(i,j) represents the pixel value in the denoised two-dimensional image block, Ct represents the three-dimensional coefficient matrix after hard threshold filtering, DCT -1 represents inverse discrete cosine transform, and T represents three-dimensional collaborative transform. The formula indicates that the denoised two-dimensional image block is restored by performing inverse transform on the three-dimensional coefficient matrix after hard threshold filtering; the inverse transform includes inverse discrete cosine transform;

[0122] Step B5: the obtained denoised image blocks are aggregated to the original positions respectively, and the intensity of each pixel is obtained by weighted average of the coefficients of the two-dimensional blocks in the corresponding positions to obtain a basic estimation image; assuming that the basic estimation image is M, the denoised image block in the corresponding position is J, and the intensity of each pixel is obtained by weighted average of the coefficients of the two-dimensional blocks in the corresponding positions;

[0123]

[0124] wherein M(x,y) represents the pixel intensity at the coordinate (x,y) in the basic estimation image; J(x,y) represents the pixel intensity at the coordinate (x,y) in the denoised image block; C(x,y) represents the coefficient of the two-dimensional block in the corresponding position; Z(x,y) represents a normalization factor for ensuring that the pixel value after weighted average is within a reasonable range; and it is the sum of the coefficients of the two-dimensional block in the corresponding position;

[0125] Step B6: based on the basic estimation image, reference blocks are selected on the image at the same step length, and matching blocks are searched below the lower difference upper limit to form a basic estimation block three-dimensional matrix with the reference blocks and the matching blocks, and a noisy original image block three-dimensional matrix is formed with the same reference blocks and matching blocks in the noisy original image;

[0126] Step B7: three-dimensional collaborative transform and filtering are performed on the two three-dimensional matrices;

[0127] Step B8: according to the power spectrum of the transform coefficients of the basic estimation block three-dimensional matrix and the noise intensity, the transform results of the noisy original image block three-dimensional matrix are scaled in coefficient by using Wiener filtering; for the transform coefficient matrix Cn(k1,k2,k3) of the noisy original block, wherein k1, k2 and k3 respectively represent three-dimensional indexes of the transform domain, the coefficient scaling formula of the Wiener filtering is:

[0128] Cw(k1,k2,k3) = Cn(k1,k2,k3)(Pw(k1,k2,k3) / (Pw(k1,k2,k3)+σ 2 / Pn(k1,k2,k3))

[0129] Formula 12;

[0130] Wherein, Cw(k1, k2, k3) represents the transform coefficient after Wiener filtering; Pw(k1, k2, k3) represents the power spectrum of the transform coefficient matrix of the basic estimation block; Pn(k1, k2, k3) represents the noise power spectrum, which can be obtained by estimating the variance of the noise; σ 2 represents the variance of the noise;

[0131] Step B9: inverse transform the processed noisy original block three-dimensional matrix to obtain a two-dimensional image block, and aggregate the two-dimensional image blocks according to their corresponding positions by using a weighted average method; for each pixel position, the pixel values of multiple two-dimensional image blocks are combined by using a weighted average method; the weighted average uses a fixed weight or is dynamically adjusted according to the similarity of the image blocks to obtain an image after optimized denoising.

[0132] Embodiment

[0133] The active polarization defogging and denoising processing cooperative defogging detection imaging method provided by the application has the basic structure of a range-gated imaging system as shown in the figure. Figure 1 A 532nm pulse laser with a repetition frequency of 2KHz is selected as the active illumination light source, and the outgoing light beam is split by a prism according to an energy ratio of 1:9. The small-energy light beam is irradiated on an avalanche photodiode to convert an electrical signal to trigger an enhanced CCD camera. The other light beam is an illumination light beam, which is homogenized by a DOE diffraction sheet and adjusted by a beam expander to ensure that the illumination light can uniformly illuminate the entire target.

[0134] In this embodiment, the pulse energy of the pulse laser used is 5uJ; the sensor resolution of the enhanced CCD camera is 1024x1024, the gating time is 5ns, the delay time control range is 0ns-10s, and the lens focal length used is 100mm. The rising edge of the APD electrical signal is used as the trigger to start the timing of the delay time.

[0135] In this embodiment, a fog tank with a size of 1.5m x 1.5m x 2.3m is built indoors, and a fog generator is used to create a uniform and controllable water mist environment in a settling mode. A 532nm continuous laser is configured in the fog tank as a monitoring light path to measure the water mist concentration at different times. A checkerboard target plate, a car model and a toy are arranged in the fog tank as targets for detection.

[0136] Reference Figure 2 The embodiment provides a dense fog scene target imaging method based on fusion of active detection and defogging algorithm of range-gated, which comprises the following steps:

[0137] Step S1: build an active illumination range-gated imaging system;

[0138] Step S2: determine the appropriate delay of the target to be measured;

[0139] Step S3: A linear polarizer is placed in front of the ICCD lens, and target images in 0°, 45°, 90° and 135° polarization directions are collected respectively;

[0140] Step S4: Polarization defogging processing is performed on the collected images according to the established scattering imaging model based on active light illumination and gating imaging mechanism;

[0141] Step S5: The image after polarization defogging is further denoised and optimized to realize clear imaging of the target in a dense fog scene.

[0142] In the embodiment, step S2 is to determine the appropriate delay of the target to be measured. The distance of the target is measured by means of a laser range finder, and then the delay time is calculated according to the speed of light. The specific calculation formula is:

[0143]

[0144] In the formula, τ is the delay time of the target to be measured, L represents the distance between the target and the detection system measured, and c represents the speed of light.

[0145] In the embodiment, step S3 is to acquire target images in 0°, 45°, 90° and 135° polarization directions based on the distance-gated imaging system, as shown in FIG. 3. The specific collection steps are as follows: Figure 3

[0146] Step S31: The polarizer is placed at the light outlet of the pulsed laser to ensure that the outgoing light is horizontally polarized light;

[0147] Step S32: The analyzer is placed in front of the lens of the ICCD camera;

[0148] Step S33: The delay time is set according to the calculated delay time, and the gain multiple of the ICCD is set;

[0149] Step S34: The collection mode and camera exposure time are set so that the echo signals of multiple pulses can be collected in one exposure time, and the signal-to-noise ratio is improved;

[0150] Step S35: The polarization direction of the analyzer is adjusted, and target images in four polarization directions are collected.

[0151] In the embodiment, step S4 is to perform polarization defogging processing on the collected images according to the established scattering imaging model based on active light illumination and gating imaging mechanism. Further, the specific method of step S4 is as follows:

[0152] Step S41: Based on the four collected polarization images, the Stokes parameters S0, S1 and S2 are calculated, and the specific formula is as follows:

[0153] ​

[0154] S1 = I(0°) - I(90°)

[0155] S2 = I(45°) - I(135°)

[0156] Step S42: based on the Stokes parameter, calculate the polarization angle of the target image, the formula is:

[0157]

[0158] Step S43: count the frequency distribution of the polarization angle, since the target light is unpolarized and the backscattered light is polarized, the polarization angle with the most occurrences represents the properties of the scattered light, and it is selected as the global polarization angle θ z ;

[0159] Step S44: based on the global polarization angle, estimate the degree of polarization and intensity of the backscattered light, the specific formula is:

[0160]

[0161]

[0162] Wherein, ξ is introduced as a parameter, which is used to avoid excessive error caused by the estimation of intensity and the collection of intensity being too close, generally taking 1.8-2.0.

[0163] Step S45: based on the calculated S0 image, select the gray value with the maximum amplitude in the target-free area to estimate the parameter B;

[0164] Step S46: based on the calculated S0 image, substitute the estimated parameters I2 and B into the formula to calculate the polarized dehazed image, as shown in Figure 4 The specific formula is:

[0165]

[0166] In this embodiment, the step S5 is further denoising and optimization based on the polarized dehazed image, to remove the influence of additive noise such as forward scattering light on image quality. Further, the step S5 specifically comprises:

[0167] Step S51: set a reference block of 8 pixels x 8 pixels in size in the obtained image with a step of 4 pixels, search for a matching block with a difference below the set upper limit in the image for each reference block, the upper limit is set to a multiple of noise variance, 3 times or 5 times noise variance, measure the similarity degree by Euclidean distance, select 8-12 matching blocks with higher similarity degree, and stack the reference block and the matching blocks into a three-dimensional matrix in any order;

[0168] Step S52: Perform 3D collaborative transform and filtering on the obtained three-dimensional matrix; use 3D discrete cosine transform (3D DCT) to perform transform and filtering, and for the input three-dimensional matrix, its transformed coefficient matrix C can be calculated by the following formula

[0169]

[0170] wherein C(k1, k1, k1) represents an element in the transformed coefficient matrix, N1, N2, N3 represent the sizes of the three dimensions of the input three-dimensional matrix.

[0171] Step S53: Perform hard threshold filtering on the transformed three-dimensional matrix, if the absolute value of the transformed coefficient exceeds the threshold, the value of the coefficient is kept unchanged; if the absolute value of the coefficient is less than or equal to the threshold, the value of the coefficient is set to 0. In this way, the hard threshold filtering can remove the smaller amplitude noise components in the coefficient matrix and retain more significant signal features;

[0172] Step S54: Perform inverse transform on the processed three-dimensional matrix to obtain a denoised two-dimensional image block;

[0173] X(i, j) = DCT -1 T(Ct)(i, j)

[0174] wherein X(i, j) represents the pixel value in the denoised two-dimensional image block, Ct represents the three-dimensional coefficient matrix after hard threshold filtering, DCT-1 represents discrete cosine inverse transform, and T represents three-dimensional collaborative transform. The formula indicates that the denoised two-dimensional image block can be restored by performing inverse transform (such as discrete cosine inverse transform) on the three-dimensional coefficient matrix after hard threshold filtering.

[0175] Step S55: Aggregate the obtained denoised image blocks to the original positions respectively, and the intensity of each pixel is obtained by weighted average of the coefficients of the corresponding two-dimensional block to obtain a basic estimation image; assuming that the basic estimation image is M, the denoised image block at the corresponding position is J, and the intensity of each pixel is obtained by weighted average of the coefficients of the corresponding two-dimensional block.

[0176]

[0177] wherein M(x, y) represents the pixel intensity at the coordinate (x, y) in the basic estimation image. J(x, y) represents the pixel intensity at the coordinate (x, y) in the denoised image block. C(x, y) represents the coefficient of the corresponding two-dimensional block. Z(x, y) represents a normalization factor for ensuring that the pixel value after weighted average is within a reasonable range. It is the sum of the coefficients of the corresponding two-dimensional block.

[0178] Step S56: Based on the image of the basic estimation, the reference block is selected on the image with the same step size, and the matching block is searched below the lower difference upper limit to form the basic estimation block three-dimensional matrix, and the same reference block and matching block are used to form the noisy original image block three-dimensional matrix in the noisy original image;

[0179] Step S57: Perform three-dimensional collaborative transformation and filtering on the two three-dimensional matrices;

[0180] Step S58: According to the power spectrum of the transformation coefficient of the basic estimation block three-dimensional matrix and the noise intensity, the transformation result of the noisy original image block three-dimensional matrix is scaled by using the Wiener filter; for the transformation coefficient matrix Cn(k1, k2, k3) of the noisy original image block, wherein k1, k2, k3 represent three dimension indexes of the transformation domain respectively, the coefficient scaling formula of the Wiener filter is:

[0181] Cw(k1, k2, k3) = Cn(k1, k2, k3)(Pw(k1, k2, k3) / (Pw(k1, k2, k3)+σ 2 / Pn(k1, k2, k3)))

[0182] Wherein, Cw(k1, k2, k3) represents the transformation coefficient after the Wiener filter; Pw(k1, k2, k3) represents the power spectrum of the transformation coefficient matrix of the basic estimation block; Pn(k1, k2, k3) represents the noise power spectrum, which can be obtained by estimating the variance of the noise; σ 2 represents the variance of the noise.

[0183] Step S59: The processed noisy original image block three-dimensional matrix is inverse transformed to obtain a two-dimensional image block, and the two-dimensional image block is aggregated according to its corresponding position by using the weighted average method. For each pixel position, the pixel values of multiple two-dimensional image blocks are combined by using the weighted average method. The weighted average can use fixed weight or dynamically adjust according to the similarity of the image blocks, so as to obtain the image after optimization denoising, as shown in Figure 5 .

[0184] In this embodiment, the same target under different water mist concentrations is detected, the water mist concentration of the mist tank is changed by controlling the working time of the mist generator, so that the optical thickness of the water mist scene is 2.12, 2.82, 3.51 and 5.87 respectively, and the method can realize good imaging, as shown in Figure 6 . The optical thickness is obtained by multiplying the water mist extinction coefficient measured by the monitoring light path and the distance of the target in the water mist. The calculation formula of the water mist extinction coefficient is:

[0185]

[0186] where β is the extinction coefficient of the water fog scene, d is the distance of the monitoring laser transmitting in the fog, I' and I are the powers of the laser after and before transmitting through the fog, respectively. c and I c are the powers of the laser after and before transmitting through the fog, respectively.

[0187] In the embodiment, the existing active and passive methods are used to detect the target in the same water fog scene, as shown in FIG. 3, the results show that the traditional passive imaging method is basically invalid in the thick fog scene, and the target image of the existing active imaging method appears serious degradation, and the information loss is serious. Compared with the straight line, the method can still realize good imaging. Figure 7

[0188] The above embodiment shows that the method provided by the application collects the original image through the distance-gated imaging system, further develops the active polarization defogging model based on the polarization difference and imaging characteristics of the target light and the scattered light, removes the forward scattered light by using the BM3D denoising algorithm, and realizes the target detection imaging in the thick fog scene with an optical thickness of 5.87. Compared with the invalidity or serious degradation of the imaging quality of the existing active and passive methods in the thick fog scene, the method provides an effective and reliable detection method, and can meet the ZCha sensing requirements in the complex ZC environment such as ZC smoke screen and the severe weather environment such as thick fog and cloud layer.

[0189] The example relates to an active polarization defogging and denoising processing method based on a distance-gated imaging system. The method comprises the following steps: building a distance-gated active imaging system; determining the delay time required for the to-be-detected target to be gated and imaged; collecting four target images with different polarization directions based on the distance-gated imaging system; performing polarization defogging processing on the collected images based on the distance-gated active imaging defogging model; and performing BM3D-based denoising optimization processing based on the polarization defogging image. The application combines the active detection imaging technology and the image enhancement algorithm, greatly improves the signal-to-noise ratio by shielding the scattered light in other non-gated regions, develops an active polarization defogging algorithm based on the distance-gated detection mechanism, and based on the form of the scattered light in the image, adds a denoising optimization algorithm, so that clear imaging of the target can be realized in a high-concentration scattering medium with an optical thickness of 5.87, the problem that the existing passive defogging imaging method cannot be applied to the thick fog scene is overcome, and the effect is better than that of most existing active and passive defogging detection methods.​

Claims

1. An active polarization defogging and denoising processing cooperative defogging detection imaging method, characterized in that: The method uses laser active illumination to increase the target surface illumination, uses the distance gated spatial gating imaging mechanism to shield the scattered light from the non-gated area, and improves the signal-to-noise ratio of target detection in the scattering medium scene; for the superimposed signal of the target light and the scattered light collected from the gated area, a scattering imaging model based on active detection is established, a polarization defogging algorithm under active detection is developed based on the polarization difference between the target light and the backscattered light, and the influence of the backscattered light is removed; Based on the additive noise property of the forward scattering light on the image, the BM3D algorithm is used to remove the influence of the additive noise of the forward scattering light to optimize the image, and the target clear imaging of the thick fog scene is realized, including the following steps. Step S1: build a distance-gated active detection imaging system; Step S2: determine the gated detection delay time required for the target to be tested; Step S3: use a polarizer and an analyzer to collect target images of different polarization directions; Step S4: based on the developed active detection mechanism scattering medium imaging model, the collected target image is processed by polarization defogging; Step S5: based on the improved BM3D algorithm, the image after defogging processing is executed by denoising optimization processing; the distance-gated active detection imaging device in step S1, comprising a laser emitting module, a synchronous control module and a detection collection module; The laser emitting module in the distance-gated active detection imaging device uses a high-repetition-rate pulsed laser as a light source, a polarizer is placed in front of the light source to ensure that completely linearly polarized light is emitted, and the emitted laser is divided into trigger light and illumination light by a prism; the illumination light is homogenized by a homogenizing element and the beam divergence angle is adjusted by a beam expander element, which is used to provide uniform illumination for the target in the distant scene; The synchronous control module in the distance-gated active detection imaging device uses a light-triggered synchronous control method, which is used to convert the trigger light into an electrical signal by an avalanche photodiode (APD) and transmit it to the detection collection module, which is used to realize the time synchronization between the laser emission and the detector; The detection collection module in the distance-gated active detection imaging device uses an intensified CCD camera (ICCD), a laser wavelength filter is placed in front of the ICCD camera lens to filter out the influence of other wavebands on imaging, and a linear polarization analyzer is configured to collect images of four polarization directions of 0°, 45°, 90° and 135°; The working process of the distance-gated active detection imaging device is as follows: the high-repetition-rate pulsed laser beam is converted into completely polarized light by a polarizer, and is divided into trigger light and illumination light by a prism, the trigger light is irradiated on the APD to generate an electrical signal, which is transmitted to the ICCD to make its time synchronized with the laser emission; after the homogenizing element and the beam expander element, the illumination light is emitted to the target; the gate of the ICCD is opened and closed according to the set delay and opening time with the synchronized laser emission time as the starting point, realizing the selective collection of the echo.

2. The method of claim 1, wherein the method is a method of fog detection imaging with fog removal and noise removal by active polarization, characterized in that: In step S3, the distance-gated active imaging device is used to image the target in the foggy scene, and the signal-to-noise ratio of the echo is improved by shielding the scattered light of the water mist and the stray light of the ambient light in the non-gated area.

3. The method of claim 1, wherein the method is a method of fog detection imaging with fog removal and noise removal processing cooperation by active polarization, characterized in that: In step S4, the echo intensity of the forward scattering light, the backscattering light and the target light is derived in detail by establishing a scattering medium imaging model based on the active illumination and the gating imaging mechanism, and the total light intensity collected by the detection camera is obtained as follows: I = I1 + I2 + I3 = (A + C)t + B·(1-δ·t) Formula 1; wherein I represents the total intensity collected by the detection system, I1, I2 and I3 represent the target light intensity, the forward scattering light intensity and the backscattering light intensity respectively; A, B and C are parameters introduced for derivation and used for describing the target light intensity, the forward scattering light intensity and the backscattering light intensity, which are related to the active laser and the detection system, and are fixed constants when the detection end, the collection end, the water mist and the target are determined; t represents the light transmittance of the water mist between the detection system and the target; and δ represents an adjusting parameter introduced due to the gating imaging.

4. The method of claim 3, wherein the method is a method of fog detection imaging with fog removal and noise removal processing cooperation by active polarization, characterized in that: The polarization properties of the forward scattering light, the backscattering light and the target light in the water mist scene are as follows: the backscattering light formed by the active polarized light beam used for illumination in the water mist scene still maintains obvious polarization characteristics, while the polarization characteristics of the forward scattering light and the reflection light of the non-specular target are seriously degraded and have no polarization information.

5. The method of claim 4, wherein the method is a method of fog detection imaging with fog removal and noise removal by active polarization, characterized in that: In step S4, based on the polarization difference between the backscattering light and the forward scattering light and the target light, a Stokes polarization defogging method based on the active detection scattering imaging model is formed.

6. The method of claim 3, wherein the method is a method of fog detection imaging with fog removal and noise removal processing cooperation by active polarization, characterized in that: The removal calculation formula of the backscattering light derived from Formula 1 is as follows: wherein E is the target signal intensity without the backscattering light, the backscattering light intensity I2 varies with the pixel; the backscattering light intensity parameter B represents the backscattering light intensity collected by the system when the target is at infinity; δ is a parameter related to the distance between the gating imaging area and the detection system, which is a constant and can be ignored in actual image calculation and reconstruction. The method comprises the following steps.

7. The method of claim 3, wherein the method is a method of fog detection imaging with fog removal and noise removal processing cooperation by active polarization, characterized in that: Step A1: acquiring the gating images with the polarization directions of 0°, 45°, 90° and 135° under the active laser illumination; Step A2: calculating the Stokes parameters and the polarization angle of the acquired images, and the formula is as follows: S1 = I(0°)-I(90°) Formula 4; S2 = I(45°)-I(135°) Formula 5; wherein S0 represents the total light intensity of the echo light beam, S1 represents the difference between the horizontal polarization light intensity and the vertical polarization light intensity, S2 represents the difference between the 45° polarization light intensity and the 135° polarization light intensity, and θ represents the polarization angle; Step A3: statistically processing the calculated polarization angle, and selecting the polarization angle with the highest occurrence probability as the global polarization angle as the polarization angle of the backscattering light; Step A4: estimating the backscattering light intensity based on the global polarization angle, and the formula is as follows: Step A5: selecting the maximum light intensity value in the image without the target, which is used to estimate the backscattering light intensity B collected by the system when the target is at infinity; where p z is the polarization degree calculated by the global polarization angle, and ξ is a parameter introduced to avoid the large error caused by the close estimation and acquisition of the intensity of individual pixels. Step A6: substituting the formula 2 to obtain the image after removing the backscattering light. In step S5, the polarized defogging image is optimized and denoised by the three-dimensional block matching algorithm BM3D to remove the influence of various additive Gaussian noises of the forward scattering light and dark noise, and the image quality is improved.

8. The method of claim 3, wherein the method is a method of fog detection imaging with fog removal and noise removal processing cooperation by active polarization, characterized in that: The optimization and denoising processing based on the three-dimensional block matching algorithm comprises the following steps.

9. The method of claim 8, wherein the method is a method of fog detection imaging with fog removal and noise removal by active polarization, characterized in that: ​ Step B1: Set a fixed size reference block in the obtained image with a certain pixel step, search for matching blocks with differences below the set upper limit in the image, set the upper limit as a multiple of the noise variance, 3 times or 5 times the noise variance, measure the similarity degree by using the Euclidean distance, select 8-12 matching blocks with higher similarity degree, and stack the reference block and the matching blocks into a three-dimensional matrix in any order; Step B2: Perform three-dimensional collaborative transformation and filtering on the obtained three-dimensional matrix; use three-dimensional discrete cosine transformation 3D DCT to perform transformation and filtering, and calculate the transformed coefficient matrix C of the input three-dimensional matrix through the following formula Formula 9; Wherein, C(k1, k1, k1) represents an element in the transformed coefficient matrix, N1, N2 and N3 represent the sizes of the three dimensions of the input three-dimensional matrix; Step B3: Perform hard threshold filtering on the transformed three-dimensional matrix, if the absolute value of the transformed coefficient exceeds the threshold, the value of the coefficient is kept unchanged; if the absolute value of the coefficient is less than or equal to the threshold, the value of the coefficient is set to 0; through the hard threshold filtering, the smaller amplitude noise components in the coefficient matrix are removed, and the more significant signal features are retained; Step B4: Perform inverse transformation on the processed three-dimensional matrix to obtain a denoised two-dimensional image block; the formula is: X(i,j) = DCT -1 T(Ct)(i,j) Equation 10; wherein X(i,j) represents a pixel value in the denoised two-dimensional image block, Ct represents a three-dimensional coefficient matrix after hard threshold filtering, DCT -1 represents inverse discrete cosine transform, T represents three-dimensional joint transform; the formula represents that the denoised two-dimensional image block is restored by performing inverse transform on the three-dimensional coefficient matrix after hard threshold filtering; the inverse transform comprises inverse discrete cosine transform; Step B5: Aggregate the obtained denoised image blocks to the original positions respectively, and obtain a basic estimation image by weighted average of the intensity of each pixel through the coefficients of the corresponding two-dimensional block; assuming that the basic estimation image is M, the denoised image block in the corresponding position is J, and the intensity of each pixel is weighted average of the coefficients of the corresponding two-dimensional block; Wherein, M(x, y) represents the pixel intensity at the coordinate (x, y) in the basic estimation image; J(x, y) represents the pixel intensity at the coordinate (x, y) in the denoised image block; C(x, y) represents the coefficient of the corresponding two-dimensional block; Z(x, y) represents a normalization factor, which is used to ensure that the pixel value after weighted average is within a reasonable range; it is the sum of the coefficients of the corresponding two-dimensional block; Step B6: Based on the basic estimation image, select a reference block on the image with the same step, search for a matching block below a lower difference upper limit, and form a basic estimation image block three-dimensional matrix with the reference block and the matching block; form a noisy original image block three-dimensional matrix with the same reference block and matching block in the noisy original image; Step B7: Perform three-dimensional collaborative transformation and filtering on the two three-dimensional matrices; Step B8: According to the power spectrum of the transformed coefficient of the basic estimation image block three-dimensional matrix and the noise intensity, perform coefficient scaling on the transformed result of the noisy original image block three-dimensional matrix by using Wiener filtering; for the transformed coefficient matrix Cn(k1, k2, k3) of the noisy original image block, wherein k1, k2 and k3 represent the three dimension indexes of the transformed domain, the coefficient scaling formula of the Wiener filtering is: Cw(kl, k2, k3) = Cn(kl, k2, k3)(Pw(kl, k2, k3) / (Pw(kl, k2, k3) + σ 2 / P n (kl, k2, k3))) Formula 12; wherein Cw(k1, k2, k3) represents the transform coefficient after Wiener filtering; Pw(k1, k2, k3) represents the power spectrum of the transform coefficient matrix of the basic estimation tile; Pn(k1, k2, k3) represents the noise power spectrum, which can be obtained by estimating the variance of the noise; σ 2 represents the variance of the noise. Step B9: inverse transform the processed noisy original image block three-dimensional matrix to obtain a two-dimensional image block, and aggregate the two-dimensional image blocks according to their corresponding positions by using a weighted average method; for each pixel position, combine the pixel values of multiple two-dimensional image blocks by using a weighted average method; the weighted average uses a fixed weight or is dynamically adjusted according to the similarity of the image blocks to obtain an image after optimized denoising.

Images