A method for removing motion blur from double-frame images based on joint control of sharpness and motion

Abstract

The present application relates to the technical field of digital image processing, and discloses a double-frame image motion blur removal method based on joint control of definition and motion, which acquires two time-adjacent images under the same scene, constructs definition distribution map and motion direction map after pretreatment, and obtains overall definition evaluation value based on direction consistency weighted statistics, so as to determine the image to be restored and the auxiliary reference image; further constructs joint confidence information of definition and motion, performs inter-frame motion estimation with blur perception, and divides adaptive blur area and non-blur area in combination with the definition distribution map and the motion estimation result; the main motion direction and the main motion amplitude are extracted in the adaptive blur area to construct the motion blur kernel, and the joint optimization objective function of the image to be restored and the motion blur kernel is established, the image to be restored and the motion blur kernel are updated through alternating iteration, and the final deblurring result image is output.

Description

Technical Field

[0001] This invention belongs to the field of image processing, and specifically relates to a method for removing motion blur from two-frame images based on sharpness-motion joint control. Background Technology

[0002] In recent years, with the widespread application of mobile terminals, vehicle-mounted equipment, security monitoring equipment, and intelligent vision systems, digital image acquisition technology has developed rapidly. However, in actual imaging processes, due to factors such as camera shake, subject movement, long exposure times, or complex imaging environments, acquired images are prone to motion blur, leading to loss of image details, weakened edge information, and degraded structural features. This, in turn, affects the accuracy and stability of subsequent visual tasks such as target detection, recognition, tracking, and scene understanding. Existing technologies for motion-blurred image recovery typically include single-frame image-based deblurring methods and multi-frame image-based deblurring methods. Single-frame image deblurring methods generally recover clear images through blur kernel estimation, image prior modeling, and optimization solutions. However, because they rely solely on a single blurred observation image, they often suffer from insufficient information, resulting in inaccurate blur kernel estimation and the recovery results being prone to ringing artifacts. In contrast, multi-frame image deblurring methods can utilize complementary information between adjacent frames to aid in restoration, thus improving the restoration quality of motion-blurred images to some extent. However, existing multi-frame deblurring methods still have the following shortcomings in practical applications: First, adjacent frames often have both blur degree differences and motion differences. If the image with a higher blur degree cannot be effectively distinguished from the reference image with a lower blur degree, the reference information may not be fully utilized. Second, existing methods usually lack collaborative constraints on blur direction, local sharpness differences, and directional reliability during inter-frame motion estimation, resulting in insufficient motion estimation accuracy. Third, existing methods often struggle to adaptively restore based on the blur degree and motion consistency of different regions in the image, easily leading to insufficient restoration of blurred areas and over-processing of non-blurred areas, affecting the overall image quality. These problems are particularly pronounced in two-frame scenarios.

[0003] Furthermore, in existing deblurring and restoration processes, the construction and updating of the motion blur kernel are often not closely integrated with local image sharpness information, motion direction information, and regional characteristics, lacking a unified joint control mechanism. This makes it difficult to simultaneously consider the restoration capability within blurred regions and the structure preservation capability within unblurred regions, thus limiting further improvement in the final deblurring effect. Therefore, it is necessary to provide a new method for removing motion blur from two-frame images to at least solve the aforementioned problems in existing technologies. This method can combine the sharpness difference and motion information of two adjacent frames to effectively determine the blurred image to be restored and the auxiliary reference image. Based on this, it can achieve more accurate inter-frame motion estimation, adaptive blurred region segmentation, and optimized updating of the motion blur kernel, thereby improving the restoration quality of motion-blurred images. Summary of the Invention

[0004] This invention proposes a method for removing motion blur from two-frame images based on joint sharpness and motion control. By introducing sharpness distribution map, motion direction map, and joint sharpness and motion confidence information, and combining inter-frame motion estimation with blur perception, adaptive blur region segmentation, motion blur kernel construction, and joint optimization iterative update, the method achieves effective recovery of motion-blurred images, improves the deblurring capability of blurred regions, and maintains the structural stability and detail authenticity of unblurred regions.

[0005] This invention provides a method for removing motion blur from two-frame images based on sharpness-motion joint control, comprising the following steps:

[0006] S1. Acquire two adjacent frames of images and preprocess them to obtain the first frame image. Second frame image ;

[0007] S2, respectively for and Perform local sharpness analysis and construct a sharpness distribution map. and Generate motion direction map based on image gradient information. Based on motion direction map right and A sharpness evaluation value is obtained by weighting the values ​​based on directional consistency. and ;

[0008] S3, according to and The size determines the blurred image to be recovered and auxiliary reference images Based on the resolution distribution map and motion direction map Constructing clarity motion joint confidence information ;

[0009] S4, according to and Establish inter-frame motion estimation and obtain initial motion estimation results. Motion Direction Diagram As a direction prior for inter-frame motion estimation and Used for The inter-frame motion estimation results are obtained by performing confidence correction. ;

[0010] S5. Define the image domain Perform joint judgment to delineate adaptive fuzzy regions. Unfuzzy regions ; and Based on the inter-frame motion estimation results Extract the main motion direction and main motion amplitude Constructing motion fuzzy kernel ;

[0011] S6. Create a clear image to be recovered. With motion fuzz kernel Joint optimization objective function ;

[0012] S7. Regarding the objective function Solve the problem to obtain the clear image to be restored. Initial deblurring result ;

[0013] S8, according to In the adaptive fuzzy region Internal main motion direction Main motion amplitude and United Credit Information Update motion blur kernel and to and Perform alternating iterative optimization until the convergence condition is met, and output the deblurred result image. .

[0014] Preferably, a sharpness distribution map is constructed in step S2. and Its features are:

[0015] set up For the first Frame preprocessing image at position Pixel value at that location, The corresponding gradient response Laplace response and local variance They are respectively: , , ,in, Indicated by position The local neighborhood centered on, This represents the number of pixels within a local neighborhood. Indicates the first The local mean of the preprocessed image within its local neighborhood; and the gradient response respectively. Laplace response and local variance Normalization is performed to obtain the normalized gradient response. Normalized Laplace response Normalized local variance And construct a sharpness distribution map based on the following formula. : ,in, , , They represent the first Frame preprocessing image at position The adaptive fusion weights corresponding to the gradient response, Laplace response, and local variance at a given point satisfy: , , in, .

[0016] Preferably, in step S2, the sharpness distribution map is constructed by first extracting the gradient response, Laplacian response, and local variance of each pixel position in each frame of the image. Then, these three types of features are normalized respectively, and the sharpness distribution map is obtained by adaptive fusion weighting. This refines the sharpness analysis into a multi-feature extraction, normalization, and adaptive fusion implementation scheme. It no longer relies on a single sharpness index, and can more comprehensively and stably characterize the local sharpness of the image. It also provides a more reliable basis for subsequent overall sharpness evaluation, blur frame and reference frame determination, joint confidence information construction, and subsequent deblurring and recovery.

[0017] Preferably, a motion direction map is constructed in step S2. Its features are:

[0018] Based on the first frame image Horizontal gradient components and vertical gradient components and the second frame image Horizontal gradient components and vertical gradient components Extract pixel positions respectively Local dominant structure tensor at the location and , among which, the Local dominant structure tensor corresponding to the frame image Represented as: ,in, , Gaussian smoothing operator; based on the local dominant structure tensor Extract the first Frame image at pixel position The local dominant structural direction Its expression is: and the local dominant structural direction The vertical direction is taken as the first Frame image at pixel position Candidate motion fuzzy direction at the location Its expression is: According to the local dominant structure tensor eigenvalues and and gradient magnitude Determine the first Frame image at pixel position Reliability of direction ,in: , ,in, , To prevent the stability constant from having a denominator of zero; according to the first Horizontal reliability of frame image and gradient magnitude For candidate motion fuzzy directions Perform reliability screening; when and At the same time, retain the corresponding candidate motion blur direction. ,in, To preset the reliability threshold, A preset gradient threshold is set; based on the candidate motion blur directions retained in the first and second frame images. , and its corresponding directional reliability , Perform fusion to determine pixel positions Fusion candidate motion fuzzy direction and reliability of fusion direction When pixel position There exist fusion candidate motion fuzzy directions that meet the reliability screening criteria. The candidate motion fuzzy directions will be fused. As a motion direction diagram At pixel position Direction value at And the reliability of the fusion direction As a motion direction diagram At pixel position The directional reliability is determined by the algorithm; otherwise, the pixel position is obtained by propagating based on the candidate motion blur directions and their directional reliability within the high-reliability neighborhood. Direction value at and directional reliability Generate motion direction map .

[0019] Preferably, the motion direction pattern is constructed in step S2. This method constructs a local dominant structure tensor based on the horizontal and vertical gradient components of two frames of images. The local dominant structure direction is then extracted from this tensor, and its vertical direction is used as a candidate motion blur direction. Subsequently, the reliability of the direction is calculated by combining the eigenvalues ​​of the structure tensor and the gradient magnitude. When the reliability and gradient magnitude at a certain location meet a threshold condition, the candidate motion blur direction is retained; otherwise, the final motion direction map is generated by propagating candidate directions from high-reliability regions in the neighborhood. This method not only provides motion blur directions that better match the local structural features of each image location but also avoids directly using unstable directions in weak texture, low-confidence, or noisy regions through reliability judgment and neighborhood propagation mechanisms. This improves the accuracy, continuity, and robustness of motion direction estimation and provides a more reliable direction prior for subsequent direction consistency weighted statistics, overall sharpness evaluation, and inter-frame motion estimation.

[0020] Preferably, a sharpness evaluation value is constructed in step S2. and Its features are:

[0021] Let the first Frame image at pixel position The sharpness distribution value at that location is ,in, Motion Direction Diagram At pixel position The direction value given at that location is Let the first Frame image at pixel position The local dominant structural direction is pixel position Directional consistency weight Represented as: ,in, Let be the directional consistency adjustment coefficient, and directional consistency weight satisfy Based on directional consistency weights No. Overall sharpness rating of frame image Represented as: ,in, , To prevent the denominator from being zero, a stability constant is used; when the pixel position... The local dominant structural direction With direction value The closer the vertical relationship, the greater the weight of directional consistency. The larger the value, the higher the sharpness distribution value at the corresponding pixel location. Overall sharpness rating The greater the weight it occupies.

[0022] Preferably, in step S3, joint confidence information on sharpness and motion is constructed. The method is characterized by:

[0023] Suppose there is a blurred image to be recovered. At pixel position The corresponding sharpness distribution value is The normalized sharpness value is obtained. : ,in, and These represent the minimum and maximum values ​​in the sharpness distribution map of the blurred image to be recovered. To prevent the stability constant from being zero in the denominator; let the motion direction pattern be... At pixel position The reliability of the direction corresponding to that location is Then the clarity motion joint confidence information Represented as: Clarity and motion joint confidence information Depending on pixel position It increases in resolution due to reduced clarity and increased directional reliability, and can be used to adaptively adjust subsequent inter-frame motion estimation corrections, adaptive blur region determination, normalized total variation regularization term weight allocation, and motion blur kernel update processes.

[0024] Preferably, in step S3, joint confidence information on sharpness and motion is constructed. Based on the sharpness distribution values ​​and their normalized results at pixel locations in the blurred image to be restored, and combined with the directional reliability corresponding to that location, joint confidence information is constructed. It is clarified that this joint confidence information will decrease as pixel sharpness decreases and directional reliability increases, and can be combined with local motion consistency for subsequent adjustments. The originally scattered sharpness information and motion reliability information are integrated into a unified control quantity, so that subsequent processing is no longer just based on sharpness or direction alone. It can be used for unified adaptive adjustment of inter-frame motion estimation correction, adaptive blurred region determination, normalized total variational regularization term weight allocation, and motion blur kernel update, thereby improving the overall deblurring recovery's pertinence, stability, and final effect.

[0025] Preferably, the inter-frame motion estimation results are constructed in step S4. Its features are:

[0026] Based on the blurred image to be recovered and auxiliary reference images A fuzzy perception-based inter-frame motion estimation model is established. This model includes inter-frame brightness consistency constraints, motion field smoothing constraints, and a motion direction map. The provided directional prior constraints are solved to obtain the initial motion estimation results. ,in: , This represents the unknown motion vector variable in the optimization process, and the initial motion estimation result. The expression is: ,in, Define the domain for the image. For feasible assembly at the sports field, For the motion vector variables to be determined during the optimization process, Indicates the sports field to be requested Spatial gradient, The coefficient for smoothing the motion field is denoted as . For the directional prior constraint coefficient, pixel position The local motion amplitude coefficient at the location; For the direction of motion diagram At pixel position The direction value given at the location The constructed prior unit vector of direction is expressed as follows: , For motion estimation weights, motion estimation weights Based on the sharpness distribution map of the blurred image to be recovered Sharpness distribution map of auxiliary reference image Adaptive determination, its expression is: ,in, To prevent the stability constant from being zero in the denominator, the weights are estimated using motion estimation. Initial motion estimation results Adaptive weighting correction is performed to obtain inter-frame motion estimation results. ;in, Indicates the final sports field. Indicates pixel position The final motion vector at point is expressed as: ,in, For correction factor, pixel position neighborhood The average value of the initial motion vector within the local area is used to characterize the initial motion vector of local consistency, and its expression is: ,in, Represented by pixel position The neighborhood centered on, This indicates the number of pixels in the neighborhood.

[0027] Preferably, in step S4, an inter-frame motion estimation result is constructed. A fuzzy-aware inter-frame motion estimation model is established between the blurred image to be restored and the auxiliary reference image. This model not only includes inter-frame brightness consistency constraints and motion field smoothing constraints, but also introduces a priori directional constraint term provided by the motion direction map. After obtaining the initial motion estimation result, the motion estimation weight is adaptively determined according to the sharpness distribution map, and the initial motion result is weighted and corrected to obtain the final inter-frame motion estimation result. The motion estimation is no longer just a simple two-frame registration, but is simultaneously constrained by the prior fuzzy direction, local sharpness difference, and local consistency information. This makes the estimated motion direction and motion amplitude more consistent with the real fuzzing process, improving the accuracy, stability, and robustness of inter-frame motion estimation. It also provides a more reliable foundation for subsequent adaptive fuzzy region segmentation and motion fuzzy kernel construction.

[0028] Preferably, the adaptive fuzzy region is defined in step S5. Unfuzzy regions Its features are:

[0029] Pixel position Neighborhood Locally consistent motion vectors within Represented as: ,in, Representing the neighborhood The number of pixels within; based on the final motion vector With local consistency motion vector The difference between them is used to calculate the pixel position. Local motion inconsistency at the location Its expression is: Based on normalized sharpness values Inconsistency with local motion Image domain A joint determination is performed, and the regions that satisfy the following formula are identified as adaptive fuzzy regions. : ,in, For the sharpness threshold, Set a motion change threshold; and define the image domain. Except for adaptive fuzzy regions The remaining areas are defined as non-fuzzy regions. .

[0030] Preferably, the adaptive fuzzy region is defined in step S5. Unfuzzy regions This method combines the normalized sharpness value of the blurred image to be restored with the local motion inconsistency reflected by the inter-frame motion estimation results to jointly determine the image domain. First, the local motion inconsistency is calculated based on the difference between the final motion vector and the locally consistent motion vector in the neighborhood. Then, based on the sharpness threshold and the motion change threshold, regions that simultaneously meet the criteria of low sharpness and large local motion change are identified as adaptive blurred regions, while the remaining regions are identified as unblurred regions. This method can more accurately separate the blurred regions that truly need to be restored from the relatively stable and sharper regions, avoiding misclassification caused by relying solely on a single sharpness index. This provides a more reliable regional basis for subsequent motion blur kernel construction and region-differentiated deblurring restoration, improving the targeting and stability of the restoration.

[0031] Preferably, in step S5, the motion estimation results are used to... Extract the main direction of motion and main motion amplitude Based on this, a motion fuzzy kernel is constructed. Its features are:

[0032] Assume adaptive fuzzy region Inner pixel position The final motion vector at point is Then the corresponding local motion direction and local motion amplitude They are represented as follows: , Combined with motion direction diagram At pixel position The direction value given at the location Calculate pixel position Directional consistency weight Its expression is: ,in, This is the directional consistency adjustment coefficient; based on the directional consistency weight... Combined confidence information with clarity and motion Determine pixel position Effective statistical weights at each location Its expression is: , will satisfy The pixels are determined as the set of valid pixels. ,in, For effective statistical thresholding; based on the set of effective pixels. Local motion direction corresponding to each pixel and local motion amplitude Weighted statistics are used to determine the main direction of motion. and main motion amplitude ,in: , ,in, To prevent the stability constant from being zero in the denominator; based on the direction of the main motion. and main motion amplitude Constructing motion fuzzy kernel Motion blur kernel Along the main direction of motion Distribution, and the core length is determined by the amplitude of the main motion. Sure.

[0033] Preferably, in step S5, within the already defined adaptive blur region, the local motion direction and local motion amplitude of each pixel are first obtained based on the inter-frame motion estimation results. Then, the direction consistency weight is calculated by combining the direction value given by the motion direction map, and the effective statistical weight is further determined by combining the sharpness and motion joint confidence information. Based on this, the effective pixel set is selected. Subsequently, the local motion direction and local motion amplitude of the effective pixels are weighted statistically to obtain the main motion direction and main motion amplitude. Finally, the motion blur kernel is constructed according to the distribution of the main motion direction and the kernel length is determined by the main motion amplitude. Incorporating direction consistency, joint confidence information, and effective motion statistics within the region into the kernel construction process makes the obtained main motion direction and main motion amplitude more stable and more consistent with the real blur characteristics, thereby improving the accuracy and robustness of the motion blur kernel and providing a more reliable foundation for subsequent joint optimization deblurring and recovery.

[0034] Preferably, the clear image to be recovered is established in step S6. With motion fuzz kernel Joint optimization objective function Its features are:

[0035] It includes at least a data fidelity term, a normalized total variation regularization term, and a region adaptive constraint term, the expression of which is: Among them, data fidelity item Represented as: ,in, and For the image at position Pixel value at that location, Indicates the image to be restored to a clear state. Through motion fuzzy kernel Degraded blurred image; normalized total variation regularization term Represented as: ,in, For adaptive fuzzy regions, Indicates the image to be restored to a clear state. At pixel position gradient at, For normalized adjustment parameters, For adaptive regularization weights, and: ,in, Based on regularization weights, The joint confidence information adjustment coefficient. Joint confidence information for resolution and motion; region adaptive constraint term Represented as: ,in, The region is unambiguous. Indicates the pixel position of the auxiliary reference image. Pixel value at that location, The reference image consistency constraint coefficient; where, the data fidelity term Used to constrain the image to be restored to clear. Through motion fuzzy kernel Degraded image and the blurred image to be restored The difference between them, normalized total variation regularization term Used to enhance adaptive blur regions Internal deblurring recovery capability, region adaptive constraint term Used to maintain non-fuzzy regions The internal structural stability and the authenticity of details.

[0036] Preferably, in step S6, the restoration process of the image to be restored and the motion blur kernel are unified. The objective function includes at least a data fidelity term, a normalized total variational regularization term, and a region adaptive constraint term. The data fidelity term is used to constrain the difference between the restored degraded image and the original blurred image. The normalized total variational regularization term acts on the adaptive blurred region, and its weight is not fixed but is determined by the basic regularization weight, the joint confidence information adjustment coefficient, and the sharpness-motion joint confidence information. The region adaptive constraint term acts on the unblurred region, using the consistency of the auxiliary reference image to constrain the restoration result. This allows the blurred region to focus on enhancing the deblurring capability, while the unblurred region focuses on maintaining structural stability and detail authenticity. At the same time, the sharpness-motion joint confidence information is introduced into the regularization weight allocation, making the restoration process more adaptive and targeted, thereby improving the overall deblurring restoration quality.

[0037] Preferably, the motion blur kernel is fixed in step S7. Then, the joint optimization objective function is... A region-adaptive optimization solution is performed to obtain the initial deblurring result of the image to be restored to a clear state. Its expression is: ,in, This represents the initial deblurring result of the image to be restored to a clear state. This represents a fixed motion fuzzy kernel. Indicates the pixel location of the blurred image to be restored. Pixel value at that location, Indicates the pixel position of the auxiliary reference image. Pixel value at that location, Indicates an adaptive fuzzy region. Indicates a non-fuzzy region. This represents the adaptive regularization weights. This represents the normalized adjustment parameter. This represents the reference image consistency constraint coefficient; where the first term represents the data fidelity constraint, used to constrain the clear image to be restored. Fixed motion fuzzy kernel Degraded image and the blurred image to be restored The difference between them; the second term represents the difference in the adaptive fuzzy region. Internally, to recover clear images Apply a normalized total variation regularization constraint to enhance the defuzzification recovery capability within the fuzzy region; the third term represents the defuzzification recovery capability within the non-fuzzy region. Introducing auxiliary reference images Consistency constraints are applied to maintain structural stability and detail authenticity within unambiguous regions.

[0038] Preferably, in step S8, based on the deblurring result... Update motion blur kernel and to restore clear images and motion blur kernel Alternating iterative optimization is characterized by:

[0039] In adaptive fuzzy region Inside, combined with the main direction of motion Main motion amplitude and resolution motion joint confidence information For motion fuzzy kernel Update to obtain the first Sub-iteration motion blur kernel Its update objective function is expressed as: ,in, Indicates the first The clear image to be recovered is obtained from the next iteration. For kernel constraint weight coefficients, To the motion fuzz kernel The constraint function is used to constrain the motion fuzzy kernel along the main motion direction. Distribution, length and amplitude of main motion Consistent with and satisfying nonnegativity and normalization conditions; in updating the motion fuzzy kernel Then, fix the motion fuzzy kernel To restore clear images Solve the problem to obtain the first... The clear image to be recovered in the next iteration Repeat the motion blur kernel update and image recovery steps until the preset convergence condition is met: in, This is the preset convergence threshold.

[0040] Preferably, in step S8, the current motion blur kernel is updated within the adaptive blur region by combining the main motion direction, main motion amplitude, and joint confidence information of sharpness motion, resulting in a new motion blur kernel. During this update process, a kernel constraint function is used to restrict the blur kernel, ensuring that it is distributed along the main motion direction, its length is consistent with the main motion amplitude, and it satisfies non-negativity and normalization conditions. After updating the blur kernel, the blur kernel is fixed, and the region adaptive optimization method is used to resolve the image to be restored. This alternating iterative step of blur kernel update and sharp image solution is repeated until the preset convergence condition is met. Instead of relying on a one-time initial kernel estimation and initial restoration result, the motion blur kernel estimation and sharp image restoration promote each other and gradually approach the real degradation process, thereby improving the accuracy and consistency of the blur kernel estimation and further enhancing the restoration quality, stability, and robustness of the final deblurring result.

[0041] Compared with the prior art, the present invention has the following technical effects:

[0042] This invention provides a method for removing motion blur from two-frame images based on joint sharpness and motion control. First, two temporally adjacent frames of images from the same scene are acquired, and a sharpness distribution map, motion direction map, and joint sharpness and motion confidence information are constructed. Then, inter-frame motion estimation with blur perception, adaptive blur region segmentation, and motion blur kernel construction are performed, and a joint optimization model including data fidelity terms, normalized total variational regularization terms, and region adaptive constraint terms is established. Finally, the sharp image to be restored and the motion blur kernel are iteratively updated alternately, and the deblurred result image is output. This method can effectively improve the restoration quality of motion-blurred images, enhance the restoration capability of blurred regions, and maintain the structural stability and detail fidelity of unblurred regions. Attached Figure Description

[0043] Figure 1 This is an overall flowchart of the method for removing motion blur from dual-frame images according to the present invention.

[0044] Figure 2 This is a schematic diagram of the construction of the sharpness distribution map of the present invention.

[0045] Figure 3 This is a schematic diagram of the motion direction diagram generation of the present invention.

[0046] Figure 4 This is a schematic diagram of the joint confidence information and inter-frame motion estimation of the present invention.

[0047] Figure 5 A schematic diagram of the adaptive fuzzy region division and initial fuzzy kernel construction of this invention.

[0048] Figure 6 A schematic diagram of the joint optimization and alternating iterative update process of this invention. Detailed Implementation

[0049] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other implementations obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0050] This invention provides a method for removing motion blur from two-frame images based on joint sharpness and motion control. First, two temporally adjacent frames of images from the same scene are acquired, and a sharpness distribution map, motion direction map, and joint sharpness and motion confidence information are constructed. Then, inter-frame motion estimation with blur perception, adaptive blur region segmentation, and motion blur kernel construction are performed, and a joint optimization model including data fidelity terms, normalized total variational regularization terms, and region adaptive constraint terms is established. Finally, the sharp image to be restored and the motion blur kernel are iteratively updated alternately, and the deblurred result image is output. This method can effectively improve the restoration quality of motion-blurred images, enhance the restoration capability of blurred regions, and maintain the structural stability and detail fidelity of unblurred regions.

[0051] Please see Figures 1 to 6 This application presents a method for removing motion blur from two-frame images based on joint sharpness and motion control. The method aims to construct a sharpness distribution map, a motion direction map, and joint sharpness and motion confidence information by jointly analyzing the local sharpness and motion information of two temporally adjacent frames in the same scene. Based on this, it achieves the determination of the blurred image to be restored and the auxiliary reference image, inter-frame motion estimation with blur perception, adaptive blur region segmentation, motion blur kernel construction, and joint optimization and iterative updating of the sharp image to be restored and the motion blur kernel. This improves the restoration quality of motion-blurred images, enhances the deblurring and restoration capability of blurred regions, and maintains the structural stability and detail authenticity of unblurred regions.

[0052] S1. Acquire two adjacent frames of images and preprocess them to obtain the first frame image. Second frame image ;

[0053] S2, respectively for and Perform local sharpness analysis and construct a sharpness distribution map. and Generate motion direction map based on image gradient information. Based on motion direction map right and A sharpness evaluation value is obtained by weighting the values ​​based on directional consistency. and ;

[0054] Furthermore, a sharpness distribution map is constructed in step S2. and The specific steps are as follows:

[0055] set up For the first Frame preprocessing image at position Pixel value at that location, The corresponding gradient response Laplace response and local variance They are respectively: , , ,in, Indicated by position The local neighborhood centered on, This represents the number of pixels within a local neighborhood. Indicates the first The local mean of the preprocessed image within its local neighborhood; and the gradient response respectively. Laplace response and local variance Normalization is performed to obtain the normalized gradient response. Normalized Laplace response Normalized local variance And construct a sharpness distribution map based on the following formula. : ,in, , , They represent the first Frame preprocessing image at position The adaptive fusion weights corresponding to the gradient response, Laplace response, and local variance at a given point satisfy: , , That , .

[0056] Furthermore, a motion direction map is constructed in step S2. The specific steps are as follows:

[0057] Based on the first frame image Horizontal gradient components and vertical gradient components and the second frame image Horizontal gradient components and vertical gradient components Extract pixel positions respectively Local dominant structure tensor at the location and , among which, the Local dominant structure tensor corresponding to the frame image Represented as: ,in, , Gaussian smoothing operator; based on the local dominant structure tensor Extract the first Frame image at pixel position The local dominant structural direction Its expression is: and the local dominant structural direction The vertical direction is taken as the first Frame image at pixel position Candidate motion fuzzy direction at the location Its expression is: According to the local dominant structure tensor eigenvalues and and gradient magnitude Determine the first Frame image at pixel position Reliability of direction ,in: , ,in, , , To prevent the stability constant from having a denominator of zero; according to the first Horizontal reliability of frame image and gradient magnitude For candidate motion fuzzy directions Perform reliability screening; when and At the same time, retain the corresponding candidate motion blur direction. ,in, , To preset the reliability threshold, , A preset gradient threshold is set; based on the candidate motion blur directions retained in the first and second frame images. , and its corresponding directional reliability , Perform fusion to determine pixel positions Fusion candidate motion fuzzy direction and reliability of fusion direction When pixel position There exist fusion candidate motion fuzzy directions that meet the reliability screening criteria. The candidate motion fuzzy directions will be fused. As a motion direction diagram At pixel position Direction value at And the reliability of the fusion direction As a motion direction diagram At pixel position The directional reliability is determined by the algorithm; otherwise, the pixel position is obtained by propagating based on the candidate motion blur directions and their directional reliability within the high-reliability neighborhood. Direction value at and directional reliability Generate motion direction map .

[0058] Furthermore, in step S2, a sharpness evaluation value is constructed. and The specific steps are as follows:

[0059] Let the first Frame image at pixel position The sharpness distribution value at that location is ,in, Motion Direction Diagram At pixel position The direction value given at that location is Let the first Frame image at pixel position The local dominant structural direction is pixel position Directional consistency weight Represented as: ,in, , This is the directional consistency adjustment coefficient, and directional consistency weight satisfy Based on directional consistency weights No. Overall sharpness rating of frame image Represented as: ,in, , , To prevent the denominator from being zero, a stability constant is used; when the pixel position... The local dominant structural direction With direction value The closer the vertical relationship, the greater the weight of directional consistency. The larger the value, the higher the sharpness distribution value at the corresponding pixel location. Overall sharpness rating The greater the weight it occupies.

[0060] S3, according to and The size determines the blurred image to be recovered and auxiliary reference images Based on the resolution distribution map and motion direction map Constructing clarity motion joint confidence information ;

[0061] Furthermore, in step S3, joint confidence information on clarity and motion is constructed. The specific steps are as follows:

[0062] Suppose there is a blurred image to be recovered. At pixel position The corresponding sharpness distribution value is The normalized sharpness value is obtained. : ,in, and These represent the minimum and maximum values ​​in the sharpness distribution map of the blurred image to be recovered. , To prevent the stability constant from being zero in the denominator; let the motion direction pattern be... At pixel position The reliability of the direction corresponding to that location is Then the clarity motion joint confidence information Represented as: Clarity and motion joint confidence information Depending on pixel position It increases in resolution due to reduced clarity and increased directional reliability, and can be used to adaptively adjust subsequent inter-frame motion estimation corrections, adaptive blur region determination, normalized total variation regularization term weight allocation, and motion blur kernel update processes.

[0063] S4, according to and Establish inter-frame motion estimation and obtain initial motion estimation results. Motion Direction Diagram As a direction prior for inter-frame motion estimation and Used for The inter-frame motion estimation results are obtained by performing confidence correction. ;

[0064] Furthermore, in step S4, the inter-frame motion estimation results are constructed. The specific steps are as follows:

[0065] Based on the blurred image to be recovered and auxiliary reference images A fuzzy perception-based inter-frame motion estimation model is established. This model includes inter-frame brightness consistency constraints, motion field smoothing constraints, and a motion direction map. The provided directional prior constraints are solved to obtain the initial motion estimation results. ,in: , This represents the unknown motion vector variable in the optimization process, and the initial motion estimation result. The expression is: ,in, Define the domain for the image. For feasible assembly at the sports field, For the motion vector variables to be determined during the optimization process, Indicates the sports field to be requested Spatial gradient, , The coefficient for smoothing the motion field is denoted as . , For the directional prior constraint coefficient, , pixel position The local motion amplitude coefficient at the location; For the direction of motion diagram At pixel position The direction value given at the location The constructed prior unit vector of direction is expressed as follows: , For motion estimation weights, motion estimation weights Based on the sharpness distribution map of the blurred image to be recovered Sharpness distribution map of auxiliary reference image Adaptive determination, its expression is: ,in, , To prevent the stability constant from being zero in the denominator, the weights are estimated using motion estimation. Initial motion estimation results Adaptive weighting correction is performed to obtain inter-frame motion estimation results. ;in, Indicates the final sports field. Indicates pixel position The final motion vector at point is expressed as: ,in, , For correction factor, pixel position neighborhood The average value of the initial motion vector within the local area is used to characterize the initial motion vector of local consistency, and its expression is: ,in, Represented by pixel position The neighborhood centered on, This indicates the number of pixels in the neighborhood.

[0066] S5. Define the image domain Perform joint judgment to delineate adaptive fuzzy regions. Unfuzzy regions ; and Based on the inter-frame motion estimation results Extract the main motion direction and main motion amplitude Constructing motion fuzzy kernel ;

[0067] Furthermore, in step S5, an adaptive blur region is defined based on inter-frame motion estimation. Unfuzzy regions The specific steps are as follows:

[0068] Pixel position Neighborhood Locally consistent motion vectors within Represented as: ,in, Representing the neighborhood The number of pixels within; based on the final motion vector With local consistency motion vector The difference between them is used to calculate the pixel position. Local motion inconsistency at the location Its expression is: Based on normalized sharpness values Inconsistency with local motion Image domain A joint determination is performed, and the regions that satisfy the following formula are identified as adaptive fuzzy regions. : ,in, , For the sharpness threshold, , Set a motion change threshold; and define the image domain. Except for adaptive fuzzy regions The remaining areas are defined as non-fuzzy regions. .

[0069] Furthermore, in step S5, a motion blur kernel is constructed. The specific steps are as follows:

[0070] Assume adaptive fuzzy region Inner pixel position The final motion vector at point is Then the corresponding local motion direction and local motion amplitude They are represented as follows: , Combined with motion direction diagram At pixel position The direction value given at the location Calculate pixel position Directional consistency weight Its expression is: ,in, , This is the directional consistency adjustment coefficient; based on the directional consistency weight... Combined confidence information with clarity and motion Determine pixel position Effective statistical weights at each location Its expression is: , will satisfy The pixels are determined as the set of valid pixels. ,in, , For effective statistical thresholding; based on the set of effective pixels. Local motion direction corresponding to each pixel and local motion amplitude Weighted statistics are used to determine the main direction of motion. and main motion amplitude ,in: , ,in, , To prevent the stability constant from being zero in the denominator; based on the direction of the main motion. and main motion amplitude Constructing motion fuzzy kernel Motion blur kernel Along the main direction of motion Distribution, and the core length is determined by the amplitude of the main motion. Sure.

[0071] S6. Create a clear image to be recovered. With motion fuzz kernel Joint optimization objective function ;

[0072] Furthermore, the objective function is constructed in step S6. The specific steps are as follows:

[0073] Joint optimization objective function It includes at least a data fidelity term, a normalized total variation regularization term, and a region adaptive constraint term, the expression of which is: Among them, data fidelity item Represented as: ,in, and For the image at position Pixel value at that location, Indicates the image to be restored to a clear state. Through motion fuzzy kernel Degraded blurred image; normalized total variation regularization term Represented as: ,in, For adaptive fuzzy regions, Indicates the image to be restored to a clear state. At pixel position gradient at, , For normalized adjustment parameters, For adaptive regularization weights, and: ,in, , Based on regularization weights, , The joint confidence information adjustment coefficient. Joint confidence information for resolution and motion; region adaptive constraint term Represented as: ,in, The region is unambiguous. Indicates the pixel position of the auxiliary reference image. Pixel value at that location, , The reference image consistency constraint coefficient; where, the data fidelity term Used to constrain the image to be restored to clear. Through motion fuzzy kernel Degraded image and the blurred image to be restored The difference between them, normalized total variation regularization term Used to enhance adaptive blur regions Internal deblurring recovery capability, region adaptive constraint term Used to maintain non-fuzzy regions The internal structural stability and the authenticity of details.

[0074] S7. Regarding the objective function Solve the problem to obtain the clear image to be restored. Initial deblurring result ;

[0075] Furthermore, the initial deblurring result is obtained in step S7. The specific steps are as follows:

[0076] In step S7, fix the motion fuzzy kernel. Then, the joint optimization objective function is... A region-adaptive optimization solution is performed to obtain the initial deblurring result of the image to be restored to a clear state. Its expression is: ,in, This represents the initial deblurring result of the image to be restored to a clear state. This represents a fixed motion fuzzy kernel. Indicates the pixel location of the blurred image to be restored. Pixel value at that location, Indicates the pixel position of the auxiliary reference image. Pixel value at that location, Indicates an adaptive fuzzy region. Indicates a non-fuzzy region. This represents the adaptive regularization weights. , This represents the normalized adjustment parameter. , This represents the reference image consistency constraint coefficient; where the first term represents the data fidelity constraint, used to constrain the clear image to be restored. Fixed motion fuzzy kernel Degraded image and the blurred image to be restored The difference between them; the second term represents the difference in the adaptive fuzzy region. Internally, to recover clear images Apply a normalized total variation regularization constraint to enhance the defuzzification recovery capability within the fuzzy region; the third term represents the defuzzification recovery capability within the non-fuzzy region. Introducing auxiliary reference images Consistency constraints are applied to maintain structural stability and detail authenticity within unambiguous regions.

[0077] S8, according to In the adaptive fuzzy region Internal main motion direction Main motion amplitude and United Credit Information Update motion blur kernel and to and Perform alternating iterative optimization until the convergence condition is met, and output the deblurred result image. ;

[0078] Furthermore, in step S8, the deblurred result image is output. The specific steps are as follows:

[0079] In step S8, based on the initial deblurring result Update motion blur kernel and to restore clear images and motion blur kernel Alternating iterative optimization methods include: in adaptive fuzzy regions Inside, combined with the main direction of motion Main motion amplitude and resolution motion joint confidence information For motion fuzzy kernel Update to obtain the first Sub-iteration motion blur kernel Its update objective function is expressed as: ,in, Indicates the first The clear image to be recovered is obtained from the next iteration. , For kernel constraint weight coefficients, To the motion fuzz kernel The constraint function is used to constrain the motion fuzzy kernel along the main motion direction. Distribution, length and amplitude of main motion Consistent with and satisfying nonnegativity and normalization conditions; in updating the motion fuzzy kernel Then, fix the motion fuzzy kernel To restore clear images Solve the problem to obtain the first... The clear image to be recovered in the next iteration Repeat the motion blur kernel update and image recovery steps until the preset convergence condition is met: in, , This is the preset convergence threshold.

[0080] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A method for removing motion blur from two-frame images based on sharpness-motion joint control, characterized in that, Includes the following steps: S1. Acquire two adjacent frames of images and preprocess them to obtain the first frame image. Second frame image ; S2, respectively for and Perform local sharpness analysis and construct a sharpness distribution map. and ; Generate motion direction map based on image gradient information Based on motion direction map right and A sharpness evaluation value is obtained by weighting the values ​​based on directional consistency. and ; S3, according to and The size determines the blurred image to be recovered and auxiliary reference images Based on the resolution distribution map and motion direction map Constructing clarity motion joint confidence information ; S4, according to and Establish inter-frame motion estimation and obtain initial motion estimation results. Motion Direction Diagram As a direction prior for inter-frame motion estimation and Used for The inter-frame motion estimation results are obtained by performing confidence correction. ; S5. Define the image domain Perform joint judgment to delineate adaptive fuzzy regions. Unfuzzy regions ; and Based on the inter-frame motion estimation results Extract the main motion direction and main motion amplitude Constructing motion fuzzy kernel ; S6. Create a clear image to be recovered. With motion fuzz kernel Joint optimization objective function ; S7. Regarding the objective function Solve the problem to obtain the clear image to be restored. Initial deblurring result ; S8, according to In the adaptive fuzzy region Internal main motion direction Main motion amplitude and United Credit Information Update motion blur kernel and to and Perform alternating iterative optimization until the convergence condition is met, and output the deblurred result image. .

2. The method for removing motion blur from dual-frame images based on sharpness-motion joint control according to claim 1, characterized in that, Sharpness distribution map in step S2 and The construction methods include: setting For the first Frame preprocessing image at position Pixel value at that location, The corresponding gradient response Laplace response and local variance They are respectively: , , ,in, Indicated by position The local neighborhood centered on, This represents the number of pixels within a local neighborhood. Indicates the first The local mean of the preprocessed image within its local neighborhood; and the gradient response respectively. Laplace response and local variance Normalization is performed to obtain the normalized gradient response. Normalized Laplace response Normalized local variance And construct a sharpness distribution map based on the following formula. : ,in, , , They represent the first Frame preprocessing image at position The adaptive fusion weights corresponding to the gradient response, Laplace response, and local variance at a given point satisfy: , , and .

3. The method for removing motion blur from dual-frame images based on sharpness-motion joint control according to claim 1, characterized in that, Motion direction diagram in step S2 The generation methods include: based on the first frame image Horizontal gradient components and vertical gradient components and the second frame image Horizontal gradient components and vertical gradient components Extract pixel positions respectively Local dominant structure tensor at the location and , among which, the Local dominant structure tensor corresponding to the frame image Represented as: ,in, , Gaussian smoothing operator; based on the local dominant structure tensor Extract the first Frame image at pixel position The local dominant structural direction Its expression is: and the local dominant structural direction The vertical direction is taken as the first Frame image at pixel position Candidate motion fuzzy direction at the location Its expression is: According to the local dominant structure tensor eigenvalues and and gradient magnitude Determine the first Frame image at pixel position Reliability of direction ,in: , ,in, , To prevent the stability constant from having a denominator of zero; according to the first Horizontal reliability of frame image and gradient magnitude For candidate motion fuzzy directions Perform reliability screening; when and At the same time, retain the corresponding candidate motion blur direction. ,in, To preset the reliability threshold, A preset gradient threshold is set; based on the candidate motion blur directions retained in the first and second frame images. , and its corresponding directional reliability , Perform fusion to determine pixel positions Fusion candidate motion fuzzy direction and reliability of fusion direction When pixel position There exist fusion candidate motion fuzzy directions that meet the reliability screening criteria. The candidate motion fuzzy directions will be fused. As a motion direction diagram At pixel position Direction value at And the reliability of the fusion direction As a motion direction diagram At pixel position The directional reliability is determined by the algorithm; otherwise, the pixel position is obtained by propagating based on the candidate motion blur directions and their directional reliability within the high-reliability neighborhood. Direction value at and directional reliability Generate motion direction map .

4. The method for removing motion blur from dual-frame images based on sharpness-motion joint control according to claim 3, characterized in that, In step S2, based on the motion direction pattern For the sharpness distribution map and We perform directional consistency weighted statistics to obtain the overall sharpness evaluation value. and The methods include: Let the first Frame image at pixel position The sharpness distribution value at that location is ,in, Motion Direction Diagram At pixel position The direction value given at that location is Let the first Frame image at pixel position The local dominant structural direction is pixel position Directional consistency weight Represented as: ,in, This is the directional consistency adjustment coefficient, and directional consistency weight satisfy Based on directional consistency weights No. Overall sharpness rating of frame image Represented as: ,in, , To prevent the denominator from being zero, a stability constant is used; when the pixel position... The local dominant structural direction With direction value The closer the vertical relationship, the greater the weight of directional consistency. The larger the value, the higher the sharpness distribution value at the corresponding pixel location. Overall sharpness rating The greater the weight it occupies.

5. The method for removing motion blur from dual-frame images based on sharpness-motion joint control according to claim 1, characterized in that, In step S3, construct the joint confidence information of sharpness and motion. The methods include: setting a blurred image to be recovered At pixel position The corresponding sharpness distribution value is The normalized sharpness value is obtained. : ,in, and These represent the minimum and maximum values ​​in the sharpness distribution map of the blurred image to be recovered. To prevent the stability constant from being zero in the denominator; let the motion direction pattern be... At pixel position The reliability of the direction corresponding to that location is Then the clarity motion joint confidence information Represented as: Clarity and motion joint confidence information Depending on pixel position It increases in resolution due to reduced clarity and increased directional reliability, and can be used to adaptively adjust subsequent inter-frame motion estimation corrections, adaptive blur region determination, normalized total variation regularization term weight allocation, and motion blur kernel update processes.

6. The method for removing motion blur from dual-frame images based on sharpness-motion joint control according to claim 1, characterized in that, In step S4, according to and Establish inter-frame motion estimation and obtain initial motion estimation results. Motion Direction Diagram As a direction prior for inter-frame motion estimation, the sharpness distribution map is used to... The inter-frame motion estimation results are obtained by performing confidence correction. ;in: , This represents the unknown motion vector variable in the optimization process, and the initial motion estimation result. The expression is: ,in, Define the domain for the image. For feasible assembly at the sports field, For the motion vector variables to be determined during the optimization process, Indicates the sports field to be requested Spatial gradient, The coefficient for smoothing the motion field is denoted as . For the directional prior constraint coefficient, pixel position The local motion amplitude coefficient at the location; For the direction of motion diagram At pixel position The direction value given at the location The constructed prior unit vector of direction is expressed as follows: , For motion estimation weights, motion estimation weights Based on the sharpness distribution map of the blurred image to be recovered Sharpness distribution map of auxiliary reference image Adaptive determination, its expression is: ,in, To prevent the stability constant from being zero in the denominator, the weights are estimated using motion estimation. Initial motion estimation results Adaptive weighting correction is performed to obtain inter-frame motion estimation results. ;in, Indicates the final sports field. Indicates pixel position The final motion vector at point is expressed as: ,in, For correction factor, pixel position neighborhood The average value of the initial motion vector within the local area is used to characterize the initial motion vector of local consistency, and its expression is: ,in, Represented by pixel position The neighborhood centered on, This indicates the number of pixels in the neighborhood.

7. The method for removing motion blur from dual-frame images based on sharpness-motion joint control according to claim 6, characterized in that, In step S5, the sharpness distribution map of the blurred image to be recovered is used as a basis. Inter-frame motion estimation results Image domain Perform joint judgment and delineate adaptive fuzzy regions. Unfuzzy regions The methods include: Indicates pixel position The final motion vector at the location; pixel position Neighborhood Locally consistent motion vectors within Represented as: ,in, Representing the neighborhood The number of pixels within; based on the final motion vector With local consistency motion vector The difference between them is used to calculate the pixel position. Local motion inconsistency at the location Its expression is: Based on normalized sharpness values Inconsistency with local motion Image domain A joint determination is performed, and the regions that satisfy the following formula are identified as adaptive fuzzy regions. : ,in, For the sharpness threshold, Set a motion change threshold; and define the image domain. Except for adaptive fuzzy regions The remaining areas are defined as non-fuzzy regions. .

8. The method for removing motion blur from dual-frame images based on sharpness-motion joint control according to claim 7, characterized in that, In step S5, in the adaptive fuzzy region Within, based on the inter-frame motion estimation results Extract the main motion direction and main motion amplitude Based on this, a motion fuzzy kernel is constructed. The methods include: setting an adaptive fuzzy region Inner pixel position The final motion vector at point is Then the corresponding local motion direction and local motion amplitude They are represented as follows: , Combined with motion direction diagram At pixel position The direction value given at the location Calculate pixel position Directional consistency weight Its expression is: ,in, This is the directional consistency adjustment coefficient; based on the directional consistency weight... Combined confidence information with clarity and motion Determine pixel position Effective statistical weights at each location Its expression is: , will satisfy The pixels are determined as the set of valid pixels. ,in, For effective statistical thresholding; based on the set of effective pixels. Local motion direction corresponding to each pixel and local motion amplitude Weighted statistics are used to determine the main direction of motion. and main motion amplitude ,in: , ,in, To prevent the stability constant from being zero in the denominator; based on the direction of the main motion. and main motion amplitude Constructing motion fuzzy kernel Motion blur kernel Along the main direction of motion Distribution, and the core length is determined by the amplitude of the main motion. Sure.

9. A method for removing motion blur from dual-frame images based on sharpness-motion joint control according to claim 1, characterized in that, In step S6, a clear image to be recovered is created. With motion fuzz kernel Joint optimization objective function Jointly optimize the objective function It includes at least a data fidelity term, a normalized total variation regularization term, and a region adaptive constraint term, the expression of which is: Among them, data fidelity item Represented as: ,in, and For the image at position Pixel value at that location, Indicates the image to be restored to a clear state. Through motion fuzzy kernel Degraded blurred image; normalized total variation regularization term Represented as: ,in, For adaptive fuzzy regions, Indicates the image to be restored to a clear state. At pixel position gradient at, For normalized adjustment parameters, For adaptive regularization weights, and: ,in, Based on regularization weights, The joint confidence information adjustment coefficient. Joint confidence information for resolution and motion; region adaptive constraint term Represented as: ,in, The region is unambiguous. Indicates the pixel position of the auxiliary reference image. Pixel value at that location, The reference image consistency constraint coefficient; where, the data fidelity term Used to constrain the image to be restored to clear. Through motion fuzzy kernel Degraded image and the blurred image to be restored The difference between them, normalized total variation regularization term Used to enhance adaptive blur regions Internal deblurring recovery capability, region adaptive constraint term Used to maintain non-fuzzy regions The internal structural stability and the authenticity of details.

10. A method for removing motion blur from dual-frame images based on sharpness-motion joint control according to claim 9, characterized in that, In step S7, the joint optimization objective function is... A region-adaptive optimization solution is performed to obtain the initial deblurring result of the image to be restored to a clear state. Its expression is: ,in, This represents the initial deblurring result of the image to be restored to a clear state. This represents a fixed motion fuzzy kernel. Indicates the pixel location of the blurred image to be restored. Pixel value at that location, Indicates the pixel position of the auxiliary reference image. Pixel value at that location, Indicates an adaptive fuzzy region. Indicates a non-fuzzy region. This represents the adaptive regularization weights. This represents the normalized adjustment parameter. This represents the reference image consistency constraint coefficient; where the first term represents the data fidelity constraint, used to constrain the clear image to be restored. Fixed motion fuzzy kernel Degraded image and the blurred image to be restored The difference between them; the second term represents the difference in the adaptive fuzzy region. Internally, to recover clear images Apply a normalized total variational regularization constraint to enhance the defuzzification recovery capability within the fuzzy region; the third term represents the defuzzification recovery capability within the non-fuzzy region. Introducing auxiliary reference images Consistency constraints are applied to maintain structural stability and detail authenticity within unambiguous regions.

11. A method for removing motion blur from dual-frame images based on sharpness-motion joint control according to claim 10, characterized in that, In step S8, based on the initial deblurring result Using it as the initial iterative image, and in subsequent iterations according to the first... The next iteration yields a clear image to be recovered. Update motion blur kernel and to restore clear images and motion blur kernel Alternating iterative optimization methods include: in adaptive fuzzy regions Inside, combined with the main direction of motion Main motion amplitude and resolution motion joint confidence information For motion fuzzy kernel Update to obtain the first Sub-iteration motion blur kernel Its update objective function is expressed as: ,in, Indicates the first The clear image to be recovered is obtained from the next iteration. For kernel constraint weight coefficients, To the motion fuzz kernel The constraint function is used to constrain the motion fuzzy kernel along the main motion direction. Distribution, length and amplitude of main motion Consistent with and satisfying nonnegativity and normalization conditions; in updating the motion fuzzy kernel Then, fix the motion fuzzy kernel To restore clear images Solve the problem to obtain the first... The clear image to be recovered in the next iteration Repeat the motion blur kernel update and image recovery steps until the preset convergence condition is met: in, This is the preset convergence threshold.

Images