Dental CBCT image filtering and denoising method and system

Abstract

The present invention relates to the field of image denoising processing technology, and in particular to a dental CBCT image filtering denoising method and system, the method comprising: obtaining a dental CBCT image to be denoised as a target image, and clustering voxel points in the target image; constructing a three-dimensional coordinate system with the voxel point representing the foreground center point in the target image as the origin; determining the symmetry distance between each two voxel points according to the coordinates of each two voxel points in the three-dimensional coordinate system; determining the symmetry plane, and determining the structural symmetry similarity between each two voxel points; determining the target symmetry similarity between each two target clusters; screening out a similar cluster set corresponding to each target cluster; using the similar cluster set corresponding to the target cluster to which each voxel point belongs as a search window, and performing non-local mean filtering on each voxel point. The present invention realizes denoising of dental CBCT images and improves the effect of denoising dental CBCT images.

Description

Technical Field

[0001] The present invention relates to the technical field of image denoising processing, and in particular to a dental CBCT image filtering denoising method and system. Background Art

[0002] With the development of science and technology, the application of image denoising is becoming more and more extensive. For example, it can be applied to dental CBCT image denoising. At present, when denoising an image, the usual method is to perform non-local mean filtering on the image to achieve image denoising. At present, in the process of non-local mean filtering, in order to improve the filtering efficiency, a fixed-size search window is often set through manual experience.

[0003] However, when a fixed-size search window is set through manual experience to perform non-local mean filtering on dental CBCT images, the search window is often set based on manual subjective experience, resulting in the setting result being affected by manual subjective factors, so the setting result may not be accurate, resulting in poor denoising effect on dental CBCT images; secondly, the situations of similar areas corresponding to different parts and regions are often different, so the search window size required for filtering voxel points in different parts and regions is often different. Therefore, if a fixed search window is set for all voxel points in the dental CBCT image, the search window set for some voxel points may be too small, which makes it difficult to accurately achieve denoising of the dental CBCT image, and further leads to poor denoising effect on the dental CBCT image. Summary of the invention

[0004] In order to solve the technical problem of poor denoising effect of dental CBCT images, the present invention proposes a dental CBCT image filtering denoising method and system.

[0005] In a first aspect, the present invention provides a dental CBCT image filtering and denoising method, the method comprising:

[0006] The dental CBCT image to be denoised is obtained as the target image, and the voxel points in the target image are clustered according to the grayscale values ​​and positions corresponding to the voxel points in the target image to obtain the target clusters;

[0007] A three-dimensional coordinate system is constructed with the voxel point representing the foreground center point in the target image as the origin, wherein the three-dimensional coordinate system is a spatial coordinate system, which takes the horizontal direction facing the face in the foreground space as the horizontal axis, the direction perpendicular to the horizontal axis on the horizontal plane as the vertical axis, and the vertical direction of the foreground space as the vertical axis;

[0008] According to the coordinates of each two voxel points in the three-dimensional coordinate system, determine the symmetry distance between each two voxel points;

[0009] The plane formed by the longitudinal axis and the vertical axis of the three-dimensional coordinate system is determined as the symmetry plane, and the structural symmetry similarity between every two voxel points is determined according to the gradient corresponding to the symmetry plane and every two voxel points;

[0010] According to the symmetry distance and structural symmetry similarity between the voxel points in each two target clusters, the target symmetry similarity between each two target clusters is determined;

[0011] According to the target symmetric similarity between each target cluster and other target clusters, the similar cluster set corresponding to each target cluster is screened out;

[0012] The similar cluster set corresponding to the target cluster to which each voxel point belongs is used as the search window, and non-local mean filtering is performed on each voxel point.

[0013] In combination with the first aspect, in a possible implementation manner, determining the symmetry distance between every two voxel points according to the coordinates of every two voxel points in the three-dimensional coordinate system includes:

[0014] The square of the sum of the horizontal coordinates of each two voxel points in the three-dimensional coordinate system is determined as the lateral symmetry factor between each two voxel points;

[0015] The square of the difference between the longitudinal coordinates of each two voxel points in the three-dimensional coordinate system is determined as the longitudinal symmetry factor between each two voxel points;

[0016] The square of the difference between the vertical coordinates of each two voxel points in the three-dimensional coordinate system is determined as the vertical symmetry factor between each two voxel points;

[0017] The symmetry distance between every two voxel points is determined according to the transverse symmetry factor, the longitudinal symmetry factor and the vertical symmetry factor between every two voxel points, wherein the transverse symmetry factor, the longitudinal symmetry factor and the vertical symmetry factor are all positively correlated with the symmetry distance.

[0018] In combination with the first aspect, in a possible implementation manner, determining the structural symmetry similarity between every two voxel points according to the symmetry plane and the gradient corresponding to every two voxel points includes:

[0019] If the gradient direction corresponding to the voxel point is not parallel to the symmetry plane, the voxel point is determined as a reference point;

[0020] If the gradient direction corresponding to the voxel point is parallel to the symmetry plane, the voxel point is determined as the target point;

[0021] Determine the structural symmetry similarity between each two reference points according to the angle between the gradient direction corresponding to each two reference points and the symmetry plane, and the difference between the gradient magnitudes corresponding to each two reference points;

[0022] According to the angle between the gradient directions corresponding to each two target points and the difference between the gradient magnitudes corresponding to each two target points, the structural symmetry similarity between each two target points is determined;

[0023] The preset value is determined as the structural symmetry similarity between any reference point and any target point.

[0024] In combination with the first aspect above, in a possible implementation manner, the formula corresponding to the structural symmetry similarity between the reference points and the structural symmetry similarity between the target points is:

[0025] ;

[0026] ; in, It is The reference point and The structural symmetry similarity between the reference points; and are the serial numbers of different reference points; It is the absolute value function; It is The gradient size corresponding to the reference point; It is The gradient size corresponding to the reference point; is the tangent function; It is The angle between the gradient direction corresponding to the reference point and the symmetry plane; It is The angle between the gradient direction corresponding to the reference point and the symmetry plane; and is a pre-set factor greater than 0; It is The target point and The structural symmetry similarity between target points; and is the serial number of different target points; It is The gradient size corresponding to the target point; It is The gradient size corresponding to the target point; It is The gradient direction corresponding to the target point is The angle between the gradient directions corresponding to the target points.

[0027] In combination with the first aspect, in a possible implementation manner, determining the target symmetry similarity between each two target clusters according to the symmetry distance and the structural symmetry similarity between the voxel points in each two target clusters includes:

[0028] Determine any two target clusters as a first marker cluster and a second marker cluster respectively, determine any voxel point in the first marker cluster as a marker voxel point, and select a voxel point with the smallest symmetry distance with the marker voxel point from the second marker cluster as a reference voxel point corresponding to the marker voxel point;

[0029] According to the structural symmetry similarity between all voxel points in the first marker cluster and the corresponding reference voxel points, the target symmetry similarity between the first marker cluster and the second marker cluster is determined.

[0030] In combination with the first aspect above, in a possible implementation manner, the formula corresponding to the target symmetric similarity between the first marker cluster and the second marker cluster is:

[0031] ;in, is the target symmetric similarity between the first and second labeled clusters; is the normalization function; is the Euclidean distance between the cluster center point of the first marker cluster and the projection point of the cluster center point of the second marker cluster onto the symmetry plane; is the absolute value function; is the distance from the cluster center of the first marker cluster to the symmetry plane; is the distance from the cluster center of the second marker cluster to the symmetry plane; is a pre-set factor greater than 0; is the variance of the Euclidean distance between the reference voxel points corresponding to all voxel points in the first marker cluster and the cluster center point of the second marker cluster; is the number of voxel points in the first labeled cluster; is the sequence number of the voxel point in the first marked cluster; is the first marker cluster The structural symmetry similarity between a voxel and its corresponding reference voxel.

[0032] In combination with the first aspect above, in a possible implementation, the step of screening out a similar cluster set corresponding to each target cluster according to the target symmetric similarity between each target cluster and other target clusters includes:

[0033] Determine any target cluster as a reference cluster, and determine each target cluster except the reference cluster in all target clusters as a candidate cluster;

[0034] The target symmetry similarities between the reference cluster and all candidate clusters are sorted in descending order to obtain a target symmetry similarity sequence corresponding to the reference cluster, the difference between each adjacent target symmetry similarity in the target symmetry similarity sequence is determined as a target difference, two target symmetry similarities corresponding to the maximum target difference are screened out from the target symmetry similarity sequence, and the larger target symmetry similarity of the two target symmetry similarities is determined as a target similarity threshold corresponding to the reference cluster;

[0035] Select candidate clusters whose target symmetric similarity with the reference cluster is greater than the target similarity threshold from all candidate clusters as candidate clusters to obtain the set of candidate clusters corresponding to the reference cluster. Similarly, determine the set of candidate clusters corresponding to each candidate cluster.

[0036] If the reference cluster and the candidate cluster are to-be-determined clusters in the corresponding to-be-determined cluster set, the candidate cluster is determined as a similar cluster of the reference cluster;

[0037] All similar clusters of the reference cluster constitute a similar cluster set corresponding to the reference cluster.

[0038] In combination with the first aspect above, in a possible implementation, the size of the neighborhood block in the process of performing non-local mean filtering on the voxel point is less than or equal to the average size of all similar clusters in the similar cluster set corresponding to the target cluster to which the voxel point belongs.

[0039] In combination with the first aspect above, in a possible implementation, the formula corresponding to the similarity error in the process of performing non-local mean filtering on voxel points is:

[0040] ;

[0041] ;in, is the first The neighborhood block corresponding to the individual pixel point and the corresponding search window The similarity error between the neighborhood blocks corresponding to the individual pixel points; is the serial number of the voxel point in the target image; is the first The sequence number of the voxel point in the search window corresponding to the voxel point; is a natural exponential function; It is The target cluster to which the individual pixel belongs is The target symmetric similarity between the target clusters to which the individual pixels belong; Characterize the rough error situation about similarity; is the number of voxel points in the neighborhood block; is the sequence number of the voxel point in the neighborhood block; It is The first Individual points, and The first Structural symmetry similarity between individual pixels; It is The first Gray value corresponding to individual pixel points; It is The first The gray value corresponding to each pixel.

[0042] In a second aspect, the present invention provides a dental CBCT image filtering and denoising system, the system comprising:

[0043] An image acquisition clustering module is used to acquire a dental CBCT image to be denoised as a target image, and cluster the voxel points in the target image according to the grayscale values ​​and positions corresponding to the voxel points in the target image to obtain a target cluster;

[0044] A coordinate system construction module, used to construct a three-dimensional coordinate system with the voxel point representing the foreground center point in the target image as the origin, wherein the three-dimensional coordinate system is a spatial coordinate system, which takes the horizontal direction facing the face in the foreground space as the horizontal axis, the direction perpendicular to the horizontal axis on the horizontal plane as the vertical axis, and the vertical direction of the foreground space as the vertical axis;

[0045] A symmetry distance determination module, used to determine the symmetry distance between every two voxel points according to the coordinates of every two voxel points in the three-dimensional coordinate system;

[0046] A structural symmetry similarity determination module is used to determine the plane formed by the longitudinal axis and the vertical axis of the three-dimensional coordinate system as the symmetry plane, and determine the structural symmetry similarity between every two voxel points according to the symmetry plane and the gradient corresponding to every two voxel points;

[0047] A target symmetry similarity determination module is used to determine the target symmetry similarity between each two target clusters according to the symmetry distance and structural symmetry similarity between the voxel points in each two target clusters;

[0048] A similar cluster screening module is used to screen out a similar cluster set corresponding to each target cluster according to the target symmetric similarity between each target cluster and other target clusters;

[0049] The non-local mean filtering module is used to perform non-local mean filtering on each voxel point by taking the similar cluster set corresponding to the target cluster to which each voxel point belongs as the search window.

[0050] In a third aspect, a server is provided, comprising a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the device executes the method in the first aspect or any possible implementation of the first aspect.

[0051] In a fourth aspect, a computer program product is provided, comprising: a computer program code, which, when executed on a computer, enables the computer to execute the method in the first aspect or any possible implementation of the first aspect.

[0052] In a fifth aspect, a computer-readable storage medium is provided, which stores a computer program code. When the computer program code runs on a computer, the computer executes the method in the above-mentioned first aspect or any possible implementation manner of the first aspect.

[0053] The present invention has the following beneficial effects:

[0054] The present invention provides a dental CBCT image filtering and denoising method, which realizes the denoising of dental CBCT images, solves the technical problem of poor denoising effect of dental CBCT images, and improves the denoising effect of dental CBCT images. Compared with setting a fixed-size search window through artificial experience, the present invention quantifies the similarity between multiple voxel points or target clusters, such as symmetry distance, structural symmetry similarity and target symmetry similarity. Secondly, the similar cluster set corresponding to the target cluster to which the quantized voxel point belongs can often characterize the area symmetrically similar to the voxel point. Therefore, the similar cluster set corresponding to the target cluster to which each voxel point belongs is used as the search window, and the search window is objectively and adaptively set for each voxel point, and the influence of human subjective factors is reduced to a certain extent, thereby improving the rationality of the search window setting, and further improving the denoising effect of the dental CBCT image to be denoised. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 A flowchart of a dental CBCT image filtering and denoising method according to the present invention;

[0057] Figure 2A schematic diagram of the structure of a dental CBCT image filtering and denoising system according to the present invention;

[0058] Figure 3 The figure is a schematic diagram of the structure of a computer device of the present invention. DETAILED DESCRIPTION

[0059] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the technical solutions proposed by the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0060] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0061] refer to Figure 1 , showing the process of some embodiments of a dental CBCT image filtering and denoising method of the present invention. The dental CBCT image filtering and denoising method comprises the following steps:

[0062] Step S1, obtaining a dental CBCT image to be denoised as a target image, clustering the voxel points in the target image according to the grayscale values ​​and positions corresponding to the voxel points in the target image to obtain a target cluster.

[0063] The dental CBCT image to be denoised may be a dental CBCT image to be denoised. A dental CBCT image, that is, a CBCT (Cone Beam Computed Tomography) image of a dental patient, mainly includes parts such as bones and tooth structures, and is a three-dimensional image. Similar to a pixel being an indivisible unit or element in the entire two-dimensional image, a voxel point can be regarded as an indivisible unit or element in the entire three-dimensional image.

[0064] It should be noted that CBCT images are three-dimensional images obtained by scanning with a cone-shaped X-ray beam and reconstructing it with a computer algorithm. Compared with traditional CT (Computed Tomography) images, CBCT uses a cone-shaped X-ray beam instead of a fan-shaped beam, which enables it to acquire data for the entire volume in one scan. Due to its lower radiation dose and higher spatial resolution, CBCT has been widely used in stomatology.

[0065] Secondly, in dental diagnosis and treatment, high-quality CBCT images are crucial to improving diagnostic accuracy and treatment effectiveness. However, due to various reasons, such as the uncertainty of the current, light field, flat panel detector, etc. in the imaging system after the equipment has been running for a long time, as well as patient movement, the acquisition process may be disturbed, resulting in a certain amount of noise in the final reconstructed CBCT image, which may affect the dentist's diagnosis and treatment decisions. Therefore, it is very important to denoise the CBCT image, and the embodiment of the present invention is mainly used to achieve denoising of the above-mentioned dental CBCT images to be denoised.

[0066] As an example, this step may include the following steps:

[0067] The first step is to obtain the dental CBCT image to be denoised as the target image.

[0068] For example, a CBCT image of a dental patient may be acquired by a cone beam computed tomography device as a target image.

[0069] It should be noted that the patient's X-ray projection data is collected using a cone beam X-ray computed tomography (CBCT) device and converted into three-dimensional data using analytical algorithms such as FDK (Feldkamp-Davis-Kress, analytical reconstruction for image reconstruction) to obtain a three-dimensional CBCT image; during the acquisition process, it is often necessary to ensure that the patient is in the center of the scanning device as much as possible in order to obtain all-round image data. About the data format: CBCT image is a three-dimensional volume data format, a stereo data set composed of a series of pixels (called voxels in three-dimensional images), each voxel represents a small volume element in space, similar to a small cube in three-dimensional space. These voxels form a three-dimensional volume according to a specific arrangement and size, representing the internal structure of the scanned object, for example, the scanned object can be the patient's maxillofacial region. Each voxel can contain the following data information: a grayscale value representing the degree of X-ray absorption at the voxel point position, and the coordinate information of the voxel point in three-dimensional space. Secondly, in the process of collecting X-ray projection data through cone beam computed tomography (CBCT) equipment, it is necessary to ensure that the data is collected evenly at multiple angles. However, in the actual collection process, factors such as long-term operation of the equipment or patient movement may cause abnormal data in the collected X-ray projection data, further causing noise in the reconstructed three-dimensional image data, affecting the dentist's diagnosis process. Therefore, it is often necessary to perform noise reduction on the image.

[0070] In the second step, the voxel points in the target image are clustered according to the grayscale values ​​and positions corresponding to the voxel points in the target image to obtain the target clusters.

[0071] For example, the voxel points in the target image can be clustered according to the grayscale values ​​and positions corresponding to the voxel points in the target image through the K-means algorithm, and the voxel points with adjacent positions and similar grayscales can be clustered into the same cluster, and each cluster can be used as the target cluster, where the number of clusters can be obtained through the elbow method.

[0072] Optionally, the method for obtaining the target cluster may also be: based on the idea of ​​superpixel segmentation, voxel points with adjacent positions and similar grayscale in the target image may be segmented into the same image block, and each image block obtained by segmentation may be used as the target cluster.

[0073] It should be noted that the CBCT images of dental patients mainly include bones and tooth structures, which are distributed in different positions in three-dimensional space, that is, the characteristics of images in different regions are different. When performing noise reduction analysis on images, the characteristics of different regions are also different when affected by noise. Therefore, clustering voxel points in the target image can achieve segmentation of different parts, for example, segmentation of each tooth can be achieved, which can improve the accuracy of subsequent noise analysis in the image.

[0074] Step S2, constructing a three-dimensional coordinate system with the voxel point representing the foreground center point in the target image as the origin.

[0075] The foreground of the target image may be a photographed part related to teeth below the nose and above the chin, for example, it may include the bones and tooth structures of dental patients. The foreground center point may be the center of gravity of the foreground of the target image. The above three-dimensional coordinate system is a spatial coordinate system, including a horizontal axis, a vertical axis and a vertical axis. Specifically, the horizontal direction facing the face in the foreground space is the horizontal axis, the direction perpendicular to the horizontal axis on the horizontal plane is the vertical axis, and the vertical direction of the foreground space is the vertical axis. The horizontal plane may be a plane perpendicular to the vertical direction.

[0076] As an example, the voxel point representing the foreground center point in the target image can be taken as the origin, the horizontal direction facing the face in the foreground space as the horizontal axis, and the horizontal direction pointing to the right of the face as the positive direction of the horizontal axis; the vertical direction of the foreground space is taken as the vertical axis, and the vertical direction pointing upward to the top of the head is taken as the positive vertical direction; the direction perpendicular to the plane formed by the horizontal and vertical axes is taken as the longitudinal axis, and the direction pointing to the back of the head is taken as the positive direction of the longitudinal axis, to construct a three-dimensional coordinate system.

[0077] It should be noted that dental CBCT images provide structural and imaging information of bones and teeth. This part of the human body structure usually has a certain degree of symmetry, and its symmetry relationship is often about left-right symmetry, such as the teeth on the left and right sides often have a certain degree of symmetry. Therefore, different parts are often symmetrical about the plane formed by the longitudinal and vertical axes of the three-dimensional coordinate system, so the plane formed by the longitudinal and vertical axes of the three-dimensional coordinate system can be called a symmetry plane. Secondly, for different target clusters, similar symmetrical structures can usually be found in the image, and similar structures can often provide a certain reference for the image denoising process.

[0078] Step S3, determining the symmetry distance between every two voxel points according to the coordinates of every two voxel points in the three-dimensional coordinate system.

[0079] As an example, this step may include the following steps:

[0080] In the first step, the square of the sum of the horizontal coordinates of every two voxel points in the three-dimensional coordinate system is determined as the lateral symmetry factor between every two voxel points.

[0081] In the second step, the square of the difference between the longitudinal coordinates of every two voxel points in the three-dimensional coordinate system is determined as the longitudinal symmetry factor between every two voxel points.

[0082] In the third step, the square of the difference between the vertical coordinates of every two voxel points in the three-dimensional coordinate system is determined as the vertical symmetry factor between every two voxel points.

[0083] The fourth step is to determine the symmetry distance between every two voxel points according to the lateral symmetry factor, longitudinal symmetry factor and vertical symmetry factor between every two voxel points.

[0084] Among them, the horizontal symmetry factor, the longitudinal symmetry factor and the vertical symmetry factor can all be positively correlated with the symmetry distance.

[0085] For example, the formula for determining the symmetry distance between any two voxel points can be:

[0086] ;in, is the first Individual points and The symmetry distance between individual points. and is the serial number of different voxel points. is the first The horizontal coordinate of the individual pixel point in the three-dimensional coordinate system. is the first The vertical coordinate of an individual pixel in a three-dimensional coordinate system. is the first The vertical coordinate of an individual pixel in a three-dimensional coordinate system. is the first The horizontal coordinate of the individual pixel point in the three-dimensional coordinate system. is the first The vertical coordinate of an individual pixel in a three-dimensional coordinate system. is the first The vertical coordinate of an individual pixel in a three-dimensional coordinate system. Characterizes the lateral symmetry factor. Characterizes the longitudinal symmetry factor. Characterizes the vertical symmetry factor.

[0087] It should be noted that since different parts are often symmetrical about the plane formed by the longitudinal and vertical axes of the three-dimensional coordinate system, when two positions are symmetrical about the plane formed by the longitudinal and vertical axes, the relationship between the two coordinates is often: the horizontal coordinates are opposite to each other, the longitudinal coordinates are equal, and the vertical coordinates are equal. The smaller the time, the more likely it is that Individual points and The more likely the individual points are to be symmetrical about the plane formed by the longitudinal and vertical axes of the three-dimensional coordinate system, the more likely it is that the Individual points and The more likely the individual pixels are, the more likely they are to be in the same location.

[0088] Step S4: determine the plane formed by the longitudinal axis and the vertical axis of the three-dimensional coordinate system as the symmetry plane, and determine the structural symmetry similarity between every two voxel points according to the symmetry plane and the gradient corresponding to every two voxel points.

[0089] As an example, this step may include the following steps:

[0090] The first step is to determine the plane formed by the longitudinal axis and the vertical axis of the three-dimensional coordinate system as the symmetry plane.

[0091] In the second step, if the gradient direction corresponding to the voxel point is not parallel to the above symmetry plane, the voxel point is determined as the reference point.

[0092] In the third step, if the gradient direction corresponding to the voxel point is parallel to the above-mentioned symmetry plane, the voxel point is determined as the target point.

[0093] The fourth step is to determine the structural symmetry similarity between each two reference points based on the angle between the gradient direction corresponding to each two reference points and the above-mentioned symmetry plane, and the difference between the gradient sizes corresponding to each two reference points.

[0094] The fifth step is to determine the structural symmetry similarity between every two target points based on the angle between the gradient directions corresponding to every two target points and the difference between the gradient sizes corresponding to every two target points.

[0095] For example, the formula for determining the structural symmetry similarity between reference points and the structural symmetry similarity between target points can be:

[0096] ;

[0097] ;in, It is The reference point and The structural symmetry similarity between the reference points. and It is the serial number of different reference points. It is the absolute value function. It is The gradient size corresponding to the reference point. It is The gradient size corresponding to the reference point. is the tangent function. It is The angle between the gradient direction corresponding to the reference point and the symmetry plane. It is The angle between the gradient direction corresponding to the reference point and the symmetry plane. and It is a pre-set factor greater than 0, mainly used to prevent the denominator from being 0, for example, and It can be 0.001. It is The target point and The structural symmetry similarity between the target points. and are the serial numbers of different target points. It is The gradient size corresponding to the target point. It is The gradient size corresponding to the target point. It is The gradient direction corresponding to the target point is The angle between the gradient directions corresponding to the target points.

[0098] It should be noted that the variation characteristics of the X-ray absorption degree of voxels in the same part are often similar. Based on the symmetry of facial bones and teeth, for symmetrical voxels, the difference in the absorption degree of X-rays in the corresponding bone or tooth area during the X-ray scanning process is relatively small. The gradient size of the voxel point can characterize the variation of the X-ray absorption degree around the voxel point. Therefore, if the difference between the gradient sizes corresponding to two voxel points is smaller, it often means that the difference in the variation of the X-ray absorption degree around the two voxel points is smaller, which often means that the two voxel points are more likely to be voxel points in the same part, and the degree of symmetry similarity between the two voxel points is higher. Secondly, in addition to the difference in the amount of change, the difference in the direction of change is also crucial to symmetry. Since the symmetry of the facial bones in the three-dimensional coordinate system is reflected by the symmetry plane, it is possible to determine whether the gradient direction is parallel to the symmetry plane. For the gradient direction parallel to the symmetry plane, the gradient directions of the two points with higher symmetry similarity should be consistent; and for the gradient direction not parallel to the symmetry plane, the gradient directions of the two points with higher symmetry similarity should be symmetrical based on the symmetry plane. When The smaller it is, the more likely it is that the two gradient directions that are not parallel to the symmetry plane are symmetric based on the symmetry plane. The smaller it is, the more likely the two gradient directions parallel to the symmetry plane are to be close. The larger the The reference point and The higher the structural symmetry similarity between the reference points, the higher the The larger the The target point and The higher the structural symmetry similarity between the target points, the higher the

[0099] In the sixth step, the preset value is determined as the structural symmetry similarity between any reference point and any target point.

[0100] The preset value may be a preset value representing the asymmetry between the reference point and the target point. For example, the preset value may be 0.

[0101] Step S5, determining the target symmetry similarity between every two target clusters according to the symmetry distance and the structural symmetry similarity between the voxel points in every two target clusters.

[0102] As an example, this step may include the following steps:

[0103] In the first step, any two target clusters are respectively determined as the first marker cluster and the second marker cluster, and any voxel point in the first marker cluster is determined as the marker voxel point. The voxel point with the smallest symmetry distance with the marker voxel point is selected from the second marker cluster as the reference voxel point corresponding to the marker voxel point.

[0104] In the second step, according to the structural symmetry similarity between all voxel points in the first marker cluster and their corresponding reference voxel points, the target symmetry similarity between the first marker cluster and the second marker cluster is determined.

[0105] For example, the formula for determining the target symmetric similarity between the first marker cluster and the second marker cluster may be:

[0106] ;in, is the target symmetric similarity between the first labeled cluster and the second labeled cluster. is a normalization function. It is the Euclidean distance between the cluster center point of the first marker cluster and the projection point of the cluster center point of the second marker cluster onto the symmetry plane. is the absolute value function. is the distance from the cluster center of the first marker cluster to the symmetry plane. is the distance from the cluster center of the second marker cluster to the symmetry plane. It is a pre-set factor greater than 0, mainly used to prevent the denominator from being 0, for example, It can be 0.001. It is the variance of the Euclidean distance between the reference voxel points corresponding to all voxel points in the first marker cluster and the cluster center point of the second marker cluster. is the number of voxel points in the first labeled cluster. is the sequence number of the voxel point in the first labeled cluster. is the first marker cluster The structural symmetry similarity between a voxel and its corresponding reference voxel.

[0107] It should be noted that when The smaller the value, the closer the distance between the projection points of the cluster center point of the first marker cluster and the cluster center point of the second marker cluster onto the symmetry plane, which often indicates that the first marker cluster and the second marker cluster are more likely to be symmetrical about the symmetry plane. The smaller the value, the closer the distance between the first marker cluster and the second marker cluster and the symmetry plane is, which means that the first marker cluster and the second marker cluster are more likely to be symmetrical about the symmetry plane. When the value is larger, it often indicates that the second marker cluster is more likely to contain The voxel points with high structural symmetry similarity between individual voxel points. When the value is larger, it often indicates that there are more voxel points with high structural symmetry similarity between the first marker cluster and the second marker cluster, which often indicates that the first marker cluster and the second marker cluster are more likely to be symmetrically similar about the symmetry plane. When the value is larger, it often indicates that the distance distribution between the reference voxel point and the cluster center point of the second marker cluster is more discrete, which often indicates that the selection result of the reference voxel point corresponding to the voxel point in the first marker cluster is relatively better, and it often indicates that fewer voxel points are repeatedly used as reference voxel points. The larger the value is, the more likely it is that the first marker cluster and the second marker cluster are symmetrically similar about the symmetry plane.

[0108] Step S6: Filter out a similar cluster set corresponding to each target cluster according to the target symmetric similarity between each target cluster and other target clusters.

[0109] As an example, this step may include the following steps:

[0110] In the first step, any target cluster is determined as a reference cluster, and each target cluster except the reference cluster in all target clusters is determined as a candidate cluster.

[0111] In the second step, the target symmetry similarities between the above reference cluster and all candidate clusters are sorted in order from large to small to obtain the target symmetry similarity sequence corresponding to the above reference cluster, and the difference between each adjacent target symmetry similarity in the above target symmetry similarity sequence is determined as the target difference, and the two target symmetry similarities corresponding to the maximum target difference are screened out from the above target symmetry similarity sequence, and the larger target symmetry similarity of the two target symmetry similarities is determined as the target similarity threshold corresponding to the above reference cluster.

[0112] The third step is to select candidate clusters whose target symmetric similarity with the above reference cluster is greater than the target similarity threshold from all candidate clusters as the pending clusters, and obtain the pending cluster set corresponding to the above reference cluster. That is, since the reference cluster is any target cluster, the pending cluster set corresponding to each target cluster can be obtained. Therefore, the same method can be used to obtain the pending cluster set corresponding to each candidate cluster.

[0113] The undetermined cluster set corresponding to the reference cluster may include: all candidate clusters whose target symmetric similarity with the reference cluster is greater than a target similarity threshold.

[0114] In the fourth step, if the reference cluster and the candidate cluster are to-be-determined clusters in the corresponding to-be-determined cluster set, the candidate cluster is determined as a similar cluster to the reference cluster.

[0115] The fifth step is to combine all similar clusters of the above reference cluster to form a similar cluster set corresponding to the above reference cluster.

[0116] Step S7: Using the similar cluster set corresponding to the target cluster to which each voxel point belongs as the search window, perform non-local mean filtering on each voxel point.

[0117] Among them, the size of the neighborhood block in the process of non-local mean filtering of the voxel point can be less than or equal to the average size of all similar clusters in the similar cluster set corresponding to the target cluster to which the voxel point belongs. The size of the neighborhood block can be selected according to actual conditions. For example, the average size of all similar clusters in the similar cluster set corresponding to the target cluster to which the voxel point belongs can be determined as the size of the neighborhood block in the process of non-local mean filtering of the voxel point.

[0118] It should be noted that, compared with conventional non-local mean filtering, the embodiment of the present invention optimizes the settings of the search window and the neighborhood block when performing non-local mean filtering on each voxel point, and improves the calculation formula of the similarity error. However, the subsequent calculation process of how to perform non-local mean filtering based on the search window, neighborhood block and similarity error calculation formula can follow the conventional non-local mean filtering calculation process.

[0119] As an example, the formula corresponding to the similarity error in the process of non-local mean filtering of voxel points can be:

[0120] ;

[0121] ;in, is the first The neighborhood block corresponding to the individual pixel point and the corresponding search window The similarity error between the neighborhood blocks corresponding to individual pixels. is the serial number of the voxel point in the target image. is the first The sequence number of the voxel point in the search window corresponding to the voxel point. is a natural exponential function. It is The target cluster to which the individual pixel belongs is The target symmetric similarity between the target clusters to which the individual pixel points belong. It should be noted that if The target cluster to which the individual pixel belongs is If the target clusters to which the individual pixels belong are the same target cluster, then Recorded as 1. Characterizes the rough error situation with respect to similarity. is the number of voxel points in the neighborhood block. is the sequence number of the voxel point in the neighborhood block. It is The first Individual points, and The first It should be noted that the size of the neighborhood blocks used in the non-local mean filtering process for a certain voxel point is the same, that is, The neighborhood block corresponding to the individual pixel is The sizes of the neighborhood blocks corresponding to individual pixel points are the same. It is The first The gray value corresponding to each pixel. It is The first The gray value corresponding to each pixel.

[0122] It should be noted that when The larger the The target cluster to which the individual pixel belongs is The more likely the target cluster to which the individual pixel belongs is symmetrically similar about the symmetry plane, the smaller the corresponding similarity error. The larger the The first Individual points, and The first The higher the structural symmetry similarity between individual points, the smaller the corresponding similarity error. Can be used as When The larger the The first Individual points, and The first The greater the grayscale difference between individual pixels, the greater the corresponding similarity error. The larger the The neighborhood block corresponding to the individual pixel point and the corresponding search window The larger the similarity error between the neighborhood blocks corresponding to the individual pixels, the larger the similarity error between the neighborhood blocks corresponding to the individual pixels. The weighted grayscale value of the voxel point in the search window corresponding to the voxel point is calculated. The filtered gray value corresponding to the individual pixel point can be Smaller weights are set for individual pixels.

[0123] Secondly, traditional similarity error calculations often only consider , that is, directly use the mean square error (MSE), which can be equal to Therefore, the present invention takes into account Based on this, we also considered and , thereby improving the rationality of similarity error calculation and further improving the image denoising effect.

[0124] Optionally, after performing optimized filtering and noise reduction processing on the dental CBCT image, the processed image is output. The processed image quality is improved, providing a reliable diagnosis basis for dentists.

[0125] refer to Figure 2 Based on the same inventive concept as the above method embodiment, the present invention provides a dental CBCT image filtering and denoising system, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the above computer program is executed by the processor, the steps of a dental CBCT image filtering and denoising method may specifically include:

[0126] The image acquisition clustering module 201 is used to acquire the dental CBCT image to be denoised as the target image, and cluster the voxel points in the target image according to the grayscale values ​​and positions corresponding to the voxel points in the target image to obtain a target cluster;

[0127] A coordinate system construction module 202 is used to construct a three-dimensional coordinate system with the voxel point representing the foreground center point in the target image as the origin, wherein the three-dimensional coordinate system is a spatial coordinate system, which has the horizontal direction facing the face in the foreground space as the horizontal axis, the direction perpendicular to the horizontal axis on the horizontal plane as the vertical axis, and the vertical direction of the foreground space as the vertical axis;

[0128] A symmetry distance determination module 203, used to determine the symmetry distance between every two voxel points according to the coordinates of every two voxel points in the three-dimensional coordinate system;

[0129] The structural symmetry similarity determination module 204 is used to determine the plane formed by the longitudinal axis and the vertical axis of the three-dimensional coordinate system as the symmetry plane, and determine the structural symmetry similarity between every two voxel points according to the symmetry plane and the gradient corresponding to every two voxel points;

[0130] The target symmetry similarity determination module 205 is used to determine the target symmetry similarity between every two target clusters according to the symmetry distance and the structural symmetry similarity between the voxel points in every two target clusters;

[0131] A similar cluster screening module 206 is used to screen out a similar cluster set corresponding to each target cluster according to the target symmetric similarity between each target cluster and other target clusters;

[0132] The non-local mean filtering module 207 is used to perform non-local mean filtering on each voxel point by using the similar cluster set corresponding to the target cluster to which each voxel point belongs as a search window.

[0133] Figure 3 is a schematic diagram of the structure of a computer device provided by an embodiment of the present invention. Figure 3 As shown, the computer device 300 includes: a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and running on the processor 302, wherein when the processor 302 executes the computer program 303, the computer device can execute any one of the dental CBCT image filtering and denoising methods introduced above.

[0134] Based on the same inventive concept as the above method embodiment, the present invention provides a server, including a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the device performs any of the above dental CBCT image filtering and denoising methods.

[0135] Based on the same inventive concept as the above method embodiment, the present invention provides a computer program product, which includes: computer program code, when the computer program code is run on a computer, the computer executes any one of the above dental CBCT image filtering and denoising methods.

[0136] Based on the same inventive concept as the above method embodiments, the present invention provides a computer-readable storage medium, which stores a computer program code. When the computer program code runs on a computer, the computer executes any one of the above dental CBCT image filtering and denoising methods.

[0137] In summary, the present invention takes into account the characteristics of dental CBCT images in the denoising process, and optimizes the selection method of the search window and the neighborhood block when performing non-local mean filtering on each voxel point according to the structural symmetry characteristics, thereby reducing the interference of the filtering process caused by the variable image characteristics of the acquisition site, thereby improving the denoising effect; when applied in the actual diagnosis process, it can reduce the display of erroneous information caused by noise interference on the CBCT image, thereby improving the reliability of the CBCT image. That is, compared with setting a fixed-size search window through artificial experience, the present invention quantifies the similarity between multiple voxel points or target clusters, such as symmetry distance, structural symmetry similarity and target symmetry similarity. Secondly, the similar cluster set corresponding to the target cluster to which the quantified voxel point belongs can often characterize the area symmetrically similar to the voxel point. Therefore, the similar cluster set corresponding to the target cluster to which each voxel point belongs is used as the search window, and the search window is objectively and adaptively set for each voxel point, and the influence of human subjective factors is reduced to a certain extent, thereby improving the rationality of the search window setting, thereby improving the denoising effect of the dental CBCT image to be denoised. Secondly, in the process of performing non-local mean filtering on each voxel point, the present invention also comprehensively considers the target symmetry similarity between the target cluster to which each voxel point belongs and other target clusters, as well as the structural symmetry similarity and grayscale difference between the voxel points, thereby improving the effect of performing non-local mean filtering on each voxel point.

[0138] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features can be replaced by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A dental CBCT image filtering and denoising method, characterized in that: The following steps are involved: The dental CBCT image to be denoised is obtained as the target image, and the voxel points in the target image are clustered according to the grayscale values ​​and positions corresponding to the voxel points in the target image to obtain the target clusters; A three-dimensional coordinate system is constructed with the voxel point representing the foreground center point in the target image as the origin, wherein the three-dimensional coordinate system is a spatial coordinate system, which takes the horizontal direction facing the face in the foreground space as the horizontal axis, the direction perpendicular to the horizontal axis on the horizontal plane as the vertical axis, and the vertical direction of the foreground space as the vertical axis; According to the coordinates of each two voxel points in the three-dimensional coordinate system, determine the symmetry distance between each two voxel points; The plane formed by the longitudinal axis and the vertical axis of the three-dimensional coordinate system is determined as the symmetry plane, and the structural symmetry similarity between every two voxel points is determined according to the gradient corresponding to the symmetry plane and every two voxel points; According to the symmetry distance and structural symmetry similarity between the voxel points in each two target clusters, the target symmetry similarity between each two target clusters is determined; According to the target symmetric similarity between each target cluster and other target clusters, the similar cluster set corresponding to each target cluster is screened out; Using the similar cluster set corresponding to the target cluster to which each voxel point belongs as the search window, non-local mean filtering is performed on each voxel point; Determining the structural symmetry similarity between every two voxel points according to the symmetry plane and the gradient corresponding to every two voxel points includes: If the gradient direction corresponding to the voxel point is not parallel to the symmetry plane, the voxel point is determined as a reference point; If the gradient direction corresponding to the voxel point is parallel to the symmetry plane, the voxel point is determined as the target point; Determine the structural symmetry similarity between each two reference points according to the angle between the gradient direction corresponding to each two reference points and the symmetry plane, and the difference between the gradient magnitudes corresponding to each two reference points; According to the angle between the gradient directions corresponding to each two target points and the difference between the gradient magnitudes corresponding to each two target points, the structural symmetry similarity between each two target points is determined; Determine the preset value as the structural symmetry similarity between any reference point and any target point; The formula corresponding to the structural symmetry similarity between reference points and the structural symmetry similarity between target points is: ； ； in, It is The reference point and The structural symmetry similarity between the reference points; and are the serial numbers of different reference points; It is the absolute value function; It is The gradient size corresponding to the reference point; It is The gradient size corresponding to the reference point; is the tangent function; It is The angle between the gradient direction corresponding to the reference point and the symmetry plane; It is The angle between the gradient direction corresponding to the reference point and the symmetry plane; and is a pre-set factor greater than 0; It is The target point and The structural symmetry similarity between target points; and is the serial number of different target points; It is The gradient size corresponding to the target point; It is The gradient size corresponding to the target point; It is The gradient direction corresponding to the target point is The angle between the gradient directions corresponding to the target points; Determining the target symmetry similarity between each two target clusters according to the symmetry distance and the structural symmetry similarity between the voxel points in each two target clusters includes: Determine any two target clusters as a first marker cluster and a second marker cluster respectively, determine any voxel point in the first marker cluster as a marker voxel point, and select a voxel point with the smallest symmetry distance with the marker voxel point from the second marker cluster as a reference voxel point corresponding to the marker voxel point; Determining the target symmetry similarity between the first marker cluster and the second marker cluster according to the structural symmetry similarity between all voxel points in the first marker cluster and their corresponding reference voxel points; The formula corresponding to the target symmetric similarity between the first labeled cluster and the second labeled cluster is: ； in, is the target symmetric similarity between the first and second labeled clusters; is the normalization function; is the Euclidean distance between the cluster center point of the first marker cluster and the projection point of the cluster center point of the second marker cluster onto the symmetry plane; is the absolute value function; is the distance from the cluster center of the first marker cluster to the symmetry plane; is the distance from the cluster center of the second marker cluster to the symmetry plane; is a pre-set factor greater than 0; is the variance of the Euclidean distance between the reference voxel points corresponding to all voxel points in the first marker cluster and the cluster center point of the second marker cluster; is the number of voxel points in the first labeled cluster; is the sequence number of the voxel point in the first marked cluster; is the first marker cluster The structural symmetry similarity between individual voxels and their corresponding reference voxels; The method of screening out a similar cluster set corresponding to each target cluster according to the target symmetric similarity between each target cluster and other target clusters includes: Determine any target cluster as a reference cluster, and determine each target cluster except the reference cluster in all target clusters as a candidate cluster; The target symmetry similarities between the reference cluster and all candidate clusters are sorted in descending order to obtain a target symmetry similarity sequence corresponding to the reference cluster, the difference between each adjacent target symmetry similarity in the target symmetry similarity sequence is determined as a target difference, two target symmetry similarities corresponding to the maximum target difference are screened out from the target symmetry similarity sequence, and the larger target symmetry similarity of the two target symmetry similarities is determined as a target similarity threshold corresponding to the reference cluster; Select candidate clusters whose target symmetric similarity with the reference cluster is greater than the target similarity threshold from all candidate clusters as candidate clusters to obtain the set of candidate clusters corresponding to the reference cluster. Similarly, determine the set of candidate clusters corresponding to each candidate cluster. If the reference cluster and the candidate cluster are to-be-determined clusters in the corresponding to-be-determined cluster set, the candidate cluster is determined as a similar cluster of the reference cluster; All similar clusters of the reference cluster constitute a similar cluster set corresponding to the reference cluster.

2. A dental CBCT image filtering and denoising method according to claim 1, characterized in that: Determining the symmetry distance between every two voxel points according to the coordinates of every two voxel points in the three-dimensional coordinate system includes: The square of the sum of the horizontal coordinates of each two voxel points in the three-dimensional coordinate system is determined as the lateral symmetry factor between each two voxel points; The square of the difference between the longitudinal coordinates of each two voxel points in the three-dimensional coordinate system is determined as the longitudinal symmetry factor between each two voxel points; The square of the difference between the vertical coordinates of each two voxel points in the three-dimensional coordinate system is determined as the vertical symmetry factor between each two voxel points; The symmetry distance between every two voxel points is determined according to the transverse symmetry factor, the longitudinal symmetry factor and the vertical symmetry factor between every two voxel points, wherein the transverse symmetry factor, the longitudinal symmetry factor and the vertical symmetry factor are all positively correlated with the symmetry distance.

3. A dental CBCT image filtering and denoising method according to claim 1, characterized in that: The size of the neighborhood block in the process of performing non-local mean filtering on the voxel point is less than or equal to the average size of all similar clusters in the similar cluster set corresponding to the target cluster to which the voxel point belongs.

4. A dental CBCT image filtering and denoising method according to claim 1, characterized in that: The formula corresponding to the similarity error in the process of non-local mean filtering of voxel points is: ； ； in, is the first The neighborhood block corresponding to the individual pixel point and the corresponding search window The similarity error between the neighborhood blocks corresponding to the individual pixel points; is the serial number of the voxel point in the target image; is the first The sequence number of the voxel point in the search window corresponding to the voxel point; is a natural exponential function; It is The target cluster to which the individual pixel belongs is The target symmetric similarity between the target clusters to which the individual pixels belong; Characterize the rough error situation about similarity; is the number of voxel points in the neighborhood block; is the sequence number of the voxel point in the neighborhood block; It is The first Individual points, and The first Structural symmetry similarity between individual pixels; It is The first Gray value corresponding to individual pixel points; It is The first The gray value corresponding to each pixel.

5. A dental CBCT image filtering and denoising system, characterized in that: The method comprises a processor and a memory, wherein the processor is used to process instructions stored in the memory to implement a dental CBCT image filtering and denoising method according to any one of claims 1 to 4.

Images