CN117372425B - A method for detecting key points in lateral cephalograms - Google Patents
A method for detecting key points in lateral cephalograms Download PDFInfo
- Publication number
- CN117372425B CN117372425B CN202311651546.9A CN202311651546A CN117372425B CN 117372425 B CN117372425 B CN 117372425B CN 202311651546 A CN202311651546 A CN 202311651546A CN 117372425 B CN117372425 B CN 117372425B
- Authority
- CN
- China
- Prior art keywords
- points
- point
- template
- key
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30036—Dental; Teeth
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a key point detection method of a skull lateral position slice, which relates to the technical field of tooth correction, and comprises the following steps of: inputting a skull lateral position slice to a detection model to obtain an initial key point position; mapping the coordinates of the key points of the template to the image to be detected; performing super-pixel segmentation on an image to be detected; judging whether the initial key point is in the super-pixel area where the corresponding template point is located; and carrying out local regression voting on the initial key points which do not meet the standard in the corresponding super pixels to obtain the final key point positions. According to the method, the position of the key point is detected by using the mapped template point position constraint model, and the position-optimized key point is obtained by carrying out local area regression voting on the initial key point which does not accord with the distance threshold, so that the accuracy of the key point detection by the model can be effectively improved.
Description
Technical Field
The invention relates to the technical field of tooth correction, in particular to a key point detection method of a skull lateral position slice.
Background
Dentists commonly use a skull side position sheet in the orthodontic treatment process to judge whether a patient is a odontopathy or a bone bulge, and the doctor marks medical index information required by calculation of some key points on the skull side position sheet, and diagnosis and orthodontic treatment scheme are formulated by using the index information.
The automation of the X-ray projection measurement analysis enables the diagnosis and treatment design of the maxillofacial deformity to be more accurate, and the burden of an operator is greatly reduced. The existing key point positioning needs to be directly judged by a doctor, so that the doctor's workload is increased while the experience of the doctor is required to be high. The accuracy of the positioning of the key points directly influences the reliability of the subsequent measurement results, and the manual marking of the key points not only requires a great deal of time, but also inevitably generates human errors and influences the accuracy and correction effect of the measurement results.
Disclosure of Invention
The invention aims to provide a method for detecting the key points of the skull side position slice aiming at the defects in the prior art so as to solve the problem of inaccurate positioning of the key points of the skull side position slice.
The technical scheme provided by the invention is as follows:
a key point detection method of a skull side position slice comprises the following steps:
step 1: inputting the skull side position slice to a detection model to obtain initial position information of key points;
step 2: the most accurate detection key points in different quadrants are collected through statistics verification, the width ratio and the height ratio of the detection points to the head where the detection points are located, the width ratio and the height ratio of the template points to the head where the detection points are located are calculated, the difference value of the width ratio and the height ratio is calculated, if the difference value is smaller than a difference threshold value, a transformation vertex is selected, and then the coordinates of the template key points are mapped to an image to be detected through calculation of an affine transformation matrix, so that the position information of the template key points in the detected image can be obtained;
step 3: performing superpixel segmentation on an image to be detected by using a DBSCAN-based method, respectively counting detection points which are in and out of the superpixels where corresponding template points are located, and performing next judgment on the points which are not in the superpixels;
step 4: selecting a local regression voting area by combining three major interest points with super pixels, selecting template points which are positioned in non-adjacent quadrants and are closest to two detection points in the step 3 as neighborhood transformation vertexes, obtaining a neighborhood mapping point rm of the point according to the coordinate ratio of the template points which are corresponding to the non-standard points to the adjacent two transformation vertexes, taking the super pixel area where the domain mapping point rm, the template point m and the detection point d are positioned as the local voting area, calculating local regression votes by using the thermodynamic diagram in the local voting area and the offset diagram corresponding to the thermodynamic diagram, and selecting the maximum value as the key point to obtain the final key point position.
In an exemplary embodiment, in step 1, the initial position information of the key point is obtained by: converting the skull side position slice to be detected into a set size, and then inputting the size into a model to obtain thermodynamic diagram information and offset diagram information of a current image; by utilizing a key point detection method based on attention feature pyramid fusion and regression voting, pi R before thermodynamic diagram is performed by traversing thermodynamic diagram of each layer of key points 2 And calculating regression voting by using the large points and the corresponding offset graphs, wherein R defaults to 40, and the maximum value is selected as a key point to obtain initial position information of the key point, and the formula is as follows:
,
wherein,front pi R in thermodynamic diagram of kth key point 2 Large set of values, +.>Is an indication Fu Hanshu of the nature of the particular,，coordinates of key points for each pixel value alone, with highest +.>Value pixelThe coordinates of the key points considered to be the most likely.
In an exemplary embodiment, in step 2, three transformed vertices are selected.
Further, the difference threshold was 0.05.
If the difference threshold value is larger than the difference threshold value, the points which do not meet the standard are replaced, and the next point with high accuracy is selected until the transformation vertex which meets the condition is selected.
Still further, the mapping the template key point coordinates to the image to be detected includes:
mapping a preset skull side piece key point template into a coordinate system of the current skull side piece, and extracting position information of a template key point;
counting the most suitable three detection key points in the verification set, and solving the position transformation relation of the corresponding template key points by utilizing an affine transformation matrix algorithm;
and inputting the position information of the rest template key points into the position transformation relation to obtain the affine transformation coordinates of the template key points.
In an exemplary embodiment, in step 3, due to the presence of transformed vertices, a minimum of three detection points can be guaranteed to be within the super-pixel where the corresponding template point is located.
In an exemplary embodiment, in step 3, the performing super-pixel segmentation on the image to be detected includes clustering the images using a density-based clustering algorithm.
In an exemplary embodiment, in step 3, it is determined whether the detection point is in the super-pixel area where the corresponding template point is located, specifically: judging whether the key points detected by the model are in the clustering area where the corresponding template key points are.
In an exemplary embodiment, in step 3, the next determination is made on the points that are not within the superpixel by: and carrying out local regression voting in the corresponding super pixels to obtain the positions of the key points, and if the key points which are not in the corresponding clustering areas exist, carrying out regression voting on the clustering areas where the template points are positioned by using the local thermodynamic diagram and the offset diagram information, and updating the position information of the detection key points again.
In an exemplary embodiment, in step 4, the formula for obtaining the final keypoint location is expressed as follows:
，
wherein,front pi Q in local thermodynamic diagram being kth key point 2 A large set of values is provided for the set of values,h and W are the height and width of the image, respectively, and the local area LA is formulated as follows:
，
wherein,DRthe regions are partitioned for the super-pixels,indicated as the super pixel area where the detection point is located,expressed as the super-pixel area where the template point is located,Represented as a super-pixel region where the neighborhood point is located,the super pixel region where the points not within the three super pixel region among the three points z, m, r are indicated.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the invention, the initial key points are obtained through the key point detection model, and then the key points which do not meet the distance requirement can be limited near the real positions by utilizing the positions of the key points of the template and the limit of the super pixel segmentation distance threshold based on the density, so that the false detection rate of the key point positions is reduced;
2. for key points which do not meet the distance requirement, the super-pixel area where three major interesting points of the template points are located, which are mapped by using the neighborhood mapping points mapped by the position information of the adjacent template points and the template points matched by the template information, is used as a local information extraction area, and the problem that the difference between the key points of the template and the human face is large is solved by extracting the interesting areas by using various position information;
3. according to the invention, the detection points are positioned by carrying out regression voting on the key points which do not meet the distance requirement by utilizing the local information, and voting positioning is carried out by utilizing the most possibly occurring area of the key points. Therefore, the method can reduce the situation of misidentification as the key points, thereby improving the accuracy of key point identification, improving the robustness of key point detection and being suitable for the skull side position plates of more factories. Meanwhile, compared with the traditional manual fixed point, the invention has higher efficiency.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
Fig. 1 is a flowchart of a method for detecting key points of a skull side position slice according to an embodiment of the present invention.
Detailed Description
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the invention. 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.
Embodiments of the invention and features of the embodiments may be combined with each other without conflict.
The automation of the X-ray projection measurement analysis enables the diagnosis and treatment design of the maxillofacial deformity to be more accurate, and the burden of an operator is greatly reduced. The existing key point positioning needs to be directly judged by a doctor, so that the doctor's workload is increased while the experience of the doctor is required to be high. The accuracy of the positioning of the key points directly influences the reliability of the subsequent measurement results, and the manual marking of the key points not only requires a great deal of time, but also inevitably generates human errors and influences the accuracy and correction effect of the measurement results.
Based on the above, specific embodiments of the present application include the following:
referring to fig. 1, the application provides an embodiment of a method for detecting a key point of a skull lateral position slice, which specifically includes:
step 1: inputting a skull side position slice into a detection model, wherein the model can accurately obtain the initial position of a key point in an image by combining a feature pyramid fusion module and regression voting by using a key point detection method based on attention feature pyramid fusion and regression voting;
(1) The skull side position piece input in the embodiment of the invention allows the skull side position pieces shot by different machine manufacturers with various bone types, different sexes, different age stages, namely the method can be suitable for various skull side position pieces with different types;
(2) The embodiment of the invention does not limit an initial key point detection model, can utilize a key point detection method based on attention feature pyramid fusion and regression voting, can more accurately extract key point information in an image by combining a feature pyramid fusion module and regression voting, and can also detect by utilizing other key point detection algorithms to obtain the initial key point position;
(3) Converting the skull side position slice to be detected into a set size, and then inputting the size into a model to obtain thermodynamic diagram information and offset diagram information of a current image;
(4) Traversing the thermodynamic diagram of each layer of key points, and leading the thermodynamic diagram to be pi R 2 And calculating regression voting by using the large points and the corresponding offset graphs, wherein R defaults to 40, and the maximum value is selected as a key point to obtain initial position information of the key point, and the formula is as follows:
，
wherein,front pi R in thermodynamic diagram of kth key point 2 Large set of values, +.>Is an indication Fu Hanshu of the nature of the particular,，for each pixel value the coordinates of the key point are individually assigned, finally, with the highest +.>Value pixel->Coordinates of the key points considered most likely;
(5) It should be noted that, in the embodiment of the present invention, the number of the 32 key points is not limited, and may be set according to the requirement. Specifically, in the embodiment of the invention, 32 points are used, namely, a butterfly saddle point, a nasogen point, an orbital point, an ear point, an upper tooth socket sitting point, a lower tooth socket sitting point, a front chin point, a lower mid-incisor edge point, a middle upper chin incisor edge point, a upper lip protrusion point, a lower nose point, a front chin point of soft tissues, a rear nasal spine, a front nasal spine, a joint point, a craniofacial point, a condylar process center point, a wing point, a nasal vertex, a lowest point of a temporomandibular joint, an inner mandibular point, a middle upper mid-incisor point, a middle upper first permanent molar near cheek point, a middle lower edge point, a middle lower incisor point, a lower jaw rising trailing edge and a lower jaw angle trailing edge.
Step 2: the most accurate detection key points in different quadrants are verified through statistics, the difference value between the detection point and the corresponding template point and the head aspect ratio value is calculated to be smaller than a difference threshold value, three transformation vertexes are selected, and then the coordinates of the template key points are mapped to an image to be detected through calculation of an affine transformation matrix;
(1) And counting and checking three most suitable detection key points in the verification set. In order to ensure that the transformation vertex is a point with accurate detection, firstly, counting the accuracy rate of key points in the verification set and corresponding quadrants, dividing the key points into four quadrants by taking the midpoint of the verification set skull as the origin, and sequencing the key points from high accuracy rate to low accuracy rate; secondly, judging whether three points with highest accuracy are in different quadrants or not, and if not, selecting the point with high secondary accuracy until the three selected points are in different quadrants; in order to ensure that the points detected by the model can be used as transformation vertexes again, the width ratio and the height ratio of the transformation vertexes of the three templates to the head where the template is positioned are calculated respectively, and the width ratio and the height ratio of the corresponding detection points to the head where the template is positioned are calculated; and finally, respectively calculating the difference value of the two width ratios and the height ratio, if the difference value is larger than 0.05, replacing the point which does not meet the standard, and selecting the point with high next accuracy rate until three transformation vertexes which meet the conditions are selected.
(2) And calculating the position transformation relation between the template key points and the corresponding template key points by using an affine transformation matrix algorithm. The three transformed vertex coordinates are respectively recorded as、、Secondly, the corresponding 32 keys are needed to be made on one skull side position sliceMarking the positions of points, wherein the sheet is marked as a template sheet, and the coordinates of the corresponding three points are marked as +.>、、And correspondingly establishing an equation set according to the same standard point, wherein the equation set comprises the following steps:
，
matrix-representing equation set (2) as:
，，，；
wherein K is affine transformation matrix, and error term is obtained by least square methodThe sum of squares of (2) is minimized, and the +.>。
(3) Mapping the template key points into the image to be detected, namely substituting the position information of the rest key points into an affine transformation matrix, and obtaining the position information of the template key points in the detected image.
Step 3: the method comprises the steps of clustering the images by using a density-based clustering algorithm, and respectively counting detection points which are in and not in super pixels where corresponding template points are located. Judging whether the detection point is in the super-pixel area where the corresponding template point is located, wherein the specific mode is as follows: judging whether the key points detected by the model are in the clustering area where the corresponding template key points are. And if the key points are not in the corresponding clustering area, carrying out regression voting on the clustering area where the template points are located by using the local thermodynamic diagram and the offset diagram information, and updating the position information of the detection key points again. Because of the existence of the transformation vertex, at least three detection points can be ensured to be positioned in the super pixel where the corresponding template point is positioned, and the point which is not positioned in the super pixel is judged in the next step.
(1) The method for carrying out super-pixel segmentation on the image is not limited in the embodiment of the invention, and a clustering algorithm DBSCAN based on density can be used. Firstly, converting the color space of an image into a Lab color space from an RGB color space, secondly, calculating the color distance and the space distance on the Lab color space one by utilizing the L, a and b values of the pixels and corresponding X, Y coordinates and other pixels, and then clustering the pixels by utilizing the color distance and the space distance of each pixel through a DBSCAN algorithm, and merging similar pixels into super pixels. Note that the Lab color space is a color-opposing space, where the dimension L represents luminance and a and b represent color-opposing dimensions. The RGB color space is any color space based on an RGB color model, wherein the RGB color model is based on three basic colors of red, green and blue, and is overlapped to different degrees to generate rich and wide colors, so the RGB color space is commonly called as a three-primary color mode. It should be noted that the DBSCAN algorithm is a representative density-based clustering algorithm, which defines clusters as the largest set of points connected by density, can divide a region with a sufficiently high density into clusters, and can find clusters of arbitrary shape in a noisy spatial database.
(2) In the embodiment of the invention, whether the key point is in the super-pixel area where the corresponding template point is located is judged. It should be noted that, a super pixel is a set of pixels with a certain similarity in color and texture within a certain area. And therefore, taking the super-pixel section where the template point corresponding to the key point is located as a judging standard of whether the point is located in the area, if the key point is located in the super-pixel section, determining the point as the position of the key point, otherwise, performing the next judgment.
Step 4: selecting a local regression voting area by combining three major interest points with super pixels, selecting template points which are positioned in non-adjacent quadrants and are closest to two detection points in the step 3 as neighborhood transformation vertexes, obtaining a neighborhood mapping point rm of the point according to the coordinate ratio of the template points which are corresponding to the non-standard points to the adjacent two transformation vertexes, taking the super pixel area where the domain mapping point rm, the template point m and the detection point d are positioned as the local voting area, calculating local regression votes by using thermodynamic diagrams in the local area and offset diagrams corresponding to the template points as the key points, and obtaining the final key point position by selecting the maximum value, wherein the formula is expressed as follows:
，
wherein,front pi Q in local thermodynamic diagram being kth key point 2 A large set of values is provided for the set of values,
h and W are the height and width of the image, respectively, and the local area LA is formulated as follows:
，
wherein,DRthe regions are partitioned for the super-pixels,indicated as the super pixel area where the detection point is located,expressed as the super-pixel area where the template point is located,Represented as a super-pixel region where the neighborhood point is located,the super pixel region where the points not within the three super pixel region among the three points z, m, r are indicated.
In the embodiment of the invention, the key points which do not accord with the standard are subjected to local regression voting in the corresponding super pixels, and the final key point positions are obtained. It should be noted that, in the full graph, the previous pi R in the thermodynamic diagram is selected in the same manner as in steps 1-4 2 The difference in the regression voting process is that global information is not used in this step, and only local information in the super-pixel area corresponding to the point is used for key positioning, namely thermodynamic diagram information of all points in the super-pixel corresponding to the point is selected, and the steps 1-4 are repeated.
According to the invention, firstly, the pixel-by-pixel regression voting is calculated by combining the thermodynamic diagram and the offset diagram, the global information is utilized to obtain the key points, the accuracy of positioning the key points is improved, the positions of the key points of the template and the limit of the super-pixel segmentation distance threshold based on density are utilized, the key points which do not meet the distance requirement can be limited near the real positions, and then the local information is utilized to carry out regression voting to position detection points, so that the robustness of the detection of the key points is improved, and the method is suitable for the skull side plates of more factories. Compared with the traditional manual fixed point, the method has the advantages of higher speed and higher efficiency, and compared with the traditional key point detection method, the method has higher accuracy and stronger robustness.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (6)
1. The key point detection method of the skull side position slice is characterized by comprising the following steps of:
step 1: inputting the skull side position slice to a detection model to obtain initial position information of key points;
step 2: the most accurate detection key points in different quadrants are collected through statistics verification, the width ratio and the height ratio of the detection points to the head where the detection points are located, the width ratio and the height ratio of the template points to the head where the detection points are located are calculated, the difference value of the width ratio and the height ratio is calculated, if the difference value is smaller than a difference threshold value, a transformation vertex is selected, and then the coordinates of the template key points are mapped to an image to be detected through calculation of an affine transformation matrix, so that the position information of the template key points in the detected image can be obtained;
step 3: performing superpixel segmentation on an image to be detected by using a DBSCAN-based method, respectively counting detection points which are in and out of the superpixels where corresponding template points are located, and performing next judgment on the points which are not in the superpixels;
step 4: selecting a local regression voting area by combining three major interest points with super pixels, selecting template points which are positioned in non-adjacent quadrants and are closest to two detection points in the detection points in accordance with the step 3 as neighborhood transformation vertexes, obtaining a neighborhood mapping point rm of the point according to the coordinate ratio of the template points which are corresponding to the non-standard points and the adjacent two transformation vertexes, taking the super pixel area where the domain mapping point rm, the template point m and the detection point d are positioned as the local voting area, calculating local regression votes by using a thermodynamic diagram in the local voting area and an offset diagram corresponding to the thermodynamic diagram, and selecting the maximum value as a key point to obtain the final key point position;
in step 1, the initial position information of the key point is obtained by: converting the skull side position slice to be detected into a set size, and then inputting the size into a model to obtain thermodynamic diagram information and offset diagram information of a current image; by utilizing a key point detection method based on attention feature pyramid fusion and regression voting, pi is performed before the thermodynamic diagram by traversing the thermodynamic diagram of each layer of key pointsR 2 Computing regression votes for large points and their corresponding offset mapsR in the rule is defaulted to 40, the maximum value is selected as a key point, and initial position information of the key point is obtained, wherein the formula is as follows:
(1)，
wherein,front pi in thermodynamic diagram that is the kth key pointR 2 Large set of values, +.>Is an indication Fu Hanshu of the nature of the particular,，coordinates of key points for each pixel value alone, with highest +.>Value pixelCoordinates of the key points considered most likely;
in step 4, the formula for obtaining the final key point position is expressed as follows:
(2)，
wherein,front pi Q in local thermodynamic diagram being kth key point 2 Large set of values, q= (LA/H W) ×piR 2 H and W are the height and width of the image, respectively, LA being denoted as local area;
the local area LA formula is expressed as follows:
(3)，
wherein,DRthe regions are partitioned for the super-pixels,denoted by the super pixel area where the detection point is located, ">Expressed as the super-pixel area where the template point is located,Represented as a super-pixel region where the neighborhood point is located,the super pixel region where the points not within the three super pixel region among the three points z, m, r are indicated.
2. The method for detecting keypoints of a skull side-slice according to claim 1, wherein in step 2, the mapping the template keypoint coordinates to the image to be detected comprises: mapping a preset skull side piece key point template into a coordinate system of the current skull side piece, and extracting position information of a template key point; counting the most suitable three detection key points in the verification set, and solving the position transformation relation of the corresponding template key points by utilizing an affine transformation matrix algorithm; and inputting the position information of the rest template key points into the position transformation relation to obtain the affine transformation coordinates of the template key points.
3. The method for detecting key points of a skull side-position slice according to claim 1, wherein in step 3, a minimum of three detection points can be ensured to be in super-pixels where corresponding template points are located due to the existence of transformation vertices.
4. The method for detecting keypoints of a skull side-position slice according to claim 1, wherein in step 3, the super-pixel segmentation of the image to be detected comprises clustering the image using a density-based clustering algorithm.
5. The method for detecting the key points of the skull side position slice according to claim 1, wherein in the step 3, whether the detection points are in the super-pixel area where the corresponding template points are located is determined by the following specific steps: judging whether the key points detected by the model are in the clustering area where the corresponding template key points are.
6. The method for detecting key points of a skull side position slice according to claim 1, wherein in step 3, the next judgment is made for the points not in the super pixel, and the method is as follows: and carrying out local regression voting in the corresponding super pixels to obtain the positions of the key points, and if the key points which are not in the corresponding clustering areas exist, carrying out regression voting on the clustering areas where the template points are positioned by using the local thermodynamic diagram and the offset diagram information, and updating the position information of the detection key points again.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311651546.9A CN117372425B (en) | 2023-12-05 | 2023-12-05 | A method for detecting key points in lateral cephalograms |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311651546.9A CN117372425B (en) | 2023-12-05 | 2023-12-05 | A method for detecting key points in lateral cephalograms |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117372425A CN117372425A (en) | 2024-01-09 |
| CN117372425B true CN117372425B (en) | 2024-03-19 |
Family
ID=89398773
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311651546.9A Active CN117372425B (en) | 2023-12-05 | 2023-12-05 | A method for detecting key points in lateral cephalograms |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117372425B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119180954B (en) * | 2024-09-02 | 2025-05-06 | 重庆邮电大学 | A cervical vertebrae segmentation and key point detection method based on diffusion model |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109166133A (en) * | 2018-07-14 | 2019-01-08 | 西北大学 | Soft tissue organs image partition method based on critical point detection and deep learning |
| CN109977767A (en) * | 2019-02-18 | 2019-07-05 | 浙江大华技术股份有限公司 | Object detection method, device and storage device based on super-pixel segmentation algorithm |
| CN110706203A (en) * | 2019-09-06 | 2020-01-17 | 成都玻尔兹曼智贝科技有限公司 | Method and system for automatically detecting key points of lateral skull tablets based on deep learning |
| CN114219897A (en) * | 2021-12-20 | 2022-03-22 | 山东大学 | A method and system for predicting orthodontic results based on feature point recognition |
| WO2022087706A1 (en) * | 2020-10-29 | 2022-05-05 | Botica Comercial Farmacêutica Ltda. | Method for detecting and segmenting the lip region |
| CN114581535A (en) * | 2022-03-03 | 2022-06-03 | 北京深光科技有限公司 | Method, device, storage medium and equipment for marking key points of user bones in image |
| CN114820517A (en) * | 2022-04-26 | 2022-07-29 | 杭州隐捷适生物科技有限公司 | System and method for automatic detection of key points in lateral cranial radiographs based on deep learning |
| WO2022188547A1 (en) * | 2021-03-09 | 2022-09-15 | 杭州朝厚信息科技有限公司 | Method for detecting key point of x-ray cephalogram image |
| CN115797730A (en) * | 2023-01-29 | 2023-03-14 | 有方（合肥）医疗科技有限公司 | Model training method and device, and head shadow measurement key point positioning method and device |
| CN116309302A (en) * | 2023-01-09 | 2023-06-23 | 四川大学 | A Method for Extracting Key Points from Cephalometric Films |
| CN116485809A (en) * | 2022-07-01 | 2023-07-25 | 山东财经大学 | Tooth instance segmentation method and system based on self-attention and receptive field adjustment |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| ES2384928B1 (en) * | 2010-06-25 | 2013-05-20 | Telefónica, S.A. | METHOD AND SYSTEM FOR THE QUICK AND ROBUST IDENTIFICATION OF SPECIFIC PRODUCTS IN IMAGES. |
| US10198858B2 (en) * | 2017-03-27 | 2019-02-05 | 3Dflow Srl | Method for 3D modelling based on structure from motion processing of sparse 2D images |
| CN112950508B (en) * | 2021-03-12 | 2022-02-11 | 中国矿业大学(北京) | A computer vision-based video data restoration method for drainage pipelines |
-
2023
- 2023-12-05 CN CN202311651546.9A patent/CN117372425B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109166133A (en) * | 2018-07-14 | 2019-01-08 | 西北大学 | Soft tissue organs image partition method based on critical point detection and deep learning |
| CN109977767A (en) * | 2019-02-18 | 2019-07-05 | 浙江大华技术股份有限公司 | Object detection method, device and storage device based on super-pixel segmentation algorithm |
| CN110706203A (en) * | 2019-09-06 | 2020-01-17 | 成都玻尔兹曼智贝科技有限公司 | Method and system for automatically detecting key points of lateral skull tablets based on deep learning |
| WO2022087706A1 (en) * | 2020-10-29 | 2022-05-05 | Botica Comercial Farmacêutica Ltda. | Method for detecting and segmenting the lip region |
| WO2022188547A1 (en) * | 2021-03-09 | 2022-09-15 | 杭州朝厚信息科技有限公司 | Method for detecting key point of x-ray cephalogram image |
| CN114219897A (en) * | 2021-12-20 | 2022-03-22 | 山东大学 | A method and system for predicting orthodontic results based on feature point recognition |
| CN114581535A (en) * | 2022-03-03 | 2022-06-03 | 北京深光科技有限公司 | Method, device, storage medium and equipment for marking key points of user bones in image |
| CN114820517A (en) * | 2022-04-26 | 2022-07-29 | 杭州隐捷适生物科技有限公司 | System and method for automatic detection of key points in lateral cranial radiographs based on deep learning |
| CN116485809A (en) * | 2022-07-01 | 2023-07-25 | 山东财经大学 | Tooth instance segmentation method and system based on self-attention and receptive field adjustment |
| CN116309302A (en) * | 2023-01-09 | 2023-06-23 | 四川大学 | A Method for Extracting Key Points from Cephalometric Films |
| CN115797730A (en) * | 2023-01-29 | 2023-03-14 | 有方（合肥）医疗科技有限公司 | Model training method and device, and head shadow measurement key point positioning method and device |
Non-Patent Citations (3)
| Title |
|---|
| 基于Rényi熵的互补尺度空间关键点医学图像配准;张少敏;支力佳;赵大哲;赵宏;林树宽;;仪器仪表学报;20101115(第11期);全文 * |
| 朱锴 ; 陶攀 ; 付忠良 ; 陈晓清 ; .利用关键点检测算法的超声图像定点测量.哈尔滨工业大学学报.2018,(第11期),全文. * |
| 潘林,胡遒生,鲍旭东,王林,周光泉.基于模板匹配的头颅侧位片标志点的自动识别.临床口腔医学杂志.2003,(第11期),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117372425A (en) | 2024-01-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yao et al. | Automatic localization of cephalometric landmarks based on convolutional neural network | |
| CN103310457B (en) | A kind of pulmonary parenchyma dividing method based on para-curve correction convex closure | |
| CN106780591B (en) | A Craniofacial Morphology Analysis and Facial Restoration Method Based on Craniofacial Dense Corresponding Point Cloud | |
| CN114511567B (en) | Tongue body and tongue coating image identification and separation method | |
| CN115272359B (en) | Tooth segmentation method, tooth surface data acquisition method and oral CT (computed tomography) equipment | |
| CN110569784B (en) | Human body size measuring method and system, storage medium and electronic equipment | |
| CN112308895B (en) | Method for constructing realistic dentition model | |
| JP2009095644A (en) | Image recognition result determination apparatus, method, and program | |
| CN107240128B (en) | X-ray and color photo registration method based on contour features | |
| CN112807108B (en) | A method for detecting the orthodontic state of teeth during orthodontic treatment | |
| CN112349391A (en) | Optimized rib automatic labeling method | |
| CN117372425B (en) | A method for detecting key points in lateral cephalograms | |
| CN115252174B (en) | Tooth arrangement method, tooth orthodontic simulation method, storage medium and oral CT device | |
| US20250009483A1 (en) | System and Method for Aligning 3-D Imagery of a Patient's Oral Cavity in an Extended Reality (XR) Environment | |
| CN113197549A (en) | System for diagnosing diseases through face recognition technology | |
| CN115713590A (en) | Three-dimensional reconstruction image processing method and system based on CT | |
| TW202103630A (en) | Multi-view mammogram analysis method, multi-view mammogram analysis system, and non-transitory computer-readable medium | |
| CN117618110A (en) | A marker-free surgical navigation method and system based on 3D structured light | |
| CN108447044A (en) | A kind of osteomyelitis lesions analysis method based on medical figure registration | |
| Zhang et al. | Dense correspondence of skull models by automatic detection of anatomical landmarks | |
| CN111612795A (en) | A method for automatic extraction of feature points from the digital model of preoperative nasal-alveolar appliance | |
| CN111091895B (en) | Breast image analysis method, system and non-transitory computer readable medium | |
| CN114549749B (en) | A structured three-dimensional face digital template and production method | |
| WO2024124790A1 (en) | Skeleton sagittal balance parameter measurement system | |
| CN103985113B (en) | Tongue is as dividing method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |