A method for classifying oracle bone characters based on key point detection

Abstract

The application discloses a kind of oracle bone characters classification method based on key point detection, it is related to image recognition technical field. Including: obtaining oracle bone characters and copy characters, according to key point definition rule, oracle bone characters are labeled, obtain oracle bone characters labeling data set and copy characters labeling data set, after obtaining oracle bone characters image key point, compare and classify with copy characters key point.The present application is helpful to realize the automatic identification and classification of oracle bone characters, solve the problem of oracle bone character recognition depending on artificial determination, and the big difference of heterograph, more universal.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method for classifying oracle bone script characters based on key point detection. Background Technology

[0002] In oracle bone script research, the identification of oracle bone characters is an important task. However, due to the unique characteristics of oracle bone characters, their automatic identification faces a series of problems (such as...). Figure 5 All oracle bone characters are variant forms of the character "dragon," but they differ significantly in form: 1) Most oracle bone characters have variant forms, and these variant forms vary greatly; 2) Oracle bone character images are too noisy, resulting in low accuracy in full-character recognition; 3) Oracle bone character recognition heavily relies on expert interpretation experience, is highly subjective, and currently lacks an effective and quantifiable automatic recognition method. This patent uses key point detection technology to solve the above problems.

[0003] Therefore, proposing a key point detection-based method for classifying oracle bone characters to solve the existing technical problems is an urgent issue that needs to be addressed by those skilled in the art. Summary of the Invention

[0004] In view of this, the present invention provides a method for classifying oracle bone characters based on key point detection, which can achieve a highly stable and balanced effect.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for classifying oracle bone script characters based on key point detection includes the following steps:

[0007] Data acquisition steps: Obtain oracle bone characters cut out from rubbings of oracle bones and tracings copied by experts;

[0008] Preprocessing steps: Using the tracing characters as the base characters, annotate the tracing characters according to the key point definition rules to obtain the base character annotation dataset. Then, randomly select 50 oracle bone characters for each character shape based on the base characters and annotate them according to the key point definition rules to obtain the oracle bone character annotation dataset.

[0009] Model training steps: Use the oracle bone script annotation dataset as training samples to train the oracle bone script key point deep learning network model, and obtain the trained oracle bone script key point deep learning network model.

[0010] Prediction calculation steps: Input the oracle bone character without key points into the trained oracle bone character key point deep learning network model, and obtain the key points of the oracle bone character image through the model prediction;

[0011] The classification calculation steps are as follows: The similarity between the key points of the predicted oracle bone character image and the key points in the baseline character annotation dataset is calculated, and the result with the highest similarity is taken. The oracle bone character corresponding to this result belongs to the same category as the baseline character.

[0012] The above method, optionally, includes the following labeling rules in the preprocessing steps:

[0013] 1) Two points define a line: Take one point at each end of a straight line segment and define it as an endpoint;

[0014] 2) Three points define an arc: the two ends of the arc are the endpoints, and the point between the two endpoints is defined as the midpoint;

[0015] 3) Four points define a circle: Take one point each at the top, bottom, left, and right of a circle (treat an approximate circle as a circle), and define each point as a circle point;

[0016] 4) Four points define a quadrilateral: Take the intersection points of the sides of the quadrilateral, and define each point as a corner point;

[0017] 5) Add a dot when encountering intersections: The point where the strokes of a character intersect is defined as the intersection point;

[0018] 6) Add a dot at the fold line: The point where the strokes of a character turn is defined as the fold point;

[0019] 7) Complex characters have their key points determined by oracle bone script experts, and these key points are defined as special points;

[0020] 8) If any of the points defined above overlap, then one of them shall be taken.

[0021] Optionally, the image similarity calculation in the classification step of the above method includes: keypoint quantity similarity and keypoint distribution similarity.

[0022] The above method can optionally include a keypoint similarity S. num Represented as:

[0023]

[0024] Where, N A and N B These represent the number of keypoints in image A and image B, respectively.

[0025] The above method, optionally, uses polar coordinates to calculate the keypoint distribution similarity, which can be expressed as:

[0026]

[0027] Among them, S dis Let k be the number of intervals and d be the similarity of keypoint distributions. i The formula for calculating the density difference of key points is as follows:

[0028]

[0029] in, and Let be the number of keypoints in the i-th interval of each image.

[0030] Optionally, in the above method, the keypoint similarity Sim in the classification step can be expressed as:

[0031] Sim=α·S num +β·S dis ;

[0032] Here, α and β are the score weights for the similarity of keypoint quantity and keypoint distribution, respectively, and α+β=1.

[0033] Optionally, in the above method, during the classification step, oracle bone characters with similar shapes are grouped into one category, and the oracle bone characters in each category are variants of each other.

[0034] As can be seen from the above technical solution, compared with the prior art, the present invention provides a method for classifying oracle bone characters based on key point detection, which has the following beneficial effects: Compared with the prior art, the above method can realize the automatic recognition and classification of oracle bone characters, and has the advantages of good adaptability, accurate recognition, and convenient classification. The reason for these advantages is that key points record the shape framework of oracle bone characters and have recognizability, effectively solving the problems of relying on manual judgment, large differences in variant characters, and difficulty in training deep learning frameworks in oracle bone character recognition. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0036] Figure 1 This is a flowchart of a key point detection-based oracle bone script classification method disclosed in this invention;

[0037] Figure 2 Here are examples of oracle bone characters to be classified in this invention, where 2a is an image of oracle bone script and 2b is a transcribed character;

[0038] Figure 3 This is a schematic diagram of the reference character annotation points disclosed in this embodiment;

[0039] Figure 4 This is a schematic diagram of the annotation rules disclosed in this invention;

[0040] Figure 5 It is an alternate form of the oracle bone script character "dragon". Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0043] Reference Figure 1 As shown, this invention discloses a method for classifying oracle bone script characters based on key point detection, including the following steps:

[0044] Data acquisition steps: Obtain oracle bone characters cut out from rubbings of oracle bones and tracings copied by experts;

[0045] Preprocessing steps: Using the tracing characters as the base characters, annotate the tracing characters according to the key point definition rules to obtain the base character annotation dataset. Then, randomly select 50 oracle bone characters for each character shape based on the base characters and annotate them according to the key point definition rules to obtain the oracle bone character annotation dataset.

[0046] Model training steps: Use the oracle bone script annotation dataset as training samples to train the oracle bone script key point deep learning network model, and obtain the trained oracle bone script key point deep learning network model.

[0047] Prediction calculation steps: Input the oracle bone character without key points into the trained oracle bone character key point deep learning network model, and obtain the key points of the oracle bone character image through the model prediction;

[0048] The classification calculation steps are as follows: The similarity between the key points of the predicted oracle bone character image and the key points in the baseline character annotation dataset is calculated, and the result with the highest similarity is taken. The oracle bone character corresponding to this result belongs to the same category as the baseline character.

[0049] Specifically, refer to Figure 2 As shown (where 2a is an image of oracle bone script and 2b is a transcribed character), in the data acquisition step, if the collected oracle bone script is determined to be an image type, then denoising, binarization, and edge detection operations are performed to obtain a binary image, thus obtaining the oracle bone script. If the collected oracle bone script is determined to be a transcribed character, then the oracle bone script is directly obtained.

[0050] The deep learning network models used in the model training process can be referenced, but are not limited to, YOLO, Faster R-CNN, SSD, etc.

[0051] Furthermore, in the preprocessing step, refer to Figure 3 and Figure 4 As shown, the labeling rules in the preprocessing steps are as follows:

[0052] 1) Two points define a line: Take one point at each end of a straight line segment and define it as an endpoint;

[0053] 2) Three points define an arc: the two ends of the arc are the endpoints, and the point between the two endpoints is defined as the midpoint;

[0054] 3) Four points define a circle: Take one point each at the top, bottom, left, and right of a circle (treat an approximate circle as a circle), and define each point as a circle point;

[0055] 4) Four points define a quadrilateral: Take the intersection points of the sides of the quadrilateral, and define each point as a corner point;

[0056] 5) Add a dot when encountering intersections: The point where the strokes of a character intersect is defined as the intersection point;

[0057] 6) Add a dot at the fold line: The point where the strokes of a character turn is defined as the fold point;

[0058] 7) Complex characters have their key points determined by oracle bone script experts, and these key points are defined as special points;

[0059] 8) If any of the points defined above overlap, then one of them shall be taken.

[0060] Furthermore, the image similarity calculation in the classification step includes: keypoint quantity similarity and keypoint distribution similarity.

[0061] Furthermore, the key point quantity similarity S num Represented as:

[0062]

[0063] Where, N A and NB These represent the number of keypoints in image A and image B, respectively.

[0064] Specifically, if S num If the score is 1, then the two images have the same number of keypoints; if S num If the score is less than 1, the number of keypoints in the two images is different.

[0065] Furthermore, the keypoint distribution similarity is implemented based on polar coordinates, and can be expressed as:

[0066]

[0067] Among them, S dis Let k be the number of intervals and d be the similarity of keypoint distributions. i For keypoint density variation, d i The calculation formula is as follows:

[0068]

[0069] in, and Let be the number of keypoints in the i-th interval of each image.

[0070] Specifically, each key point is represented in polar coordinates as (r i θ i The method involves sorting all keypoints according to their polar angle θ and calculating the distribution density of keypoints in two images at their respective polar angles. Specifically, within a given bandwidth, the polar angle distribution density is calculated with a certain step size. For example, if the bandwidth length is k and the step size is Δθ, the keypoints can be divided into... There are several intervals; for each interval, the number of keypoints located within that interval in both images is calculated (corresponding to the polar angle distribution density). and Using the inverse of the keypoint density difference, the smaller the density difference, the more similar the keypoint distributions, and the higher the similarity score.

[0071] Furthermore, the keypoint similarity Sim in the classification step is represented as:

[0072] Sim=α·S num +β·S dis .

[0073] Here, α and β are the score weights for the similarity of keypoint quantity and keypoint distribution, respectively, and α+β=1.

[0074] Specifically, the values ​​of α and β were determined experimentally.

[0075] Furthermore, in the calculation and classification steps, oracle bone characters with similar shapes are grouped into one category, and the oracle bone characters in each category are variants of each other.

[0076] Specifically, in this embodiment of the invention, the invention is not limited to oracle bone characters on rubbings of oracle bones; other forms of oracle bone characters can also use the same method, including but not limited to image oracle bone characters.

[0077] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for classifying oracle bone script characters based on key point detection, characterized in that, Includes the following steps: Data acquisition steps: Obtain oracle bone characters cut out from rubbings of oracle bones and tracings copied by experts; Preprocessing steps: Using the tracing characters as the base characters, annotate the tracing characters according to the key point definition rules to obtain the base character annotation dataset. Then, randomly select 50 oracle bone characters for each character shape based on the base characters and annotate them according to the key point definition rules to obtain the oracle bone character annotation dataset. Model training steps: Use the oracle bone script annotation dataset as training samples to train the oracle bone script key point deep learning network model, and obtain the trained oracle bone script key point deep learning network model. Prediction calculation steps: Input the oracle bone character without key points into the trained oracle bone character key point deep learning network model, and obtain the key points of the oracle bone character image through the model prediction; The classification calculation steps are as follows: The similarity between the key points of the predicted oracle bone character image and the key points in the baseline character annotation dataset is calculated, and the result with the highest similarity is taken. The oracle bone character corresponding to this result belongs to the same category as the baseline character. The image similarity calculation in the classification step includes: keypoint quantity similarity and keypoint distribution similarity; Key point quantity similarity Represented as: ， in, and These represent the number of keypoints in image A and image B, respectively. Keypoint distribution similarity is implemented based on polar coordinates, and is represented as follows: ， in, Let k be the number of intervals and d be the similarity of keypoint distributions. i For keypoint density variation, d i The calculation formula is as follows: ， in, and Let be the number of keypoints in the i-th interval of the two images, respectively; Key point similarity Represented as: ； in, and These are the score weights for the similarity in the number of key points and the similarity in the distribution of key points, respectively. .

2. The oracle bone script classification method based on key point detection according to claim 1, characterized in that, The definition rules for key points in the preprocessing step are as follows: 1) Two points define a line: Take one point at each end of a straight line segment and define it as an endpoint; 2) Three points define an arc: the two ends of the arc are the endpoints, and the point between the two endpoints is defined as the midpoint; 3) Four points define a circle: Take one point each at the top, bottom, left, and right of the circle, and define each point as the circle point; 4) Four points define a quadrilateral: Take the intersection points of the sides of the quadrilateral, and define each point as a corner point; 5) Add a dot when encountering intersections: The point where the strokes of a character intersect is defined as the intersection point; 6) Add a dot at the fold line: The point where the strokes of a character turn is defined as the fold point; 7) For complex characters, the key points are determined by oracle bone script experts, and these key points are defined as special points; 8) If any of the points defined above overlap, then one of them shall be taken.

3. The oracle bone script classification method based on key point detection according to claim 1, characterized in that, In the classification process, oracle bone characters with similar shapes are grouped together, and the oracle bone characters in each group are variants of each other.

Images