Facial wrinkle three-dimensional evaluation method and system based on gaussian process and random transformation

Abstract

The application discloses a kind of based on Gaussian process and random transform facial wrinkle three-dimensional evaluation method and system, it is related to facial wrinkle detection field, first, based on self-defined neural network model in facial image positioning facial feature and facial contour, according to the positioning result, remove facial feature, background from facial image;And combining Gaussian process and random transform algorithm, realize facial wrinkle detection;Finally, according to the three-dimensional body density of facial wrinkle calculated by the wrinkle evaluation algorithm proposed in the application.The application can detect and measure wrinkle length, width, depth, quantity in two-dimensional facial image, and calculate three-dimensional wrinkle body density by this, so as to realize the purpose of three-dimensional evaluation of facial wrinkle of subject.

Description

Technical Field

[0001] This invention relates to the field of facial wrinkle detection, and in particular to a three-dimensional evaluation method and system for facial wrinkles based on Gaussian processes and random transformations. Background Technology

[0002] In recent years, facial wrinkle detection has been applied in several areas: first, in related face recognition, expression recognition, and expression synthesis; second, in facial retouching, which is widely used in the media and entertainment industries to eliminate facial wrinkles and blemishes, making the skin look beautiful and smooth; and third, in age estimation, age simulation, and aging recognition. Facial wrinkle detection can provide some basic information for the above-mentioned related application research; therefore, facial wrinkle detection has great research significance.

[0003] Most facial wrinkle detection methods treat wrinkles as skin texture, but they are completely different. Skin texture is innate, uniform, and repetitive, and has a low correlation with physiological age. Wrinkles, on the other hand, are caused by skin aging and long-term facial muscle contraction, and are acquired. Therefore, wrinkles are curved objects with certain intensity and geometric constraints; their length, width, depth, and distribution rate are all closely related to the degree of skin aging.

[0004] Currently, the main methods for facial wrinkle detection include: First, a stochastic wrinkle generation model using Markov processes. This model has the advantage of good localization results, but requires significant computation time, and the algorithm's accuracy depends on the position of the initial line segments. To improve this, a second detection method, a deterministic method based on image morphology, was proposed. This method achieves rapid localization of facial wrinkles, and its accuracy and computational cost are significantly better than the stochastic wrinkle generation model using Markov processes. However, for images with rough skin or high resolution, there is a lot of interference, leading to poor wrinkle feature extraction and a low detection rate. A third method uses a hybrid Hessian filter to detect wrinkles. This method is only suitable for detecting rough wrinkles in the horizontal direction, thus having significant limitations. Based on this, a line-tracking method based on the original hybrid Hessian filter was proposed. Compared with the above methods, the detection rate is improved, but the problem of low detection rate for fine lines has not been completely solved.

[0005] Besides the facial wrinkle detection methods mentioned above, how to scientifically and effectively evaluate wrinkles is also a technical challenge. Currently, wrinkle evaluation methods can be divided into clinical evaluation methods (such as photo grading and descriptive grading methods) and non-clinical evaluation methods (such as the wrinkle index evaluation method). Clinical evaluation methods are convenient to operate and widely applicable, but most are manual methods, requiring physicians to have high levels of professional knowledge and rich diagnostic experience. The final qualitative conclusions can vary from person to person, raising questions about accuracy and objectivity. Non-clinical evaluation methods mainly refer to measuring wrinkle morphology using detection instruments or software. Currently, commonly used wrinkle measurement instruments include ultrasound diagnostic instruments and laser surface photometers. In summary, compared to clinical evaluation methods, non-clinical evaluation methods play an important role in assessing the effectiveness of wrinkle treatment because they can provide relatively objective data. However, non-clinical evaluation methods, such as the wrinkle index evaluation method, ignore the influence of wrinkle width and number on the measurement of wrinkle severity, and do not consider that different image resolutions and face sizes can lead to significant variations in the wrinkle index. Furthermore, the wrinkle depth in this method is the filtered response value extracted from the image features during the detection process. This value is only a relative depth and cannot objectively reflect the actual depth of the wrinkle. In summary, this method uses wrinkle depth and wrinkle length on the imaging surface for two-dimensional evaluation of wrinkles, without conducting three-dimensional evaluation. Summary of the Invention

[0006] To address the issues of low fine line detection rate and the inability to objectively evaluate facial wrinkles using the detection results, this invention provides a three-dimensional evaluation method and system for facial wrinkles based on Gaussian processes and random transformations.

[0007] To achieve the above objectives, the present invention provides the following solution:

[0008] This invention provides a three-dimensional evaluation method for facial wrinkles based on Gaussian processes and stochastic transformations, comprising:

[0009] Read an original frontal face image of a subject and standardize the original frontal face image to obtain a standard frontal face image.

[0010] Based on a custom neural network model, facial features and facial contours are located in a standard frontal face image to obtain facial feature coordinate data and facial contour coordinate data.

[0011] Based on the facial feature coordinate data and facial contour coordinate data, the facial feature area and background area are removed from the frontal standard face image to obtain the facial skin image.

[0012] Based on the Gaussian process and stochastic transformation algorithm, facial wrinkles are detected on a facial skin map, and wrinkle detection result map is obtained.

[0013] Extract facial position information from the original frontal face image, and crop the face region from the wrinkle detection result image based on the facial position information;

[0014] Image processing is performed on the face region to obtain the wrinkle region, and the number of wrinkles in the wrinkle region, as well as the length, width and depth of each wrinkle, are determined.

[0015] Based on the number of wrinkles in the wrinkle area and the length, width and depth of each wrinkle, calculate the wrinkle volume level of the whole face, and calculate the three-dimensional wrinkle volume density based on the wrinkle volume level of the whole face and the face area.

[0016] Based on the three-dimensional wrinkle density, the facial wrinkles of the subjects were evaluated in three dimensions.

[0017] This invention also provides a three-dimensional evaluation system for facial wrinkles based on Gaussian processes and random transformations, comprising:

[0018] The image standardization processing module is used to read an original frontal face image of a subject and standardize the original frontal face image to obtain a standard frontal face image.

[0019] The coordinate data extraction module is used to locate facial features and facial contours in a standard frontal face image based on a custom neural network model, and obtain facial feature coordinate data and facial contour coordinate data.

[0020] The facial skin map determination module is used to remove the facial features and background areas from the standard frontal face image based on the facial feature coordinate data and facial contour coordinate data to obtain the facial skin map.

[0021] The wrinkle detection result image determination module is used to perform facial wrinkle detection on facial skin images based on Gaussian process and random transformation algorithm, and obtain wrinkle detection result images;

[0022] The face region cropping module is used to extract the face position information from the original frontal face image and crop the face region from the wrinkle detection result image based on the face position information.

[0023] The wrinkle information determination module is used to perform image processing on the face region to obtain the wrinkle region, and determine the number of wrinkles in the wrinkle region as well as the length, width and depth of each wrinkle.

[0024] The 3D wrinkle density calculation module is used to calculate the number of wrinkle levels on the whole face based on the number of wrinkles in the wrinkle area and the length, width and depth of each wrinkle, and to calculate the 3D wrinkle density based on the number of wrinkle levels on the whole face and the face area.

[0025] The three-dimensional evaluation module is used to evaluate the facial wrinkles of the subject in three dimensions based on the three-dimensional wrinkle density.

[0026] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0027] To address the main shortcomings of existing wrinkle detection and evaluation technologies, this invention provides a three-dimensional facial wrinkle evaluation method and system based on Gaussian processes and random transformations. This invention uses Gaussian processes and random transformation algorithms to quickly locate facial wrinkles and can identify wrinkles in different directions. It can also detect wrinkles when their direction changes, thus avoiding wrinkle discontinuities. This invention achieves a more scientific evaluation of facial wrinkles by calculating their three-dimensional volume density. Specifically, it involves low computational cost, high wrinkle detection rate, and the ability to identify wrinkles with different directional trends. Furthermore, it fills the gap in current methods for evaluating wrinkles by using the three-dimensional volume density of facial wrinkles. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 A flowchart illustrating the three-dimensional evaluation method for facial wrinkles based on Gaussian processes and random transformations provided in this embodiment of the invention;

[0030] Figure 2 This is a schematic diagram of the colored area provided in an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of the wrinkle detection effect provided in an embodiment of the present invention;

[0032] Figure 4 A schematic diagram showing the positions of the highest, lowest, leftmost, and rightmost points of a face provided in an embodiment of the present invention;

[0033] Figure 5 This is a schematic diagram of the face region cropping result provided in an embodiment of the present invention;

[0034] Figure 6 This is a diagram illustrating the specific steps involved in removing non-wrinkle areas according to an embodiment of the present invention.

[0035] Figure 7 This is a schematic diagram illustrating the effect of removing non-wrinkle areas provided in an embodiment of the present invention;

[0036] Figure 8 A schematic diagram of four directions for calculating curvature provided for embodiments of the present invention;

[0037] Figure 9 The system flowchart of the three-dimensional evaluation method for facial wrinkles based on Gaussian process and random transformation provided in the embodiments of the present invention is shown. Detailed Implementation

[0038] 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.

[0039] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0040] Example 1

[0041] like Figure 1 As shown in the figure, this embodiment provides a three-dimensional evaluation method for facial wrinkles based on Gaussian processes and random transformations, including the following steps.

[0042] Step 101: Read an original frontal face image of the subject and standardize the original frontal face image to obtain a standard frontal face image.

[0043] In this embodiment, step 101 specifically includes:

[0044] The original frontal face image is resized to a standard size; the standard size is X pixels wide and Y pixels long. If the original frontal face image is smaller than the standard size, a two-dimensional interpolation algorithm is used to standardize it, resulting in a standard frontal face image. If the original frontal face image is larger than the standard size, a two-dimensional downsampling algorithm is used to standardize it, resulting in a standard frontal face image. If the original frontal face image is equal to the standard size, it is determined as the standard frontal face image.

[0045] Step 102: Based on a custom neural network model, locate facial features and facial contours in a standard frontal face image to obtain facial feature coordinate data and facial contour coordinate data.

[0046] The custom neural network model was obtained through preliminary experiments. The preliminary experimental steps involved inputting a public dataset (e.g., the Helen face dataset) into an effective face parsing hierarchical aggregation network for model training. The trained model was then stored on a cloud server for use in the proposed three-dimensional facial wrinkle evaluation method. The loss function of the face parsing hierarchical aggregation network is shown in equations (1)-(3).

[0047]

[0048]

[0049] L=λ a ·L a +λ b ·L b (3);

[0050] Among them, L a L b Let S represent the coarse segmentation loss and the boundary-aware loss, respectively. S^(e.g.) ) and S (e.g.) Let (i, j) represent the N-channel confidence map and the N-channel ground truth, respectively. Let (i, j) represent the 2D coordinates of the pixel. Based on binary cross-entropy, an appropriate weight ratio γ is assigned to the boundary-aware loss to alleviate the imbalance between foreground and background classes. L represents the total loss function. λ a , λ b The weights used to balance the losses during training are the coarse segmentation loss and the boundary-aware loss. After calling the model, the results returned by the model are facial feature coordinate data and facial contour coordinate data.

[0051] Step 103: Based on the facial feature coordinate data and facial contour coordinate data, remove the facial feature area and background area from the frontal standard face image to obtain the facial skin image.

[0052] In this embodiment, step 103 specifically includes:

[0053] Based on facial feature coordinate data, the facial feature areas (including eyebrows, eyes, nose, mouth, and ears) are removed from a standard frontal face image. The background area (area outside the facial contour) is removed based on facial contour coordinate data. This involves resetting the pixel values ​​of both the facial feature areas and the background area to 0, resulting in a facial skin image. This achieves the effect of removing the background and facial feature areas. Colored areas are shown below. Figure 2 As shown, 0 represents the outer background area, and 1, 2, 3, and 4 represent the facial features area. The coloring operation is shown in formula (4).

[0054] f(x,y)=0 (4).

[0055] In the formula, f(x,y) is the pixel value at (x,y) in the image.

[0056] Step 104: Based on the Gaussian process and stochastic transformation algorithm, perform facial wrinkle detection on the facial skin map to obtain the wrinkle detection result map.

[0057] In this embodiment, step 104 specifically includes:

[0058] (1) Perform grayscale conversion on the facial skin image to obtain a grayscale image of the face.

[0059] (2) Using the Gaussian process and stochastic transformation algorithm, a method that tracks the center line and diameter of wrinkles, facial wrinkles are detected on a grayscale image of the face, resulting in a wrinkle detection image. The specific steps are as follows:

[0060] First, a seed point is randomly selected on the facial grayscale image, and the feature vector corresponding to the selected seed point is calculated using a random transformation algorithm, as shown in formula (5). The feature vector R(ρ, θ) is obtained by integrating along a straight line with different distances (ρ) and different angles (θ) from the origin.

[0061]

[0062] Then, the process of solving for wrinkle curvature and wrinkle diameter is assumed to be a Gaussian process, with wrinkle curvature and wrinkle diameter being the outputs of the Gaussian process, and the eigenvectors obtained above being used as the inputs of the Gaussian process.

[0063] Next, the wrinkle curvature and wrinkle diameter obtained above are used as prior information. For a single wrinkle without branching, the wrinkle curvature and wrinkle diameter usually change smoothly, so new wrinkle curvature and wrinkle diameter can be statistically predicted from past values.

[0064] The change in wrinkle direction is calculated using formulas (6) and (7).

[0065]

[0066] C(x n ,x m )=k(x n ,x m )+α -1 (7).

[0067] Where, x N It is the input vector of the Gaussian process, y N It is its corresponding output variable, C N It has element k(x) n ,x m The covariance matrix of x, where k is a matrix with elements k(x). n ,x N+1 A vector of type k, n = 1, ..., N. And the scalar z = k(x) N+1 ,x N+1 )+α (-1)α represents the precision of the random noise. These equations are key results in defining Gaussian process regression.

[0068] Finally, by adding the wrinkle direction change calculated above to the wrinkle forward direction, a new wrinkle direction is calculated. Moving forward one step along the new wrinkle direction, the prediction of wrinkle direction changes continues until the end of the wrinkle is reached. After one wrinkle is detected, a seed point is randomly selected again until all wrinkles are detected, resulting in the following... Figure 3 The results are shown in the figure.

[0069] Step 105: Extract the facial position information from the original frontal face image, and crop the face region from the wrinkle detection result image based on the facial position information.

[0070] In this embodiment, extracting facial position information from the original frontal face image specifically includes:

[0071] This paper utilizes facial landmark detection technology to extract facial location information from the original frontal face image. Facial landmark detection is a general-purpose technology, and the implementation process can be found in the open-source code at: https: / / github.com / codeniko / shape_predictor_81_face_landmarks.

[0072] Furthermore, using facial landmark detection technology, the facial contour is marked as 81 points in the original frontal face image. The positions of the highest point A on the forehead, the lowest point B on the chin, the leftmost point C on the face, and the rightmost point D on the face are obtained from the original frontal face image. The positions of points A, B, C, and D are then read and stored on a cloud server. The positions of the four points are as follows: Figure 4 As shown.

[0073] In this embodiment, the face region is cropped from the wrinkle detection result image based on the face location information, specifically including:

[0074] Extract the stored location information of points A, B, C, and D, and mark the positions of the four points on the cedar tree in the wrinkle detection result image, such as... Figure 4 As shown in the image. Then connect these four points to form a rectangular area; this rectangular area is the face region. Cropping this area yields the face region, as shown in the image. Figure 5 As shown.

[0075] Step 106: Perform image processing on the face region to obtain the wrinkle region, and determine the number of wrinkles in the wrinkle region as well as the length, width and depth of each wrinkle.

[0076] In this embodiment, image processing is performed on the face region to obtain the wrinkle region, specifically including:

[0077] The face region is binarized, and connected components within the binarized region are calculated. Then, the number of pixels and eccentricity of each connected component are calculated. Finally, based on the number of pixels and eccentricity of the connected components, non-wrinkle areas are removed from the face region, while wrinkle areas are retained. For example, if the number of pixels in a connected component is less than 100 or the eccentricity is less than 0.95, that connected component is removed, treating it as a non-wrinkle area. Specific operation steps and removal conditions are as follows... Figure 6 As shown, the effect after removal is as follows Figure 7 As shown.

[0078] In this embodiment, determining the number of wrinkles within the wrinkle area, as well as the length, width, and depth of each wrinkle, specifically includes:

[0079] A connected component identification operation is performed on the wrinkle region to obtain the connected components located within the wrinkle region; where a connected component located within the wrinkle region represents a wrinkle, and the number of connected components located within the wrinkle region is the number of wrinkles.

[0080] Calculate the minimum bounding rectangle of each connected component, and determine the length of the minimum bounding rectangle as the length of the wrinkle (length(i)) and the width of the minimum bounding rectangle as the width of the wrinkle (width(i)) to obtain the length and width of each wrinkle.

[0081] Calculate the curvature K(i) of the cross-section of each connected component in four directions, and determine the maximum curvature value as the wrinkle depth, thus obtaining the depth Depth(i) of each wrinkle. The four directions are as follows: Figure 8 As shown, the curvature calculation method is as shown in formula (8):

[0082]

[0083] The method for calculating the depth of wrinkles is shown in formula (9):

[0084] Depth(i) = max(K(i)) (9).

[0085] Where i = 0, ..., N, and N is the number of connected components. n represents the four directions, K(n) represents the curvature of the cross-section in the four directions, f(n) represents the edge profile of the cross-section obtained at the four directions, and fvr represents the connected component for which the curvature calculation operation is to be performed.

[0086] Step 107: Calculate the number of wrinkles in the wrinkle area and the length, width and depth of each wrinkle, and calculate the three-dimensional wrinkle density based on the number of wrinkles in the face and the face area.

[0087] In this embodiment, the number of wrinkle levels on the entire face is calculated according to formula (10).

[0088] F=∑length(i)*width(i)*Depth(i) (10).

[0089] In this embodiment, based on the number of wrinkle levels on the entire face and the size of the face region (i.e., attached...) Figure 5 The density M of the three-dimensional wrinkle is calculated by taking the number of pixels and the number of pixels. The calculation method is shown in formula (11):

[0090] M = F / (x*y) (11).

[0091] Where i = 0, ..., N, and N is the number of connected components. x and y are the width and height of the face region for wrinkle detection, respectively.

[0092] Step 108: Evaluate the facial wrinkles of the subject in three dimensions based on the three-dimensional wrinkle density.

[0093] In this embodiment, the three-dimensional wrinkle density is converted into a score representing the severity of wrinkles, thereby realizing a three-dimensional evaluation of the subject's facial wrinkles. The calculation method is shown in formula (12):

[0094]

[0095] Where M1 is the conversion threshold, which is defined as shown in formula (13):

[0096] M1 = M * 10 4 (13).

[0097] The system flow of the three-dimensional evaluation method for facial wrinkles based on Gaussian processes and random transformations provided in this embodiment is as follows: Figure 9 As shown.

[0098] Example 2

[0099] This embodiment provides a three-dimensional evaluation method for facial wrinkles based on Gaussian processes and random transformations, including:

[0100] Step 1: Use imaging equipment to capture the original frontal facial image of the subject, and then transmit the original image to the facial wrinkle 3D evaluation system.

[0101] Step 2: Upon receiving the original image, the facial wrinkle 3D evaluation system receives a facial detection request. The original image is then resized to a standard size—1024 pixels wide and 1024 pixels long.

[0102] Step 3: The corresponding model, which has been trained by FPHANET, is called from the cloud server to process the standardized original image and return the coordinate data of facial features and facial contours.

[0103] Step 4: Based on the coordinate data of facial features and facial contours obtained in Step 3, remove facial features from the facial image and remove the background area based on the facial contour. That is, reset the pixel values ​​of facial features (including eyebrows, eyes, nose, mouth, and ears) and background area (area outside the facial contour) to 0, thereby achieving the effect of removing the background and facial features.

[0104] Step 5: Using the facial skin image obtained in step 4, perform facial wrinkle detection to obtain the wrinkle detection result image.

[0105] Step 6: Using facial landmark detection technology, the facial contour is marked as 81 points in the original facial image mentioned in Step 1. The positions of the highest point A on the forehead, the lowest point B on the chin, the leftmost point C on the face, and the rightmost point D on the face are obtained in the image. The positions of the four points A, B, C, and D are read and stored in the cloud server.

[0106] Step 7: Extract the location information of the four points A, B, C, and D on the face. Mark the positions of the four points in the wrinkle detection result image obtained in step 5, connect these four points to form a rectangle, and crop out the rectangular area, i.e. the face area, to obtain the face image.

[0107] Step 8: Binarize the cropped image obtained in step 7 to obtain a binarized image, and then calculate the connected components of the image. Calculate the number of pixels and the eccentricity of each connected component. If the number of pixels in a connected component is less than 100 or the eccentricity is less than 0.95, remove the connected component, treating it as a non-wrinkle region.

[0108] Step 9: Perform connected component recognition on the image obtained in Step 8 after removing non-wrinkle regions. Each connected component represents a wrinkle. Calculate the minimum bounding rectangle of each connected component and determine the length of the wrinkle as length(i) and the width of the minimum bounding rectangle as width(i), thus obtaining the length and width of each wrinkle. Calculate the curvature K(i) of the cross-section of each connected component in four directions and determine the maximum curvature value as the depth of the wrinkle, thus obtaining the depth Depth(i) of each wrinkle.

[0109] Step 10: Calculate the number of wrinkles in the wrinkle area and the length, width and depth of each wrinkle, and calculate the three-dimensional wrinkle density based on the number of wrinkles in the face and the face area.

[0110] Step 11: Based on the three-dimensional wrinkle density, a three-dimensional evaluation of the subject's facial wrinkles is performed.

[0111] Calculated Figure 5 The three-dimensional wrinkle density (retaining four significant figures) is 0.0002223. Then, the wrinkle distribution rate is converted using the three-dimensional wrinkle density, and the evaluation result is expressed as a percentage score. Finally... Figure 5 The calculated wrinkle score was 67.77.

[0112] Step 12: After visualizing the final score and the image of the completed wrinkle detection, return it to the system front end.

[0113] Example 3

[0114] In order to implement the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, a three-dimensional evaluation system for facial wrinkles based on Gaussian processes and random transformations is provided below.

[0115] This embodiment provides a three-dimensional facial wrinkle evaluation system based on Gaussian processes and random transformations, including:

[0116] The image standardization processing module is used to read an original frontal face image of a subject and standardize the original frontal face image to obtain a standard frontal face image.

[0117] The coordinate data extraction module is used to locate facial features and facial contours in a standard frontal face image based on a custom neural network model, and obtain the coordinate data of facial features and facial contours.

[0118] The facial skin map determination module is used to remove the facial features and background areas from the standard frontal face image based on the facial feature coordinate data and facial contour coordinate data to obtain the facial skin map.

[0119] The wrinkle detection result image determination module is used to perform facial wrinkle detection on facial skin images based on Gaussian process and random transformation algorithm, and obtain wrinkle detection result images.

[0120] The face region cropping module is used to extract the facial position information from the original frontal face image and crop the face region from the wrinkle detection result image based on the facial position information.

[0121] The wrinkle information determination module is used to perform image processing on the face region to obtain the wrinkle region, and to determine the number of wrinkles in the wrinkle region as well as the length, width and depth of each wrinkle.

[0122] The 3D wrinkle density calculation module is used to calculate the number of wrinkle levels on the whole face based on the number of wrinkles in the wrinkle area and the length, width and depth of each wrinkle, and to calculate the 3D wrinkle density based on the number of wrinkle levels on the whole face and the face area.

[0123] The three-dimensional evaluation module is used to evaluate the facial wrinkles of the subject in three dimensions based on the three-dimensional wrinkle density.

[0124] The present invention has the following advantages:

[0125] 1. By utilizing a novel neural network structure to remove background and facial features, the accuracy of facial feature localization is improved, thus enabling better image processing.

[0126] 2. By using Gaussian processes and random transformations for wrinkle recognition, the robustness of the system is improved, and the direction of wrinkle changes can be identified, solving the problem that other methods can only identify wrinkles in a certain direction.

[0127] 3. A scientific evaluation and scoring mechanism for wrinkle severity has been established, solving the problem of the lack of clear evaluation standards in the market.

[0128] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0129] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A three-dimensional evaluation method for facial wrinkles based on Gaussian processes and stochastic transformations, characterized in that, include: Read an original frontal face image of a subject and standardize the original frontal face image to obtain a standard frontal face image. Based on a custom neural network model, facial features and facial contours are located in a standard frontal face image to obtain facial feature coordinate data and facial contour coordinate data. The custom neural network model is obtained by inputting a public dataset into a face parsing hierarchical aggregation network for model training. Based on the facial feature coordinate data and facial contour coordinate data, the facial feature area and background area are removed from the frontal standard face image to obtain the facial skin image. Based on the Gaussian process and stochastic transformation algorithm, facial wrinkle detection is performed on a facial skin image to obtain a wrinkle detection result image; specifically including: A seed point is randomly selected on the facial grayscale image, and the feature vector corresponding to the selected seed point is calculated using a random transformation algorithm; the feature vector is obtained by integrating along a straight line with different distances and angles from the origin. The feature vector is used as the input of the Gaussian process, and the wrinkle curvature and wrinkle diameter are used as prior information to predict the change in wrinkle direction. The predicted change in wrinkle direction is then added to the wrinkle forward direction to calculate the new wrinkle direction. Move forward one step along the new wrinkle direction and continue to predict the change in wrinkle direction until the end of the wrinkle is reached. After one wrinkle is detected, randomly select a seed point from the facial grayscale image until all wrinkles are detected and obtain the wrinkle detection result image. Extract facial position information from the original frontal face image, and crop the face region from the wrinkle detection result image based on the facial position information; Image processing is performed on the face region to obtain the wrinkle region, and the number of wrinkles in the wrinkle region, as well as the length, width and depth of each wrinkle, are determined. Based on the number of wrinkles within the wrinkle area and the length, width, and depth of each wrinkle, the total facial wrinkle volume level is calculated, and the three-dimensional wrinkle volume density is calculated based on the total facial wrinkle volume level and the facial area; the formula for calculating the three-dimensional wrinkle volume density is as follows: M = F / (x*y); Where M represents the three-dimensional wrinkle density, F represents the wrinkle level of the whole face, and x and y represent the width and height of the face region; Based on the three-dimensional wrinkle density, the facial wrinkles of the subjects were evaluated in three dimensions.

2. The three-dimensional evaluation method for facial wrinkles based on Gaussian processes and random transformations according to claim 1, characterized in that, The original frontal face image is standardized to obtain a standard frontal face image, specifically including: If the original frontal face image is smaller than the standard size, a two-dimensional interpolation algorithm is used to standardize the original frontal face image to obtain a standard frontal face image. If the original frontal face image is larger than the standard size, a two-dimensional downsampling algorithm is used to standardize the original frontal face image to obtain a standard frontal face image. If the original image of a frontal face is equal to the standard size, then the original image of a frontal face is determined as a standard frontal face image.

3. The three-dimensional evaluation method for facial wrinkles based on Gaussian processes and random transformations according to claim 1, characterized in that, Extract facial location information from the original frontal face image, specifically including: Using facial landmark detection technology, facial position information is extracted from the original frontal face image; the facial position information includes the position of the highest point of the forehead, the lowest point of the chin, the position of the leftmost point of the face, and the position of the rightmost point of the face.

4. The three-dimensional evaluation method for facial wrinkles based on Gaussian processes and random transformations according to claim 1, characterized in that, Image processing is performed on the facial region to obtain the wrinkle region, specifically including: Binarize the face region and calculate the connected components within the binarized face region. Calculate the number of pixels and eccentricity of each connected component, and based on the number of pixels and eccentricity of the connected components, remove non-wrinkle regions and retain wrinkle regions in the face region.

5. The three-dimensional evaluation method for facial wrinkles based on Gaussian processes and random transformations according to claim 1, characterized in that, Determine the number of wrinkles within the wrinkle area, as well as the length, width, and depth of each wrinkle, specifically including: A connected component identification operation is performed on the wrinkle region to obtain the connected components located within the wrinkle region; where a connected component located within the wrinkle region represents a wrinkle, and the number of connected components located within the wrinkle region is the number of wrinkles. Calculate the minimum bounding rectangle of each connected component, and determine the length of the minimum bounding rectangle as the length of the wrinkle, and determine the width of the minimum bounding rectangle as the width of the wrinkle, thus obtaining the length and width of each wrinkle; The curvature of the cross-section of each connected region in four directions is calculated, and the maximum curvature value is determined as the depth of the wrinkle, thereby obtaining the depth of each wrinkle; the four directions are the horizontal direction, the 45° direction, the vertical direction, and the 135° direction.

6. The three-dimensional evaluation method for facial wrinkles based on Gaussian processes and random transformations according to claim 1, characterized in that, The formula for calculating the grade of wrinkles on the entire face is: F=∑length(i)*width(i)*Depth(i); Where F represents the number of wrinkle levels on the whole face, length(i) represents the length of the i-th wrinkle, width(i) represents the width of the i-th wrinkle, and depth(i) represents the depth of the i-th wrinkle.

7. A three-dimensional evaluation system for facial wrinkles based on Gaussian processes and stochastic transformations, characterized in that, include: The image standardization processing module is used to read an original frontal face image of a subject and standardize the original frontal face image to obtain a standard frontal face image. The coordinate data extraction module is used to locate facial features and facial contours in a frontal standard face image based on a custom neural network model, and obtain facial feature coordinate data and facial contour coordinate data. The custom neural network model is obtained by inputting a public dataset into a face parsing hierarchical aggregation network for model training. The facial skin map determination module is used to remove the facial features and background areas from the standard frontal face image based on the facial feature coordinate data and facial contour coordinate data to obtain the facial skin map. The wrinkle detection result image determination module is used to perform facial wrinkle detection on a facial skin image based on a Gaussian process and stochastic transformation algorithm, and obtain a wrinkle detection result image; specifically, it includes: A seed point is randomly selected on the facial grayscale image, and the feature vector corresponding to the selected seed point is calculated using a random transformation algorithm; the feature vector is obtained by integrating along a straight line with different distances and angles from the origin. The feature vector is used as the input of the Gaussian process, and the wrinkle curvature and wrinkle diameter are used as prior information to predict the change in wrinkle direction. The predicted change in wrinkle direction is then added to the wrinkle forward direction to calculate the new wrinkle direction. Move forward one step along the new wrinkle direction and continue to predict the change in wrinkle direction until the end of the wrinkle is reached. After one wrinkle is detected, randomly select a seed point from the facial grayscale image until all wrinkles are detected and obtain the wrinkle detection result image. The face region cropping module is used to extract the face position information from the original frontal face image and crop the face region from the wrinkle detection result image based on the face position information. The wrinkle information determination module is used to perform image processing on the face region to obtain the wrinkle region, and determine the number of wrinkles in the wrinkle region as well as the length, width and depth of each wrinkle. The three-dimensional wrinkle density calculation module is used to calculate the number of wrinkle levels on the entire face based on the number of wrinkles within the wrinkle area and the length, width, and depth of each wrinkle, and to calculate the three-dimensional wrinkle density based on the number of wrinkle levels and the facial area; the calculation formula for the three-dimensional wrinkle density is as follows: M = F / (x*y); Where M represents the three-dimensional wrinkle density, F represents the wrinkle level of the whole face, and x and y represent the width and height of the face region; The three-dimensional evaluation module is used to evaluate the facial wrinkles of the subject in three dimensions based on the three-dimensional wrinkle density.

Images