A face skin quality recognition system fusing acupoint position features

Abstract

The application relates to the technical field of face skin quality recognition, in particular to a face skin quality recognition system fusing meridian point position features. The system comprises an image acquisition module for acquiring a skin image of a face; an acupoint partition and skin quality confidence evaluation module for partitioning acupoint regions based on meridian point positions, extracting features such as ruddiness and dullness by analyzing pixel distribution to construct a skin quality vector; classifying regions according to meridians, quantifying skin quality confidence of each acupoint region by using feature similarity in the region and similarity between other regions in the class; and a skin quality recognition module for performing weighted fusion processing on skin quality feature vectors of all acupoint regions and corresponding skin quality confidences, and recognizing the face skin quality type by using the fused result. The application solves the problem of inaccurate recognition caused by difficulty in distinguishing internal constitution and external interference, and improves the accuracy and anti-interference ability of skin quality recognition.

Description

Technical Field

[0001] This application relates to the field of facial skin texture recognition technology, specifically to a facial skin texture recognition system that integrates acupoint location features. Background Technology

[0002] With the development of artificial intelligence and digital health technologies, facial skin texture recognition has been widely used in the fields of intelligent beauty and health management. Traditional methods mainly rely on texture modeling and color analysis to extract superficial features such as oil and pores to automatically identify skin type. However, these data-driven technologies lack modeling of the relationship between internal physiological state and external manifestation, especially neglecting the systematic connection between skin texture and organ function under the TCM theories of "holistic view" and "inferring internal from external manifestation".

[0003] Existing skin texture recognition methods generally suffer from a key limitation: they primarily rely on image feature extraction from the entire face or empirically defined regions (such as the T-zone and U-zone) to determine skin texture. This strategy is highly susceptible to interference from non-constitutional factors such as shooting lighting, angle, ambient temperature and humidity, recent skincare practices, or temporary physiological states (such as staying up late or diet). This can lead to the skin texture being mistaken for a fundamental change in skin type. Furthermore, current methods cannot distinguish whether skin texture fluctuations stem from long-term evolution of an individual's internal constitution (such as worsening of spleen and stomach damp-heat or progression of liver and kidney yin deficiency) or short-term superficial changes caused by external disturbances. This confusion reduces the accuracy of facial skin texture recognition. Summary of the Invention

[0004] To address the aforementioned technical problems, the purpose of this application is to provide a facial skin texture recognition system that integrates acupoint location features. The specific technical solution adopted is as follows: This application proposes a facial skin texture recognition system that integrates acupoint location features, the system comprising: The image acquisition module is used to acquire skin images of human faces; The acupoint zoning and skin texture confidence assessment module is used to divide the skin image into multiple acupoint regions based on the acupoint location; by analyzing the distribution of pixel values ​​in each acupoint region, the module assesses the redness, dullness, oiliness, and roughness of the facial skin in each acupoint region, thereby constructing a skin texture feature vector for each acupoint region. Acupoint regions belonging to the same meridian are divided into the same category. By analyzing the similarity of each acupoint region with the other acupoint regions in its category on the skin texture feature vector, as well as the similarity of any two acupoint regions in its category other than each acupoint region, the skin texture confidence of each acupoint region is quantified. The skin texture recognition module is used to perform weighted fusion processing on the skin texture feature vectors of all acupoint areas and the corresponding skin texture confidence scores, and to identify the skin texture type of the face using the fused result.

[0005] Preferably, the step of dividing the skin image into multiple acupoint regions based on acupoint distribution according to the acupoint location includes: The facial feature points and acupoints in the skin image are obtained. All facial feature points and acupoints are used as input to the Voronoi diagram. Boundary constraints of the effective facial region are constructed based on the facial feature points. All acupoints are used as seed points. The polygonal region of each seed point is output as the acupoint region.

[0006] Preferably, the assessment of the redness, dullness, oiliness, and roughness of the facial skin within each acupoint area includes: The skin image in Lab space is acquired. By analyzing the difference between the mean a-channel pixel value and the mean b-channel pixel value of all pixels in each acupoint area, as well as the dispersion of the a-channel pixel value of all pixels, the redness coefficient is determined to evaluate the redness of the facial skin. By analyzing the difference between the mean value of the b channel and the mean value of the L channel of all pixels in each acupoint area, as well as the contrast of all pixels in the L channel, a dullness coefficient is determined to assess the dullness of facial skin. The number of highlight pixels in each acupoint area is counted. Based on the proportion of all highlight pixels in each acupoint area and the distribution of gray values ​​of pixels in all pore areas in each acupoint area, an oiliness coefficient is determined to assess the oiliness of facial skin. Based on the contrast of all pixels in the L channel within each acupoint area and the area ratio of the pore area, a roughness coefficient is determined to assess the roughness of facial skin.

[0007] Preferably, the skin texture feature vector of each acupoint area consists of a redness coefficient, a dullness coefficient, an oiliness coefficient, and a roughness coefficient.

[0008] The preferred method for determining highlight pixels is as follows: Calculate the product of the maximum value of the L-channel pixel value of all pixels in the skin image and a preset value, and record it as the brightness feature value. Calculate the preset quantile of the L-channel pixel value of all pixels in the skin image. Select the maximum value between the preset quantile and the brightness feature value as the highlight threshold. Pixels with L-channel pixel values ​​greater than the highlight threshold are designated as highlight pixels.

[0009] Preferably, the pore region is obtained by image segmentation of the skin image using an adaptive region growing algorithm.

[0010] Preferably, the method for quantifying the skin quality confidence level of each acupoint region is as follows: Calculate the mean similarity of each acupoint region with all other acupoint regions in its category on the skin texture feature vector, and denote it as the first feature value; Calculate the mean similarity of all acupoint regions within the category to which each acupoint region belongs, excluding each acupoint region itself, on the skin texture feature vector, and denote it as the second feature value; Calculate the ratio of the first eigenvalue to the second eigenvalue, and use the sum of the ratio and 1 as the confidence score of the skin quality for each acupoint region.

[0011] Preferably, the weighted fusion processing of the skin texture feature vectors and corresponding skin texture confidence scores for all acupoint regions includes: Calculate the scalar multiplication of the skin texture feature vector of all acupoint regions in the skin image with the corresponding skin texture confidence normalization value, concatenate the vectors after logarithmic multiplication, and record the concatenated result as the weighted skin texture feature vector.

[0012] Preferably, the step of identifying facial skin type using the fused result includes: The weighted skin texture feature vector of the person to be detected is used as the input of the pre-trained face skin texture recognition model, and the output is the skin texture result of the person to be detected.

[0013] Preferably, the process of obtaining the pre-trained facial skin texture recognition model is as follows: Obtain the weighted skin feature vectors and corresponding skin labels of a preset number of people with known skin types, and use them as input to train a machine learning classifier. The trained machine learning classifier is then used as a pre-trained face skin recognition model.

[0014] This application has the following beneficial effects: Based on the intrinsic connection between Traditional Chinese Medicine (TCM) meridian theory and facial skin texture distribution, this application utilizes MediaPipe and Voronoi diagram algorithms to construct acupoint regions based on acupoint distribution, overcoming the poor anti-interference defects of traditional empirical partitioning. By deeply analyzing the distribution characteristics of pixels within each region in the Lab color space and fluorescence images, the coefficients for four dimensions—redness, dullness, oiliness, and roughness—are quantified, thereby constructing a high-dimensional skin texture feature vector that accurately reflects local physiological states. This achieves a comprehensive, objective, and robust digital representation of an individual's true skin condition. Furthermore, this application classifies acupoint regions based on TCM meridian theory. By calculating the average similarity between each acupoint region and other regions of the same type, as well as the average overall similarity of the same type excluding that region, the skin texture confidence score, which represents the degree of interference from non-physical factors, is obtained. This enables precise screening and adaptive weight adjustment of an individual's intrinsic basic skin texture characteristics, effectively suppressing the interference of local temporary fluctuations on the recognition results. Finally, this application uses the skin texture confidence score to perform weighted normalization and fusion of the feature vectors of acupoint regions, obtaining a weighted feature vector that accurately reflects an individual's intrinsic basic skin texture. The trained model is then used to identify the skin texture type of the person being tested, improving the accuracy of facial skin texture recognition. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of this application 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A block diagram of a facial skin texture recognition system that integrates acupoint location features, provided as an embodiment of this application; Figure 2 This is a flowchart illustrating the process of obtaining skin quality confidence in one embodiment of this application. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a facial skin texture recognition system integrating acupoint location features proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, 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 application pertains.

[0019] The following description, in conjunction with the accompanying drawings, details a specific scheme for a facial skin texture recognition system that integrates acupoint location features, as provided in this application.

[0020] Please see Figure 1 The diagram illustrates a facial skin texture recognition system that integrates acupoint location features according to an embodiment of this application. The system includes: an image acquisition module 101, an acupoint partitioning and skin texture confidence assessment module 102, and a skin texture recognition module 103.

[0021] Image acquisition module 101 is used to acquire skin images of human faces.

[0022] Facial skin images are acquired using a CMOS image sensor. Image acquisition utilizes a white LED light source with a dominant peak of 450nm and a UV light source as induction light sources. The LED light source is used to acquire visible light images of the facial skin for epidermal detection, while the UV light source is used to excite hair follicles and pores to obtain skin fluorescence images. During skin texture recognition, given the limitations of RGB and HSV color spaces in color representation methods, it is difficult to accurately describe the color characteristics of the skin surface. The Lab color space, with its uniform distribution characteristics in all directions, is closer to human visual perception. Therefore, the system first converts the acquired skin image (RGB image) under the LED light source to the XYZ color space, and then to the Lab color space, thus obtaining the skin image in the Lab color space. The specific color space conversion process is a well-known technique and will not be elaborated here.

[0023] The acupoint zoning and skin texture confidence assessment module 102 is used to divide the skin image into multiple acupoint regions based on the acupoint location; by analyzing the distribution of pixel values ​​in each acupoint region, the redness, dullness, oiliness, and roughness of the facial skin in each acupoint region are assessed, thereby constructing a skin texture feature vector for each acupoint region; acupoint regions belonging to the same meridian are divided into the same category, and the skin texture confidence of each acupoint region is quantified by analyzing the similarity of each acupoint region with the other acupoint regions in its category on the skin texture feature vector, as well as the similarity of any two acupoint regions in its category other than the acupoint regions mentioned above on the skin texture feature vector.

[0024] With the development of computer vision technology and artificial intelligence, image processing-based facial skin texture recognition technology has achieved a relatively better balance between "usability-cost-information content" due to its convenience, non-contact nature, and high information richness, and has become one of the main means of large-scale, daily skin texture monitoring. However, existing skin texture recognition methods mainly determine skin texture by extracting image features from the entire face or empirically defined regions. This recognition result is highly susceptible to interference from non-constitutional factors, resulting in the identified skin texture type failing to accurately reflect the individual's true physiological characteristics and exhibiting significant perturbation. This also makes it difficult to determine whether skin texture fluctuations stem from long-term evolution of an individual's internal constitution or short-term superficial changes caused by external perturbations.

[0025] Therefore, to more accurately determine an individual's actual skin condition and reduce interference from non-constitutional factors, this embodiment divides the skin image into multiple acupoint regions based on acupoint distribution according to the location of acupoints. By analyzing the distribution of pixel values ​​within each acupoint region, the redness, dullness, oiliness, and roughness of the facial skin within each acupoint region are evaluated, thereby constructing a skin quality feature vector for each acupoint region. Acupoint regions belonging to the same meridian are grouped into the same category. By analyzing the similarity of each acupoint region with the other acupoint regions within its category on the skin quality feature vector, and the similarity of any two acupoint regions within its category other than the aforementioned acupoint regions on the skin quality feature vector, the skin quality confidence of each acupoint region is quantified. The flowchart of the skin quality confidence acquisition process provided in this embodiment is as follows: Figure 2 As shown, the specific process for obtaining skin type confidence is as follows: S1: Divide the skin image into multiple acupoint regions based on the acupoint location; by analyzing the distribution of pixel values ​​in each acupoint region, assess the redness, dullness, oiliness, and roughness of the facial skin in each acupoint region, thereby constructing a skin texture feature vector for each acupoint region.

[0026] Normally, the skin condition of different areas on the face is not completely consistent, and there can even be significant differences. In current technology, fixed empirical partitioning methods such as T-zone and U-zone are commonly used to detect the skin condition of different areas. However, skin changes based on such fixed empirical partitioning often have weak or even no regularity, and are easily disturbed by non-constitutional factors, resulting in skin fluctuations. This not only leads to poor consistency of skin condition detection results in different situations, but also makes it difficult for the detection results to accurately represent the individual's true skin condition.

[0027] In fact, the skin condition at facial acupoints can reflect individual physiological characteristics and actual skin fluctuations. According to traditional Chinese medicine theory, the state of the Qi and blood of the internal organs is reflected on the body surface through the meridians, and the areas of facial skin quality difference highly overlap with the locations of classic acupoints. For example, the high sebum and large pores in the T-zone (forehead, nose, and nasal wings) correspond to damp-heat in the spleen and stomach, which overlaps with the meridian patterns of the Du meridian and the Stomach meridian of Foot Yangming; pigmentation, sensitivity, and dryness in the cheekbone area are attributed to imbalances in the liver, spleen, and kidneys, which is consistent with the functions of the Small Intestine meridian, the Triple Energizer meridian, and the Gallbladder meridian; the skin characteristics of the perioral and peri-ocular areas also highly match the meridian patterns of the Stomach meridian, the Large Intestine meridian, the Ren meridian, and the Bladder meridian.

[0028] Therefore, based on the above analysis, this embodiment divides the skin image into multiple acupoint regions based on acupoint distribution according to the location of acupoints; by analyzing the distribution of pixel values ​​in each acupoint region, the redness, dullness, oiliness, and roughness of the facial skin in each acupoint region are evaluated, thereby constructing a skin texture feature vector for each acupoint region. The specific process is as follows: First, the skin image is divided into multiple acupoint regions based on the location of acupoints, specifically: In this embodiment, the skin image is used as the input of the MediaPipe algorithm, and the output is the feature points covering the entire face on the skin image, which are recorded as facial feature points. Further, in the skin image, the lower left corner pixel is used as the origin to construct image coordinates, and the coordinates of each acupoint are obtained in the skin image through a machine learning-based location prediction model. The process of locating acupoints using a machine learning-based location prediction model is a well-known technique and will not be described in detail here.

[0029] Furthermore, all facial feature points and acupoints are used as input to the Voronoi diagram. To ensure that the divided area is limited to the effective skin surface of the face and avoids the area extending into irrelevant areas such as the background or hairline, this embodiment sequentially connects the feature points located at the edge of the facial contour to construct a closed polygonal boundary, and the area inside this boundary is used as the effective analysis area. Furthermore, all acupoints are used as seed points, and Voronoi diagrams are performed within the effective analysis area. The polygonal area of ​​each seed point is output as the acupoint area. The process of using Voronoi diagrams to divide the image into regions is a well-known technique and will not be described in detail here.

[0030] Furthermore, this embodiment analyzes the distribution of pixel values ​​within each acupoint region to assess the redness, dullness, oiliness, and roughness of the facial skin within each acupoint region, thereby constructing a skin texture feature vector for each acupoint region. Specifically: (1) In this embodiment, the redness coefficient is determined by analyzing the difference between the mean value of the a-channel pixel and the mean value of the b-channel pixel of all pixels in each acupoint area. This coefficient is used to assess the redness of the facial skin. Specifically: As one implementation method, in this embodiment, the redness coefficient of the facial skin in region i is... The expression is: In the formula, These represent the average pixel value of channel a and the average pixel value of channel b for all pixels within the acupoint region i, respectively. This represents the standard deviation of the pixel values ​​in channel a of all pixels within the acupoint region i. This represents a preset constant greater than 0, used to prevent the denominator from being 0. In this embodiment... The value is set manually, in this embodiment. The value of is 0.01. Provided that the denominator is not zero and does not excessively affect the calculation result, the implementer may also set it according to the specific situation. This embodiment does not impose any special restrictions.

[0031] The rosiness coefficient, as understood from the diagram, characterizes the vibrancy of red tones in the skin color of acupoint areas and the rosy complexion, reflecting the local blood circulation and the abundance of Qi and blood in the face. The calculation of the rosiness coefficient is primarily influenced by the difference between the means of the a and b channels in the Lab color space, as well as the standard deviation of the a channel. A larger difference between the a and b channel means indicates that red components are more dominant than yellow / blue components in the skin color, resulting in a more rosy complexion. This difference, as the numerator, directly contributes to the rosiness coefficient, making a higher rosiness coefficient indicate healthier and rosier skin. Conversely, a smaller difference indicates that the skin is more yellowish or bluish-white, with a lower rosiness. Meanwhile, the standard deviation of the a channel, as the denominator, reflects the uniformity of skin color in the area. A larger standard deviation indicates drastic color fluctuations and uneven skin tone, leading to a lower rosiness coefficient and reflecting instability or the presence of pigmentation. Conversely, a smaller standard deviation indicates a more uniform and stable skin tone, helping to maintain a higher rosiness coefficient and reflecting delicate and pure skin texture.

[0032] (2) In this embodiment, the dullness coefficient is determined by analyzing the difference between the average value of the b-channel pixel and the average value of the L-channel pixel of all pixels in each acupoint area, as well as the contrast of all pixels in the L-channel, to evaluate the dullness of facial skin. Specifically: As one implementation method, in this embodiment, the dullness coefficient of the facial skin in region i is... The expression is: In the formula, These represent the average pixel value of the b channel and the average pixel value of the L channel of all pixels within the acupoint region i, respectively. This represents the contrast of the gray-level co-occurrence matrix of all pixels within the acupoint region i on the L channel; This represents a preset constant greater than 0, used to prevent the denominator from being 0. In this embodiment... The value is set manually, in this embodiment. The value of is 0.01. Provided that the denominator is not zero and does not excessively affect the calculation result, the implementer may also set it according to the specific situation. This embodiment does not impose any special restrictions.

[0033] The method for calculating the contrast of the gray-level co-occurrence matrix is ​​a well-known technique and will not be elaborated further.

[0034] The dullness coefficient, as understood from the context, characterizes the degree of dullness and sallowness of skin luster in acupoint areas. It reflects the skin's aging state, the accumulation of stratum corneum, or poor blood circulation. The calculation of the dullness coefficient is influenced by the ratio of the average pixel value of the b-channel to the average pixel value of the L-channel in the Lab color space, as well as the contrast of the L-channel gray-level co-occurrence matrix. The ratio of the b-channel to the L-channel measures the saturation of yellow relative to brightness. A larger ratio indicates a more pronounced yellowing of the skin at the same brightness, or that lower brightness makes the yellow appear more prominent. A higher contrast ratio directly increases the dullness index, reflecting a lack of translucency and a sickly or tired yellowish appearance. Conversely, a smaller ratio indicates brighter and whiter skin with a lower dullness index. In addition, the L channel contrast reflects the difference in texture depth. A higher contrast ratio means that the skin surface texture is uneven and rough, and the diffuse reflection of light makes it appear darker, thus increasing the dullness index and indicating a rough skin texture. Conversely, a lower contrast ratio indicates smoother and more radiant skin with a lower dullness index, reflecting good skin texture.

[0035] (3) In this embodiment, the oiliness coefficient is determined by statistically analyzing the highlight pixels within each acupoint area, based on the proportion of all highlight pixels within each acupoint area, and the distribution of all pore areas within each acupoint area. This coefficient is used to assess the oiliness of facial skin. Specifically: In this embodiment, firstly, the product of the maximum value of the L-channel pixel value of all pixels in the skin image and a preset value is calculated and recorded as the brightness feature value. Then, a preset quantile of the L-channel pixel value of all pixels in the skin image is calculated. The maximum value between the preset quantile and the brightness feature value is selected as the highlight threshold.

[0036] The preset quantile and preset value are both set manually. In this embodiment, the preset quantile is set to 95% and the preset value is set to 0.9. The reason for this setting is to accurately filter out the strong highlight areas that can reflect sebum secretion through a dual threshold strategy, while effectively eliminating the interference of false highlights caused by ordinary skin texture or slight uneven lighting. This can ensure that the identified highlight pixels are indeed due to the specular reflection formed by sebum on the skin surface, thereby improving the accuracy and anti-interference ability of the oiliness coefficient calculation.

[0037] Furthermore, a UV light source is used to excite hair follicles and pores to obtain a fluorescence image of facial skin. Since pores are usually dark (low grayscale value), this embodiment converts the fluorescence image into a grayscale image. The grayscale value of all pixels in the grayscale image is used as the input of a threshold segmentation algorithm, and the output is a segmentation threshold. Pixels with grayscale values ​​less than the segmentation threshold are used as initial seed points. The grayscale image is used as the input of an adaptive region growing algorithm, and the final output growing region is used as the pore region. The pore region is then mapped to the corresponding position in the skin image.

[0038] Furthermore, this embodiment determines the oiliness coefficient based on the proportion of all highlight pixels within each acupoint area and the distribution of all pore areas within each acupoint area. Specifically: As one implementation method, in this embodiment, the oiliness coefficient of the facial skin in region i is... The expression is: In the formula, This represents the number of all highlight pixels within the acupoint region i; This represents the total number of pixels within the acupoint region i. This represents the average grayscale value of all pixels within all pore areas in acupoint region i. This represents the standard deviation of the grayscale values ​​of all pixels within all pore areas of acupoint region i; This represents a preset constant greater than 0, used to prevent the denominator from being 0. In this embodiment... The value is set manually, in this embodiment. The value of is 0.01. Provided that the denominator is not zero and does not excessively affect the calculation result, the implementer may also set it according to the specific situation. This embodiment does not impose any special restrictions.

[0039] Based on the oiliness coefficient, it can be understood that the oiliness coefficient is used to characterize the degree of sebum secretion and pore blockage in the acupoint area, reflecting the oil level and cleanliness of the skin. The calculation of the oiliness coefficient is mainly affected by the proportion of highlight pixels and the fluorescence intensity of the pore area. Among them, the proportion of highlight pixels and the mean gray value of pixels in the pore area are positive factors. The larger the proportion of highlight pixels, the more highlight areas are formed on the skin surface due to oil reflection, or the fluorescence is enhanced due to the accumulation of oil and keratin in the pores. The higher the oiliness coefficient, the more oily the skin is, reflecting that the skin is oily, the pores are large, or the surface is greasy. Conversely, the smaller these values ​​are, the more matte the skin surface or the cleaner the pores. The lower the oiliness coefficient, the drier the skin is. The higher the mean gray value of pixels within the pore area and the smaller the standard deviation of the gray values ​​within the pore area, the darker and more concentrated the pore area is, indicating a higher level of skin oil. Therefore, the corresponding oiliness coefficient is higher. Conversely, the lower the mean gray value of pixels within the pore area and the larger the standard deviation of the gray values ​​within the pore area, the brighter and more dispersed the pore area is, indicating a lower level of skin oil. Therefore, the corresponding oiliness coefficient is lower.

[0040] (4) In this embodiment, the roughness coefficient is determined based on the contrast of all pixels in the L channel within each acupoint area and the area ratio of the pore area, which is used to evaluate the roughness of facial skin. Specifically: As one implementation method, the roughness coefficient of the facial skin in region i in this embodiment is... The expression is: In the formula, This represents the contrast of the gray-level co-occurrence matrix of all pixels within the acupoint region i on the L channel; This represents the percentage of the total area of ​​all pore areas within acupoint region i.

[0041] The roughness coefficient is used to characterize the texture and unevenness of the skin surface in the acupoint area, reflecting the smoothness, gloss, and aging of the skin. The roughness coefficient is jointly affected by the contrast of the L channel grayscale co-occurrence matrix in the Lab color space and the area ratio of the pore region. Both factors have a positive impact. That is, the larger the L channel contrast, the deeper the grooves and the more dramatic the changes in light and dark on the skin surface, or the larger the area ratio of the pores, the larger and denser the hair follicles. Both will lead to a significant increase in the roughness coefficient, reflecting that the skin surface is extremely rough, lacks elasticity, or has severely enlarged pores. Conversely, the smaller the L channel contrast and the smaller the area ratio of the pores, the finer the skin texture, the smaller or almost invisible the pores, and the lower the roughness coefficient, reflecting the excellent condition of smooth and delicate skin.

[0042] Furthermore, the vector composed of the redness coefficient, dullness coefficient, oiliness coefficient, and roughness coefficient of each acupoint area is denoted as the skin texture feature vector.

[0043] Thus, this embodiment, based on the intrinsic connection between the theory of meridians in traditional Chinese medicine and the distribution of facial skin texture, uses MediaPipe and Voronoi diagram algorithms to construct acupoint regions based on the distribution of acupoints, thus eliminating the shortcomings of traditional empirical partitioning in terms of poor anti-interference. By deeply analyzing the distribution characteristics of pixels in each region in the Lab color space and fluorescence image, the coefficients of four dimensions—redness, dullness, oiliness, and roughness—are quantitatively calculated, thereby constructing a high-dimensional skin texture feature vector that can accurately reflect the local physiological state, achieving a comprehensive, objective, and robust digital representation of an individual's true skin texture.

[0044] S2: Divide acupoint regions belonging to the same meridian into the same category. Quantify the skin quality confidence of each acupoint region by analyzing the similarity of each acupoint region with the other acupoint regions in its category on the skin quality feature vector, as well as the similarity of any two acupoint regions in its category other than the acupoint regions mentioned above on the skin quality feature vector.

[0045] While the aforementioned skin texture feature vectors can measure the characteristics of each acupoint region, skin texture is prone to short-term fluctuations due to non-constitutional factors. Directly using these vectors for identification will result in significant errors and cannot represent the true underlying skin texture. In reality, multiple facial acupoints on the same meridian and symmetrically distributed acupoints are physiologically synchronous, and the skin texture characteristics of their surrounding areas should be highly similar. However, local fluctuations caused by non-constitutional factors can not only temporarily change local skin texture but may even mask the original characteristics, disrupting this similarity between regions. The goal of this embodiment is to identify the underlying skin texture determined by individual physiological characteristics. Therefore, this embodiment quantifies the skin texture confidence of each acupoint region by dividing acupoint regions belonging to the same meridian into the same category and analyzing the similarity of each acupoint region with the other acupoint regions in its category on the skin texture feature vector, as well as the similarity of any two acupoint regions in its category other than the aforementioned acupoint regions on the skin texture feature vector. The specific process is as follows: First, this embodiment divides acupoint regions belonging to the same meridian into the same category. Specifically, it analyzes the similarity of each acupoint region with the other acupoint regions within its category on the skin texture feature vector. In this embodiment, based on the theory of meridians in traditional Chinese medicine and the standards of human anatomy, a mapping relationship table between facial acupoints and meridian names is established. The mapping relationship table is queried according to the spatial location of the acupoint area to obtain the meridian label corresponding to each acupoint area, and acupoint areas with the same meridian label are grouped into the same category.

[0046] Furthermore, in this embodiment, the average similarity of each acupoint region with all other acupoint regions in its category on the skin texture feature vector is calculated and denoted as the first feature value. Calculate the mean similarity of all acupoint regions within the category to which each acupoint region belongs, excluding each acupoint region itself, on the skin texture feature vector, and denote it as the second feature value; The ratio of the first feature value to the second feature value is calculated, and the sum of the ratio and 1 is used as the confidence score of the skin quality for each acupoint region. Since the second feature value can be 0, a preset factor is added to the second feature value during the calculation process to prevent the ratio from being meaningless due to a denominator of 0. In this embodiment, the preset factor is set to 0.01. Under the premise of ensuring that the denominator is not 0 and does not excessively affect the calculation result, the implementer can also set it according to the specific situation. This embodiment does not impose any special restrictions.

[0047] It should be noted that there are many methods for measuring the similarity between vectors. In this embodiment, the normalized value of the cosine similarity between each acupoint region and all other acupoint regions within its category on the skin texture feature vector is used as the similarity between each acupoint region and all other acupoint regions within its category on the skin texture feature vector. Similarly, the normalized value of the cosine similarity between all acupoint regions within the category of each acupoint region (excluding the acupoint region itself) on the skin texture feature vector is used as the similarity between all acupoint regions within the category of each acupoint region (excluding the acupoint region itself) on the skin texture feature vector. In practical applications, as other implementation methods, implementers may also adopt other similarity measurement methods according to specific circumstances. This embodiment does not impose any special restrictions.

[0048] In this embodiment, the maximum-minimum normalization method is used to normalize the cosine similarity. In practical applications, implementers may also use other normalization methods depending on the specific circumstances. This embodiment does not impose any special restrictions on the selection of normalization methods. The calculation process of cosine similarity and the process of normalizing data using the maximum-minimum normalization method are well-known technologies and will not be described in detail here.

[0049] Optionally, when the total number of acupoint regions within the same category is less than 3, the second feature value adopts a preset default value, or adopts other alternative algorithms based on global data.

[0050] Based on the skin texture confidence score, it can be understood that the skin texture characteristics of a specific acupoint area truly reflect the individual's internal constitution, that is, the degree to which the feature data is interfered with by non-constitutional factors (such as temporary environment or local pollution). The calculation of skin texture confidence score is affected by the first feature value (the average similarity between the acupoint area and other acupoint areas of the same type) and the second feature value (the average overall similarity among other acupoint areas of the same type). The larger the ratio of the first feature value to the second feature value, the more consistent the characteristics of the acupoint area are with the characteristics of other acupoint areas on the same meridian, which conforms to the physiological law of meridian synchronization in traditional Chinese medicine. This means that the acupoint area has not been significantly interfered with, and the higher the skin texture confidence score, the greater the weight of its feature data should be in the subsequent weighted fusion. Conversely, the smaller the ratio of the first feature value to the second feature value, the more the characteristics of the acupoint area deviate from the common characteristics of similar areas, suggesting that it may have been disturbed by local non-constitutional factors. The lower the skin texture confidence score, the less reliable the data of this area is, and its weight should be reduced during fusion to suppress interference.

[0051] Thus, this embodiment classifies acupoint regions based on the theory of meridians in traditional Chinese medicine. By calculating the average similarity between each acupoint region and other regions of the same type, as well as the average overall similarity of the same type after removing the acupoint region, the skin quality confidence level, which represents the degree of interference of non-physical factors, is obtained. This achieves accurate screening and adaptive weight control of the individual's inherent basic skin quality characteristics, effectively suppressing the interference of local temporary fluctuations on the recognition results.

[0052] The skin texture recognition module 103 is used to perform weighted fusion processing on the skin texture feature vectors of all acupoint areas and the corresponding skin texture confidence scores, and to use the fusion result to identify the skin texture type of the face.

[0053] Based on the acupoint zoning and skin texture confidence assessment module, skin texture feature vectors and their corresponding skin texture confidence scores for all acupoint regions in the skin image were obtained. Therefore, this embodiment further performs weighted fusion processing on the skin texture feature vectors and corresponding skin texture confidence scores for all acupoint regions, and uses the fused results to identify the skin texture type of the face. The specific process is as follows: Calculate the scalar multiplication of the skin texture feature vector of all acupoint regions in the skin image with the corresponding skin texture confidence normalization value, concatenate the vectors after logarithmic multiplication, and record the concatenated result as the weighted skin texture feature vector.

[0054] It should be noted that there are many commonly used normalization methods. In this embodiment, the maximum-minimum normalization method is used to normalize the skin quality confidence score, mapping the skin quality confidence score to the range of [0,1]. In practical applications, as other implementation methods, implementers may also use other normalization methods according to specific circumstances. This embodiment does not impose any special restrictions.

[0055] Furthermore, in this embodiment, a preset number of weighted skin feature vectors and their corresponding skin labels are collected from individuals with known skin types. In this embodiment, the skin label types include "dry", "oily", "combination", "sensitive", and "neutral". Furthermore, the weighted skin feature vectors and their corresponding skin labels are used as input to a machine learning classifier. In this embodiment, a support vector machine is used to train it. The kernel function is a radial basis function, the regularization parameter is 1, and the kernel parameter is 0.25. The trained machine learning classifier is used as a pre-trained face skin recognition model.

[0056] It should be noted that the preset quantity is set manually. In this embodiment, the preset quantity is 500. In actual application, as other implementation methods, implementers can also set it according to specific circumstances. This embodiment does not impose any special restrictions.

[0057] The process of training a machine learning classifier is a well-known technique and will not be described in detail here.

[0058] Furthermore, the weighted skin texture feature vector of the person to be detected is used as the input of the pre-trained face skin texture recognition model, and the skin texture result of the person to be detected is output.

[0059] Thus, this embodiment utilizes skin texture confidence to perform weighted normalization and fusion of feature vectors in acupoint regions, obtaining a weighted feature vector that accurately reflects an individual's underlying skin texture. The trained model is then used to identify the skin texture type of the person being tested, thereby improving the accuracy of facial skin texture recognition.

[0060] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0061] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0062] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A facial skin texture recognition system integrating acupoint location features, characterized in that, The system includes: The image acquisition module is used to acquire skin images of human faces; The acupoint zoning and skin texture confidence assessment module is used to divide the skin image into multiple acupoint regions based on the acupoint location; by analyzing the distribution of pixel values ​​in each acupoint region, the module assesses the redness, dullness, oiliness, and roughness of the facial skin in each acupoint region, thereby constructing a skin texture feature vector for each acupoint region. Acupoint regions belonging to the same meridian are divided into the same category. By analyzing the similarity of each acupoint region with the other acupoint regions in its category on the skin texture feature vector, as well as the similarity of any two acupoint regions in its category other than each acupoint region, the skin texture confidence of each acupoint region is quantified. The skin texture recognition module is used to perform weighted fusion processing on the skin texture feature vectors of all acupoint areas and the corresponding skin texture confidence scores, and to identify the skin texture type of the face using the fused result.

2. The facial skin texture recognition system integrating acupoint location features according to claim 1, characterized in that, The process of dividing the skin image into multiple acupoint regions based on acupoint distribution according to the location of acupoints includes: The facial feature points and acupoints in the skin image are obtained. All facial feature points and acupoints are used as input to the Voronoi diagram. Boundary constraints of the effective facial region are constructed based on the facial feature points. All acupoints are used as seed points. The polygonal region of each seed point is output as the acupoint region.

3. The facial skin texture recognition system integrating acupoint location features according to claim 1, characterized in that, The assessment of the facial skin's redness, dullness, oiliness, and roughness within each acupoint area includes: The skin image in Lab space is acquired. By analyzing the difference between the mean a-channel pixel value and the mean b-channel pixel value of all pixels in each acupoint area, as well as the dispersion of the a-channel pixel value of all pixels, the redness coefficient is determined to evaluate the redness of the facial skin. By analyzing the difference between the mean value of the b channel and the mean value of the L channel of all pixels in each acupoint area, as well as the contrast of all pixels in the L channel, a dullness coefficient is determined to assess the dullness of facial skin. The number of highlight pixels in each acupoint area is counted. Based on the proportion of all highlight pixels in each acupoint area and the distribution of gray values ​​of pixels in all pore areas in each acupoint area, an oiliness coefficient is determined to assess the oiliness of facial skin. Based on the contrast of all pixels in the L channel within each acupoint area and the area ratio of the pore area, a roughness coefficient is determined to assess the roughness of facial skin.

4. A facial skin texture recognition system integrating acupoint location features according to claim 3, characterized in that, The skin texture feature vector for each acupoint area consists of a redness coefficient, a dullness coefficient, an oiliness coefficient, and a roughness coefficient.

5. A facial skin texture recognition system integrating acupoint location features according to claim 3, characterized in that, The method for determining highlight pixels is as follows: Calculate the product of the maximum value of the L-channel pixel value of all pixels in the skin image and a preset value, and record it as the brightness feature value. Calculate the preset quantile of the L-channel pixel value of all pixels in the skin image. Select the maximum value between the preset quantile and the brightness feature value as the highlight threshold. Pixels with L-channel pixel values ​​greater than the highlight threshold are designated as highlight pixels.

6. A facial skin texture recognition system integrating acupoint location features according to claim 3, characterized in that, The pore region was obtained by image segmentation of the skin image using an adaptive region growing algorithm.

7. A facial skin texture recognition system integrating acupoint location features according to claim 1, characterized in that, The method for quantifying the skin quality confidence level of each acupoint area is as follows: Calculate the mean similarity of each acupoint region with all other acupoint regions in its category on the skin texture feature vector, and denote it as the first feature value; Calculate the mean similarity of all acupoint regions within the category to which each acupoint region belongs, excluding each acupoint region itself, on the skin texture feature vector, and denote it as the second feature value; Calculate the ratio of the first eigenvalue to the second eigenvalue, and use the sum of the ratio and 1 as the confidence score of the skin quality for each acupoint region.

8. A facial skin texture recognition system integrating acupoint location features according to claim 1, characterized in that, The weighted fusion process of skin texture feature vectors and corresponding skin texture confidence scores for all acupoint regions includes: Calculate the scalar multiplication of the skin texture feature vector of all acupoint regions in the skin image with the corresponding skin texture confidence normalization value, concatenate the vectors after logarithmic multiplication, and record the concatenated result as the weighted skin texture feature vector.

9. A facial skin texture recognition system integrating acupoint location features according to claim 8, characterized in that, The method of identifying facial skin texture types using the fused results includes: The weighted skin texture feature vector of the person to be detected is used as the input of the pre-trained face skin texture recognition model, and the output is the skin texture result of the person to be detected.

10. A facial skin texture recognition system integrating acupoint location features according to claim 9, characterized in that, The process of obtaining a pre-trained facial skin texture recognition model is as follows: Obtain the weighted skin feature vectors and corresponding skin labels of a preset number of people with known skin types, and use them as input to train a machine learning classifier. The trained machine learning classifier is then used as a pre-trained face skin recognition model.

Images