Skin image red blood streak marking method, evaluation method and system

Abstract

The present invention discloses a skin image red blood streak marking method, evaluation method and system thereof. The marking method comprises the following steps: S1: obtaining a color skin image, and graying the skin image to obtain a G component image; S2: for the gray value in the G component image, based on the gray value probability density function, obtaining the expected value and variance of the probability density function of the gray value of the red blood streak part, and the expected value and variance of the probability density function of the gray value of the normal part; S3: based on the expected value and variance of the probability density function of the gray value of the red blood streak part, and the expected value and variance of the probability density function of the gray value of the normal part, marking the red blood streak in the G component image. The skin image red blood streak marking method and evaluation method provided by the present invention have high accuracy and reliability.

Description

Technical Field

[0001] The invention relates to a skin image red blood streak marking method, and also relates to a corresponding skin image red blood streak evaluation method and a detection system thereof, belonging to the technical field of skin image processing. Background Art

[0002] Red blood streaks, scientifically known as facial telangiectasia, are caused by the strong fragility of the capillary walls. The uneven expansion and rupture of the blood vessels cause the skin to turn red. The capillaries are visible to the naked eye. They are manifested as reddish-purple spots on the skin, which are mostly linear, branch-shaped, spider-shaped, star-shaped, dot-shaped, and papular-shaped. The diameter is generally less than 0.5 mm. Red blood streaks affect the patient's daily life. The patient's face is sensitive to temperature and light, and usually has a burning sensation. At the same time, red blood streaks affect the patient's appearance and cause a certain psychological burden on the patient.

[0003] As living standards continue to improve, people pay more and more attention to beauty and skin care. In recent years, the quality and performance of cosmetics have been continuously improved, and the technology of the medical beauty industry has been continuously developed. It is particularly important to evaluate the efficacy of cosmetics and detect and diagnose the problem of skin red blood streaks in medical beauty. At present, the evaluation of cosmetic efficacy and medical beauty diagnosis are mainly tested and analyzed through various methods such as biochemistry, cell biology, and clinical evaluation. Among them, the detection of red blood streaks is generally carried out through the visual diagnosis of clinical physicians. There is a lack of quantitative analysis between a large number of samples, which is affected by the subjective influence of physicians and the objectivity of the evaluation is insufficient. Although professional skin measurement instruments and equipment have high accuracy, the cost of use is relatively high. Summary of the invention

[0004] The primary technical problem to be solved by the present invention is to provide a method for marking red blood streaks in skin images.

[0005] Another technical problem to be solved by the present invention is to provide a method for evaluating red blood streaks in skin images.

[0006] Another technical problem to be solved by the present invention is to provide a skin image red blood streak detection system.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] According to a first aspect of an embodiment of the present invention, a method for marking red blood streaks in a skin image is provided, comprising the following steps:

[0009] S1: Obtain a color skin image, and perform grayscale preprocessing on the skin image to obtain a G component image;

[0010] S2: for the gray value in the G component image, based on the gray value probability density function, obtaining the expected value and variance of the probability density function of the gray value of the red blood part, and the expected value and variance of the probability density function of the gray value of the normal part;

[0011] S3: Marking the red blood streaks in the G component image based on the expected value and variance of the probability density function of the grayscale value of the red blood streak portion and the expected value and variance of the probability density function of the grayscale value of the normal portion.

[0012] Preferably, the method of obtaining the expected value and variance of the probability density function of the gray value of the red blood part, and the expected value and variance of the probability density function of the gray value of the normal part, comprises the following steps:

[0013] S21: establishing a probability model for the G component image, wherein the probability distributions of the red blood part and the normal part are both Gaussian distributions;

[0014] S22: Based on the probability model, curve fitting is used to obtain the expected value estimate and variance estimate of the probability density function of the grayscale value of the red blood part of the grayscale value probability density function, as well as the expected value estimate and variance estimate of the probability density function of the grayscale value of the normal part.

[0015] Preferably, before step S3, a step of enhancing the red blood feature of the gray value in the G component image is also included.

[0016] Preferably, the step of enhancing the red blood streak feature of the gray value in the G component image comprises the following steps:

[0017] After obtaining the expected value and variance of the probability density function of the gray value of the red blood streak part and the expected value and variance of the probability density function of the gray value of the normal part, the skin image is segmented using the gray interval μ1±2σ1 corresponding to the red blood streak as the segmentation boundary;

[0018] The grayscale value after segmentation is linearly stretched from the interval μ1±2σ1 to 0 to 255 to improve the contrast of the image.

[0019] Among them, the probability density function f1(x) of the gray value of the red blood part in the G component image has an expected value of μ1 and a variance of

[0020] According to a second aspect of an embodiment of the present invention, a method for evaluating red blood streaks in a skin image is provided, comprising the following steps:

[0021] Using the above-mentioned skin image red blood streak marking method, marking the red blood streak in the G component image;

[0022] The grayscale values ​​of the normal part and the red blood part are standardized so that the normal parts of different skin images have the same grayscale value expectation and variance;

[0023] Calculate the red blood streak significance evaluation index according to the gray value expectation and variance of the normal part after standardization and the gray value expectation and variance of the red blood streak part after standardization;

[0024] According to the red blood streak prominence evaluation index, the red blood streak evaluation value is calculated to evaluate the skin image.

[0025] Preferably, the step of marking the red blood streaks in the G component image further includes the step of enhancing the red blood streak features of the gray value in the G component image:

[0026] After obtaining the expected value and variance of the probability density function of the gray value of the red blood streak part and the expected value and variance of the probability density function of the gray value of the normal part, the skin image is segmented using the gray interval μ1±2σ1 corresponding to the red blood streak as the segmentation boundary;

[0027] The grayscale value after segmentation is linearly stretched from the interval μ1±2σ1 to 0 to 255 to improve the contrast of the image;

[0028] Among them, the probability density function f1(x) of the gray value of the red blood part in the G component image has an expected value of μ1 and a variance of

[0029] Preferably, the grayscale values ​​of the normal part and the red blood part are standardized by performing a linear transformation on the grayscale values.

[0030] Preferably, the red blood streak significance evaluation index is a comparison between the grayscale value of the red blood streak portion and the grayscale value distribution of normal skin; wherein, the more the grayscale value of the red blood streak portion deviates from the grayscale value of the normal portion, the more severe the red blood streak is.

[0031] According to a third aspect of an embodiment of the present invention, a skin image red blood streak detection system is provided, comprising:

[0032] An acquisition module, used to take a picture of the skin with standard shooting parameters to obtain an original skin image of a predetermined pixel size;

[0033] An image processing module, used for executing the aforementioned skin image red blood streak marking method or skin image red blood streak evaluation method on the original skin image provided by the acquisition module;

[0034] The display module is used to display the mark or evaluation results.

[0035] Compared with the prior art, the present invention is based on the double Gaussian fitting method, applies digital image processing technology, combines probability theory and mathematical statistics theory, and conducts experimental analysis on the pixel gray value distribution characteristics of the red blood streak image; and implements the algorithm for feature enhancement, identification mark and quantitative evaluation functions of the red blood streak part. Experiments have proved that the present invention has the following technical effects: (1) Under cross-polarized light irradiation, the red blood streak features in the G component of the facial skin image containing red blood streak features are more obvious, and the gray value range of the red blood streak is lower than the gray mean of the image; (2) The probability distribution form of the image gray value approximately obeys the double Gaussian distribution. When the pixel resolution is 1400*1400 and above, the frequency distribution histogram of the image gray value is fitted with double Gaussians, and the probability distribution parameters of the image can be effectively estimated; (3) The proposed red blood streak detection algorithm has significant effects on feature enhancement and identification mark of the red blood streak in the skin image; the evaluation value calculated by the algorithm has a high accuracy in quantitative evaluation of the severity of the red blood streak in the skin image. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flowchart of a method for marking red blood streaks in a skin image provided by the first to third embodiments of the present invention;

[0037] Figure 2A is the original color image of the original skin image;

[0038] Figure 2B for Figure 2A The R component image after grayscale processing;

[0039] Figure 2C for Figure 2A The G component image after grayscale processing;

[0040] Figure 2D for Figure 2A B component image after grayscale processing;

[0041] Figure 3A is a schematic diagram of the G component graph;

[0042] Figure 3B For Figure 3A Schematic diagram of the segmentation range of grayscale values ​​from 0 to 39 after image segmentation;

[0043] Figure 3C For Figure 3A Schematic diagram of the segmentation range of grayscale values ​​between 40 and 52 after image segmentation;

[0044] Figure 3D For Figure 3A Schematic diagram of the segmentation range of grayscale values ​​between 53 and 255 after image segmentation;

[0045] Figure 4 A schematic diagram showing the bimodal characteristics of the grayscale histogram of the grayscale image;

[0046] Figure 5A An example of a double Gaussian fitting curve and parameter estimation value obtained by fitting the gray value data of the red blood part in a set of skin images;

[0047] Figure 5B is an example of a double Gaussian fitting curve and parameter estimation value obtained by fitting the gray value data of the red blood part in another set of skin images;

[0048] Figure 6 A flowchart of a method for marking red blood streaks in a skin image provided by the fourth and fifth embodiments of the present invention;

[0049] Fig. 7A This is a schematic diagram of the red blood part in the original skin image;

[0050] Figure 7B For Fig. 7A Schematic diagram of the feature enhancement effect of the red blood part;

[0051] Figure 7C For Figure 7B Schematic diagram of the marking effect of the red blood part;

[0052] Figure 8 To verify the expert evaluation value in the effect - calculate the evaluation value scatter plot;

[0053] Fig. 9 This is a structural block diagram of the skin image red blood streak detection system provided by the present invention. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0055] The basic idea of ​​the present invention is that under cross-polarized light illumination, the red blood streak features are more obvious in the G component of the facial skin image containing red blood streak features, and the grayscale value range of the red blood streak is lower than the grayscale mean value of the image. On this basis, the present invention performs a characteristic analysis of the grayscale value distribution of the skin image containing red blood streak features based on the double Gaussian fitting method, and proposes a skin image red blood streak marking method for feature enhancement, identification marking and quantitative evaluation of the red blood streak part in the image.

[0056] <First Embodiment>

[0057] like Figure 1 As shown, the method for marking red blood streaks in skin images provided in this embodiment includes the following steps.

[0058] S1: Acquire a color skin image and perform grayscale preprocessing on the skin image.

[0059] Since each pixel of a color image is composed of three components, RGB, such as Figure 2A Therefore, by separating the RGB components in the image, the original image can be divided into three component images, among which, Figure 2B Display the R component graph; Figure 2C Display the G component graph; Figure 2D Displays the B component graph.

[0060] Because the skin color reflectivity in the green light band is affected by the hemoglobin in the dermis, the grayscale value of the G component of the skin image can reflect the content and distribution of hemoglobin in the dermis. Under cross-polarized light, the skin red blood image has no obvious features in the R and B components, but has obvious features in the G component. This is shown by Figure 2B , Figure 2C , Figure 2D Therefore, the present invention uses the G component of each pixel in the original image as the gray value of the pixel, grays the image, and obtains a grayscale image.

[0061] S2: For the grayscale values ​​in the preprocessed G component image, based on the grayscale value probability density function, obtain the expected value and variance of the probability density function of the grayscale values ​​of the red blood part, and the expected value and variance of the probability density function of the grayscale values ​​of the normal part.

[0062] As is known to all, hemoglobin has an additional absorption effect on the green light band, so the area with a higher hemoglobin content has a lower reflectivity to green light. Therefore, in the G component image obtained after preprocessing, the grayscale of the red blood part should be darker than that of the normal part. Here, assuming that the original image has no obvious features except the red blood (that is, there are no other skin problems in the original skin image that may affect the grayscale G component), the image can be divided into a normal part and a red blood part.

[0063] To verify that the grayscale value of the red blood part is lower than that of the normal part in the G component grayscale value, the inventor conducted an experiment. The specific experimental process is as follows:

[0064] 1) Visually mark the pixels of the red blood part in the original image.

[0065] 2) The grayscale value of the G component is segmented according to a specific grayscale threshold, and pixels within each threshold range are extracted.

[0066] like Figure 3A to Figure 3D As shown, in this embodiment, the grayscale threshold of the G component image is set to 0-39, 40-52, 53-255. Figure 3A) are segmented according to the grayscale value of the pixel, and the segmentation range is: pixels with grayscale values ​​of 0 to 39 ( Figure 3B ), a segmentation range of grayscale values ​​of 40 to 52, and a segmentation range of grayscale values ​​of 53 to 255. Then, pixels within each threshold range are extracted.

[0067] 3) Position matching of the extracted pixels and the marked red blood streak pixels.

[0068] Figure 3B to Figure 3D One of the images is used to schematically show the segmentation and extraction effect. As can be seen from the figure, the grayscale value of the pixel with successful pixel position matching is generally 10 to 20 lower than the grayscale average of all pixels in the G component image, which confirms that the color of the red blood part in the G component grayscale value is darker than the normal part and the grayscale value is smaller. The aforementioned 10 to 20 lower is a rough estimate of multiple images. The actual grayscale value difference varies depending on the image quality. The position matching is only to verify that the grayscale value of the red blood part is lower.

[0069] Based on the aforementioned characteristics of the G component image, the grayscale histogram of the G component grayscale value presents a bimodal feature (the two peak areas have different grayscale means and variances), such as Figure 4 shown. Figure 4 The peak area in the dotted box has a smaller grayscale value, which corresponds to the red blood part in the skin image; the peak area on the right side of the dotted box corresponds to the normal part of the skin image. Therefore, this feature can be used to identify and mark the red blood part of the skin image.

[0070] Based on the above description, the specific steps of performing double Gaussian fitting on the preprocessed G component map are introduced below:

[0071] S21: Establish a probability model for the grayscale value of the G component of each pixel in the image

[0072] In image processing technology, the gray value of each pixel in an image can be regarded as a random variable, which obeys a specific probability distribution. In the embodiment of the present invention, the image is divided into a red blood part and a normal part, and it is assumed that the gray values ​​of pixels in the same part obey independent distribution, and a probability model is established for the gray value of the G component of each pixel in the whole image.

[0073] When the image is captured, the grayscale optical signal acquired by the device is a continuous value. When digitally stored, the optical signal is discretized, resulting in the stored data being discrete values. Taking into account the continuity of its signal value, in an embodiment of the present invention, the probability distribution form of the grayscale value is processed as a continuous random variable with a value range of [0,256). The grayscale value interval [0,256) is divided into 256 groups (n=0,1,…,255) with a group interval of 1 (d=1), and the interval boundaries of each group are closed on the left and open on the right. The interval [n,n+1) corresponds to the discrete grayscale value n.

[0074] Since skin images are affected by a large number of environmental factors such as skin condition and light intensity, the present invention assumes that the influencing factors are independent of each other and each influencing factor does not have a dominant influence on the grayscale value of the image. The grayscale value of the image should satisfy the central limit theorem and approximately obey the normal distribution.

[0075] The red blood part or the normal part of the skin (analyzed separately), according to the central limit theorem, the probability distribution of the gray value of the two parts is the same Gaussian distribution. The red blood part is more sensitive than the normal part and is affected by some additional factors such as ambient temperature. Therefore, the two parts have different expected μ and variance σ 2 That is, the expected value of the probability density function f1(x) of the gray value of the red blood part in the G component image is μ1, and the variance is The expected value of the probability density function f2(x) of the grayscale value of the normal part of the G component image is μ2, and the variance is

[0076] The probability density function of the grayscale value of the red blood part is expressed by the following equation (1), and the probability density function of the grayscale value of the normal part is expressed by the following equation (2).

[0077]

[0078]

[0079] Where x1 represents the gray value of each pixel in the red blood part; x2 represents the gray value of each pixel in the normal part.

[0080] When the entire image is considered, an additional parameter, the probability of the red blood part, p, is introduced to represent the probability of the event "when a pixel point in the image is selected, the pixel point belongs to the red blood part". Then the probability density function f(x) of the gray value of each pixel in the entire skin image can be expressed by formula (3):

[0081] f(x)=pf1(x)+(1-p)f2(x) Formula (3)

[0082] Wherein, x is the gray value of each pixel in a skin image, and p is the expected value of the proportion of the number of pixels of the red blood part in all pixels of the skin image.

[0083] S22: Based on the probability model, the parameter estimation value of the gray value probability density function is obtained by curve fitting.

[0084] The frequency distribution histogram reflects the frequency distribution law of the sample. When the sample size tends to infinity, the frequency distribution of the random variable is its probability distribution. Therefore, if the sample size is large enough, the frequency distribution histogram can be curve fitted to estimate the parameters of the probability distribution with a known form.

[0085] The pixel resolution of the original skin image processed by the embodiment of the present invention is 1400*1400, and the sample capacity of each original image is large enough to approximate the frequency distribution of the grayscale value of the skin image to its probability distribution. At the same time, the probability density function of the grayscale value is known to be a double Gaussian distribution, and the sample population composed of all pixels in the entire image can be used to estimate the parameters of its probability density function.

[0086] The double Gaussian curve is fitted to the discrete points in the gray value frequency distribution histogram. The fitting formula is shown in formula (4):

[0087]

[0088] According to the parameters a1, a2, b1, b2, c1, c2 obtained by curve fitting, the parameter estimation value of the gray value probability density function can be solved, as shown in formula (5):

[0089]

[0090] in, represents the estimated value of the probability p of the red blood part; The estimated value of the expected value μ1 of the probability density function f1(x) representing the gray value of the red blood part; The variance of the probability density function f1(x) of the gray value of the red blood part is the estimated value of σ1; The expected value μ2 of the probability density function f2(x) representing the gray value of the normal part; Represents the estimated value of the variance σ2 of the probability density function f2(x) of the grayscale values ​​of the normal part.

[0091] For the convenience of symbolic representation, parameter symbols are used to represent both parameters and parameter estimates in the following text; for the convenience of observing the curve fitting effect, the histogram is represented in the form of scattered points. The parameter estimates of the frequency distribution histogram, double Gaussian fitting curve and probability density function (expected value, variance and expected value of the number of pixels in the red blood part) are shown in the figure. Figure 5A and Figure 5B shown. Figure 5A and Figure 5B They respectively represent the double Gaussian fitting curves obtained by fitting the gray value data of the red blood part in a group of skin images, and further obtain the parameter estimation values.

[0092] S3: Marking the red blood streaks in the G component image based on the expected value and variance of the probability density function of the grayscale value of the red blood streak part and the expected value and variance of the probability density function of the grayscale value of the normal part.

[0093] There is a common overlap between the grayscale value distribution intervals of the red blood part and the normal part. In order to ensure that the marking accuracy is high enough, when marking the red blood part according to the grayscale value, the confidence of the grayscale value of the normal part in the grayscale value segmentation interval should be low. For this reason, the lower limit of the threshold interval is μ1-2σ1, and the upper limit is min{μ1+2σ1,μ2-2σ2}. The pixels in the original image whose grayscale value is within the threshold interval are marked as black (grayscale value is 0), and the pixels outside the threshold interval are marked as white (grayscale value is 255), so as to effectively mark the red blood part in the skin image. Based on the marked image, beauty, makeup or treatment operations can be performed on the skin of the red blood part.

[0094] <Second Embodiment>

[0095] Another method for marking red blood streaks in a skin image provided by this embodiment includes the following steps:

[0096] S1: Acquire a color skin image and perform grayscale preprocessing on the skin image.

[0097] This step is the same as that in the first embodiment and will not be described in detail here.

[0098] S2A: For the grayscale values ​​in the preprocessed G component image, based on the grayscale value probability density function, the expected value and variance of the probability density function of the grayscale values ​​of the red blood part and the expected value and variance of the probability density function of the grayscale values ​​of the normal part are obtained.

[0099] Here, the expected value and variance of the probability density function of the gray value of the red blood part and the expected value and variance of the probability density function of the gray value of the normal part can be obtained directly based on empirical values; or, through data collection, in a neural network deep learning manner, to obtain the parameter estimation value (expected value, variance) of the probability density function of the red blood part and the parameter estimation value (expected value, variance) of the probability density function of the normal part. It can be understood that other methods can also be used to calculate the expected value and variance of the probability density function of the red blood part and the expected value and variance of the probability density function of the normal part.

[0100] S3: Marking the red blood streaks in the G component image based on the expected value and variance of the probability density function of the grayscale value of the red blood streak part and the expected value and variance of the probability density function of the grayscale value of the normal part.

[0101] Based on the expected value and variance of the probability density function of the red blood part obtained in step S2A and the expected value and variance of the probability density function of the normal part, a threshold is set, with the lower limit of the threshold interval being μ1-2σ1 and the upper limit being min{μ1+2σ1,μ2-2σ2}. Pixels in the original image whose grayscale values ​​are within the threshold interval are marked as black (grayscale value is 0), and pixels outside the threshold interval are marked as white (grayscale value is 255), thereby effectively marking the red blood part in the skin image.

[0102] <Third Embodiment>

[0103] like Figure 1 As shown, the method for evaluating red blood streaks in skin images provided in this embodiment includes the following steps:

[0104] S1: Acquire a color skin image and perform grayscale preprocessing on the skin image.

[0105] S2: For the grayscale values ​​in the preprocessed G component image, based on the grayscale value probability density function, obtain the expected value and variance of the probability density function of the grayscale values ​​of the red blood part, and the expected value and variance of the probability density function of the grayscale values ​​of the normal part.

[0106] S3: Marking the red blood streaks in the G component image based on the expected value and variance of the probability density function of the grayscale value of the red blood streak part and the expected value and variance of the probability density function of the grayscale value of the normal part.

[0107] The above steps may use the skin image red blood streak marking method of the first embodiment or the second embodiment to mark the red blood streak in the G component image.

[0108] S4: Standardize the grayscale values ​​of the normal part and the red blood part

[0109] Affected by factors such as individual skin color and shooting environment, the grayscale value distribution of normal skin varies in different images. Therefore, before quantitative evaluation, the grayscale value of the skin image needs to be standardized and the grayscale value needs to be linearly transformed so that the normal parts of different skin images have the same grayscale value expectation and variance. The probability density function of the grayscale value of the G component image after standardization is shown in formula (6):

[0110] f st (x)=αf(α(x+β))=αpf1(α(x1+β))+α(Ip)f2(α(x2+β))

[0111] = pf r +(1-p)f n Formula (6)

[0112] Among them, α and β are transformation coefficients, which are set according to experience; f r is the probability density function of the gray value of the standardized red blood part (also recorded as f r (x1)), f n is the probability density function of the gray value of the normal part after standardization (also recorded as f n (x2)), as shown in formula (7):

[0113]

[0114] where μ n , is the expected value and variance of the gray value of the normal part after standardization; μ r , is the expected gray value and variance of the red blood part after standardization. The expected gray value of the normal part is μ n and variance σ n is an empirical value. r , σ r The calculation formula of is shown in formula (8):

[0115]

[0116] S5: Calculate the red blood streak significance evaluation index according to the gray value expectation and variance of the normal part after standardization and the gray value expectation and variance of the red blood streak part after standardization.

[0117] After the skin image is standardized, the probability of a pixel belonging to the red blood part is still p, which reflects the area of ​​the red blood region in the skin image; the grayscale expected value of the red blood part is μ r , reflects the gray depth of the red blood part in the skin image; the gray standard deviation of the red blood part σ r , reflecting the grayscale distribution range of the red blood part in the skin image.

[0118] Where p, μ r It quantitatively reflects the severity of red blood streaks on the skin from a certain perspective, and is used to quantify the area and depth of red blood streaks. r , due to the interference of the gray value distribution of the normal part, the gray value distribution of the red blood part cannot directly reflect the severity of the red blood, and needs to be compared with the gray value distribution of normal skin. The more the gray value of the red blood part deviates from the gray value of the normal part, the more serious the red blood is. Here, the gray significance function k(x) is introduced, as shown in formula (9):

[0119]

[0120] The grayscale saliency function reflects the grayscale value x and μ of the red blood part in the skin image. n As the gray value x gradually approaches μ from low to high n , the value of k(x) decays exponentially. When x is μ n -2σ n , μ n When the k value is 1 and 0.5, the grayscale value of the pixels in the normal part of the corresponding skin is greater than the confidence level of x by 98% and 50%.

[0121] The grayscale significance function k(x) is used as the integral weight of the grayscale value of each pixel in the red blood part of the skin image when evaluating the severity of the red blood. The red blood part in the standardized probability density function is integrated to obtain the red blood significance evaluation index K. From formula (7), we know that the grayscale value of the red blood part is mainly distributed in μ r ±2σ r Let μ r ±2σ r is the integral interval, and the gray significance evaluation index K is approximately calculated. The calculation formula is shown in formula (10):

[0122]

[0123] S6: Calculate the red blood streak evaluation value according to the red blood streak prominence evaluation index and evaluate the skin image.

[0124] In this embodiment, five skin images with different degrees of red blood streaks are selected as the reference group. Three experts are asked to conduct subjective evaluation of the red blood streak status of each image in the reference group (on a 100-point scale, and the evaluation results are averaged to reduce the evaluation error), and the expert evaluation value of red blood streaks is obtained. r , K are respectively used as the evaluation indexes of the area, depth and prominence of red blood streaks in each image, and are calculated by the algorithm. The reference group images are shown in Figure 7, and the parameter estimation values, evaluation index calculation values ​​and expert evaluation values ​​of each image are shown in Table 1.

[0125] Table 1 Parameter estimation values, evaluation index calculation values, and expert evaluation values ​​of the reference group images

[0126]

[0127] Combined with the calculated values ​​of the evaluation indicators, a multivariate linear regression is performed on the expert evaluation values. The regression equation is shown in formula (11):

[0128] Calculated Evaluation=a1p+a2μr +a3K+a4 Formula (11)

[0129] The least squares method is used to estimate the regression parameters and obtain the final calculation evaluation function. The calculation results of regression parameters a1, a2, a3, and a4 are: 33.0537, -3.1564, 13.1156, and 243.0869 respectively.

[0130] Therefore, by calculating the evaluation function (Formula 11), after marking each pixel of the red blood streak part in the skin image, the red blood streak evaluation value (Calculated Evaluation) can be directly calculated to evaluate the skin image.

[0131] <Fourth Embodiment>

[0132] As mentioned above, the quality of skin images is not uniform due to skin image shooting environment, shooting parameters, etc. Therefore, in this embodiment, before marking the red blood streaks, a red blood streak feature enhancement step is added to improve the accuracy and wide applicability of the red blood streak marking.

[0133] like Figure 6 As shown, the skin image red blood streak marking method provided in this embodiment includes the following steps:

[0134] S1: Acquire a color skin image and perform grayscale preprocessing on the skin image.

[0135] S2: For the grayscale values ​​in the preprocessed G component image, based on the grayscale value probability density function, obtain the expected value and variance of the probability density function of the grayscale values ​​of the red blood part, and the expected value and variance of the probability density function of the grayscale values ​​of the normal part.

[0136] The above steps may adopt the method of the skin image red blood streak marking method of the first embodiment or the second embodiment, which will not be described in detail here.

[0137] S2B: Enhance the red blood feature of the gray value in the G component image.

[0138] In the above steps, after obtaining the probability density functions of the gray values ​​of the red blood part and the normal part of the skin image, μ1, σ1, μ2, and σ2, the image is segmented by the threshold method to separate the red blood part from the entire image. It is known that for the normal distribution X~N(μ, σ), the confidence interval μ±2σ is 95%. Therefore, from formula (1) and formula (2), it can be seen that the gray value of the red blood part is mainly distributed in μ1±2σ1; the gray value of the normal part is mainly distributed in μ2±2σ2. Based on the above gray value distribution range, the skin image is enhanced and marked with red blood features.

[0139] The skin image is segmented using the grayscale interval μ1±2σ1 corresponding to the red blood streaks as the segmentation boundary. The grayscale value after segmentation is linearly stretched from the interval μ1±2σ1 to 0 to 255 to improve the image contrast, thereby enhancing the features of the red blood streaks in the skin image. The effect is shown in Figure 7.

[0140] S3: Marking the red blood streaks in the G component image based on the expected value and variance of the probability density function of the grayscale value of the red blood streak part and the expected value and variance of the probability density function of the grayscale value of the normal part.

[0141] <Fifth Embodiment>

[0142] like Figure 6 As shown in Figure 7, based on the third embodiment, the skin image red blood streak evaluation method provided in this embodiment can also add a red blood streak feature enhancement step for skin images with insufficient image quality. The operation of this step is similar to step S2B in the fourth embodiment and will not be repeated here.

[0143] The measured effects of the present invention are described below in conjunction with specific experimental data.

[0144] 228 skin images with red blood streaks were selected as the test group to test the reliability of the evaluation function.

[0145] First, three experts were asked to subjectively evaluate the red blood state of each image in the test group using the reference group images and corresponding evaluation values ​​as reference standards to obtain expert evaluation values; then, the method of the fifth embodiment was used for each image to calculate the red blood evaluation value (Calculated Evaluation) according to the evaluation function as the red blood score. The evaluation indicators and evaluation values ​​corresponding to some images in the test group are shown in Table 2.

[0146] Table 2 Evaluation index calculation values, expert evaluation values ​​and calculated evaluation values ​​of the test group images (parts)

[0147] serial number p <![CDATA[μ r ]]> K Expert evaluation value Calculate the evaluation value 019 0.444 72.81 0.601 18.8 35.8 035 0.517 70.84 0.725 25.6 46.1 067 0.334 73.24 0.577 27.1 30.5 080 0.641 70.50 0.870 39.6 53.2 124 0.469 73.10 0.624 40.0 36.0 141 0.11 61.79 0.278 65.9 55.4 163 0.533 67.72 0.956 54.2 59.5 196 0.122 66.38 0.256 47.8 40.9 202 0.406 68.55 0.615 58.8 48.2

[0148] The calculated evaluation value and the expert evaluation value were regressed to verify the reliability of the algorithm. Figure 8 As shown, the correlation coefficient calculation result is 0.8742, which has good consistency; the normalized residual standard error is 0.0861, which has high accuracy. It can be seen that the skin image red blood streak marking method and evaluation method provided by the present invention have high accuracy and reliability.

[0149] like Fig. 9As shown, the present invention also provides a skin image red blood streak detection system, which includes an acquisition module, an image processing module and a display module. Among them, the acquisition module can be implemented by a high-definition camera, which is used to take pictures of the skin with standard shooting parameters to obtain an original skin image of a predetermined pixel size. The image processing module can be implemented by a single-chip microcomputer or a microcontroller system. The image processing module is connected to the acquisition module and is used to perform image processing on the original skin image provided by the acquisition module according to the skin image red blood streak marking method or the skin image red blood streak evaluation method in the first to fifth embodiments described above. The display module can be implemented by an LCD or OLED display screen, which is used to connect the image processing module to display the marking or evaluation results.

[0150] The above is a detailed description of the skin image red blood streak marking method, evaluation method and system provided by the present invention. For those skilled in the art, any obvious changes made to it without departing from the essence of the present invention will constitute an infringement of the patent right of the present invention and will bear corresponding legal responsibilities.

Claims

1. A method for evaluating red blood streaks in skin images, characterized in that The following steps are involved: S1: obtaining a color skin image, and performing grayscale preprocessing on the skin image to obtain a G component image; S2: for the gray value in the G component image, based on the gray value probability density function, obtaining the expected value and variance of the probability density function of the gray value of the red blood part, and the expected value and variance of the probability density function of the gray value of the normal part; S3: marking the red blood streaks in the G component image based on the expected value and variance of the probability density function of the grayscale value of the red blood streak portion and the expected value and variance of the probability density function of the grayscale value of the normal portion; S4: Standardize the grayscale values ​​of the normal part and the red blood part so that the normal parts of different skin images have the same grayscale value expectation and variance; S5: Calculate the red blood streak significance evaluation index according to the gray value expectation and variance of the normal part after standardization and the gray value expectation and variance of the red blood streak part after standardization; S6: Calculate the red blood evaluation value based on the red blood prominence evaluation index and evaluate the skin image. Probability density function of the gray value of the red blood part The probability density function of the gray value of the normal part is expressed by the following formula (1): It is represented by the following formula (2): In the formula, x1 represents the gray value of each pixel in the red blood part; x2 represents the gray value of each pixel in the normal part; The expected value of And the variance is ; The expected value of And the variance is ; The grayscale values ​​of the normal part and the red blood part are standardized by calculating the probability density function of the grayscale value of the G component image after standardization: ， in is the transformation coefficient, which is set according to the empirical value; is the probability density function of the gray value of the red blood part after standardization, is the probability density function of the gray value of the normal part after standardization, is the probability of gray value of the red blood part; Satisfies formula (7): in is the expected value and variance of the gray value of the normal part after standardization; , is the expected gray value and variance of the red blood part after standardization; the expected gray value of the normal part is and variance is the experience value, , The calculation formula of is shown in formula (8): 。 2. The method for evaluating red blood streaks in skin images according to claim 1, characterized in that The method of obtaining the expected value and variance of the probability density function of the gray value of the red blood part, and the expected value and variance of the probability density function of the gray value of the normal part, comprises the following steps: S21: establishing a probability model for the grayscale value of the G component of each pixel in the G component image, wherein the probability distribution of the red blood part and the normal part are both Gaussian distributions; S22: Based on the probability model, curve fitting is used to obtain the expected value estimate and variance estimate of the probability density function of the grayscale value of the red blood part of the grayscale value probability density function, as well as the expected value estimate and variance estimate of the probability density function of the grayscale value of the normal part.

3. The method for evaluating red blood streaks in skin images according to claim 1 or 2, characterized in that Before step S3, the method further includes a step of enhancing the red blood feature of the gray value in the G component image.

4. The method for evaluating red blood streaks in skin images according to claim 3, characterized in that The step of enhancing the red blood streak feature of the gray value in the G component image comprises the following steps: After obtaining the expected value and variance of the probability density function of the gray value of the red blood part and the expected value and variance of the probability density function of the gray value of the normal part, the gray interval corresponding to the red blood part is calculated. As the segmentation boundary, the skin image is segmented; The gray value after segmentation is divided into Linearly stretch from 0 to 255 to increase the contrast of the image. Among them, the probability density function of the gray value of the red blood part in the G component image is The expected value of , the variance is .

5. The method for evaluating red blood streaks in skin images according to claim 4, characterized in that: The red blood streak significance evaluation index is a comparison of the grayscale value of the red blood streak portion with the grayscale value distribution of normal skin; wherein, the more the grayscale value of the red blood streak portion deviates from the grayscale value of the normal portion, the more serious the red blood streak is.

6. A skin image red blood streak detection system, characterized in that It includes an acquisition module, an image processing module and a display module; wherein, An acquisition module, used to take a picture of the skin with standard shooting parameters to obtain an original skin image of a predetermined pixel size; An image processing module is connected to the acquisition module and the display module respectively, and is used to execute the skin image red blood streak evaluation method according to any one of claims 1 to 5 on the original skin image provided by the acquisition module; and a display module is used to display a mark or an evaluation result.

Images