CN103034874A - Face gloss analytical method based on inspection diagnosis of traditional Chinese medical science - Google Patents

Abstract

The invention discloses a face gloss analytical method based on inspection diagnosis of traditional Chinese medical science. The method sequentially comprises the following steps of automatically collecting face images by using a digital camera controlled by a computer in a stable light environment and automatically storing the images in the computer; obtaining the face images from an area at the position of the cheek and adjusting resolution to be same; using a hue, saturation and value color pattern to conduct feature extraction on test samples and training samples of the face images by using an improved two-dimensional principal component analysis method on a standard face image gloss image and a test sample image; and using a nearest neighbor method to conduct qualitative and quantitative analysis on the test samples by calculating a cosine distance of characteristics of the test samples and the training samples after dimensionality reduction. The face gloss analytical method can directly conduct quantitative analysis and qualitative description of the gloss of the face images and can assist diagnosis of traditional Chinese medical.

Description

Facial gloss analysis method based on traditional Chinese medicine inspection

technical field

The invention belongs to the field of traditional Chinese medicine diagnosis digitization and computer medical image processing, and specifically relates to a facial gloss analysis method based on traditional Chinese medicine inspection. The method of the invention can directly perform quantitative analysis and qualitative description of the gloss on facial images, and assists traditional Chinese medicine diagnosis.

Background technique

Facial inspection is a method of TCM diagnosis. Traditional Chinese medicine has always attached great importance to facial inspection, which mainly includes two parts: inspection of color and inspection of luster. "Lingshu" says that "the twelve meridians, three hundred and sixty-five collaterals, the blood and qi all go up the face and go through the hole", the heart governs the blood, its flowers are on the face, and the three yang meridians of the hands and feet all go up on the head and face, so the facial color Diagnosis can diagnose the ups and downs of visceral essence and the profit and loss of meridian qi and blood. "Plain Questions Theory on the Essence of the Pulse" says: "A man who is astute in the five colors has the brilliance of Qi." It is believed that the essence of qi and blood ascends to the head and face and outwards to the skin, and can show different colors. "Looking and Diagnosis Zunjing · Color Based on Moisture" says: "Those who are bright and moist have qi, and those who are blue, red, yellow, white and black have color. If there is qi, there is no color, and if there is color, there is no qi." Observing the changes in facial luster can diagnose the ups and downs of visceral essence, which is very important for judging the severity of the disease and predicting the prognosis. However, the traditional diagnosis of facial gloss mainly relies on clinicians’ subjective descriptions, which are described as shiny, less shiny, and dull, lacking objective data support, and has strong subjectivity and ambiguity, which will inevitably affect the overall development of TCM diagnosis. . The modernization and objective research of TCM face-to-face diagnosis has important theoretical value and clinical significance for the standardization of TCM syndrome differentiation, clinical curative effect evaluation and further development of TCM face-to-face diagnosis.

With the continuous advancement of science and technology, computer technology, especially pattern recognition, computer vision, data mining and other technologies, has been gradually introduced into the objectivity and standardization of TCM research, and staged results have been achieved. The gloss quantification of facial examination is an important aspect of facial examination. After searching the prior art literature, it is found that there is no method and technical report on gloss analysis of facial examination.

Contents of the invention

The technical problem to be solved by the present invention is to provide a facial gloss analysis method for face-to-face diagnosis of traditional Chinese medicine, aiming to automatically analyze the gloss of collected facial images of patients. From the perspective of pattern recognition, the present invention assists the diagnosis of traditional Chinese medicine by reducing the dimension of sample images, calculating the cosine distance between test samples and training samples, and using the nearest neighbor method to distinguish the category and quantitative value of the test sample gloss.

In order to solve the above-mentioned technical problems, the idea of the present invention is: use the qualitative glossy sample picture as the training sample, obtain the projection matrix by using an improved two-dimensional principal component analysis on the training sample, and use this projection matrix to train the glossy picture Samples and test samples are used for feature extraction, and the cosine distance of all training sample features is calculated for the test sample features. Finally, the nearest neighbor method is used to perform gloss analysis on the test samples.

According to above-mentioned inventive conception, the technical solution adopted in the present invention comprises the following steps:

(1) The facial image is automatically collected by a computer-controlled digital camera under a stable lighting environment to ensure the stability of the collection environment;

(2) The sample images are taken from an area on the cheek of the facial image, and all are adjusted to the same resolution;

(3) Use HSV color mode, if the sample picture is in RGB mode, it will be converted to HSV color mode;

(4) Select a part of the picture as a training sample, and manually classify the gloss;

(5) Adopt the improved two-dimensional principal component analysis method to carry out feature extraction to the test sample and the training sample of facial image;

(6) Calculate the cosine distance between the test sample feature and each training sample feature;

(7) Perform facial gloss analysis on the test sample using the nearest neighbor method.

The method for feature extraction of the test sample and the training sample of the face image using the improved two-dimensional principal component analysis method is as follows:

① Use matrix I to represent samples: if I _{a, b} represent the elements at row a and column b of matrix I, then use matrix I with m rows and 3n columns to represent a picture in HSV color mode with a resolution of n×m, I _{3i-2, j} , I _{3i-1, j} , I _{3i, j} respectively store the H, S, and V values of the pixel at row i, column j of the image (i=1, 2, 3, ..., m, j=1,2,3,...,n);

其中k表示样本的类别数，n_i为第i类的训练样本数，N为训练样本总数，为第i类训练样本的样本均值，I_i为训练样本总体均值；②Use an improved method to calculate the covariance matrix of the two-dimensional analysis based on the training samples:

Where k represents the number of categories of samples, n _i is the number of training samples of the i-th category, N is the total number of training samples, Is the sample mean value of the i-th type of training sample, and I _i is the overall mean value of the training sample;

③ Find the eigenvalue decomposition of the covariance matrix Cov, and arrange the obtained eigenvalues in descending order: λ ₁ , λ ₂ K λ _3n , and the eigenvectors are sorted in the order of the corresponding eigenvalues, so the eigenvector matrix is obtained: V=[v ₁ v ₂ K v _3n ] ^T ;

④ select the first d rows of V to form a projection matrix: V _project = [v ₁ v ₂ K v _d ] ^T ;

⑤ Perform feature extraction on the sample according to the projection matrix V _project to obtain the sample feature F = V _project ^IT , so the feature set of the training sample is: F _train = {F ₁ , F ₂ K F _N }, and the test sample feature is F _test .

The above calculation of the cosine distance between the test sample feature and each training sample feature is to find the cosine value of the angle between all samples in F _test and F _train , and take the absolute value of the result.

The above-mentioned nearest neighbor method performs facial gloss analysis on the test sample, that is, selects the category of the training sample feature with the smallest cosine distance from Ftest as the category of the test sample.

The present invention has the following obvious outstanding substantive features and significant advantages:

The method of the present invention can obtain a higher matching accuracy rate: due to the glossy image samples of known categories, the existing samples are fully utilized, and the improved two-dimensional principal component analysis is a supervised learning method, which can quickly It is better to use training samples to extract more discriminative gloss features. This method is suitable for automatic gloss quantitative and qualitative analysis of face test sample pictures when there are a certain number of qualitative gloss pictures, and can obtain a relatively high test accuracy rate, which has a certain role in promoting the objectification and modernization of TCM face-to-face diagnosis. It has more practical use value.

Description of drawings

Fig. 1 is the flow chart of the facial gloss analysis method based on traditional Chinese medicine inspection of the present invention.

Fig. 2 is a computer automatic interpretation result diagram of the facial gloss analysis method based on inspection in traditional Chinese medicine according to the present invention.

Fig. 3 is an example diagram of sample picture interception positions in the facial gloss analysis method based on traditional Chinese medicine inspection of the present invention.

Detailed ways

The present invention will be further elaborated below in conjunction with the accompanying drawings and specific embodiments. These examples should be understood as only for illustrating the present invention but not for limiting the protection scope of the present invention. After reading the contents of the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

In the following embodiments of the present invention, facial image samples are divided into three categories: "glossy", "less glossy" and "dull", with 30 training samples for each category.

As shown in Figure 1, in the facial gloss analysis method based on traditional Chinese medicine inspection provided by the preferred embodiment of the present invention, the whole facial gloss analysis process includes the following steps:

(1) Collect the required facial images under a uniform lighting environment.

(2) All facial images crop the image at the cheek and face as the sample image, and adjust the sample resolution to 100×100 (as shown in Figure 3).

(3) Classify the training samples into glossy categories, and divide them into three types of pictures: "glossy", "less glossy" and "dull", and select 30 pictures for each category as training samples.

(4) Adjust all pictures to HSV color mode.

(5) Adopt the improved two-dimensional principal component analysis method to carry out feature extraction to the test sample and the training sample of face image, the method is as follows:

①Each sample is represented by a 100×300 matrix I: Assume that I _{a, b} represent the elements at row a and column b of matrix I, then I _{i, 3j-2} , I _{i, 3j-1} , I _{i, 3j} respectively stores the H, S, and V values (i=1, 2, 3,..., 100, j=1, 2, 3,..., 100) of the pixels at the i-th row and j-column of the image;

为第i类训练样本的样本均值，I_i为训练样本总体均值；② Calculate an improved method based on the training samples to calculate the covariance matrix of the two-dimensional analysis: in

Is the sample mean value of the i-th type of training sample, and I _i is the overall mean value of the training sample;

③ Find the eigenvalue decomposition of the covariance matrix Cov, and arrange the obtained eigenvalues in descending order: λ ₁ , λ ₂ K λ ₃₀₀ , and the eigenvectors are sorted in the order of the corresponding eigenvalues, so the eigenvector matrix is obtained: V=[v ₁ v ₂ K v ₃₀₀ ] ^T ;

④ Select the first 10 rows of V to form a projection matrix: V _project = [v ₁ v ₂ K v ₁₀ ] ^T ;

⑤ According to the projection matrix V _project , perform feature extraction on the sample to obtain the sample feature F=V _project ^IT , so the feature set of the training sample is: F _train ={F ₁ , F ₂ K F ₃₀ }, and the test sample feature is F _test .

The above calculation of the cosine distance between the test sample feature and each training sample feature is to find the absolute value of the cosine value of the angle between F _test and all samples in F _train , and subtract the absolute value from 1.

The above-mentioned nearest neighbor method performs quantitative and qualitative analysis on the facial gloss of the test sample, that is, selects the category of the training sample feature with the smallest cosine distance from the F _test as the category of the test sample.

As shown in Figure 2, the cosine distance between a certain test sample and a certain training sample feature is 0.933, and this distance is the minimum value of all training samples. The category of this training sample is "less glossy", then this test The sample also belongs to the "less glossy" category; the category of the training sample is originally "glossy", then this test sample also belongs to the "glossy" category; the category of the training sample is "dull", then this test sample also belongs to "Matte" category.

The test samples belong to "shiny", "less shiny" or "dull", which can assist in the diagnosis of traditional Chinese medicine.

Claims (3)

1. a facial gloss analysis method based on traditional Chinese medicine inspection, it is characterized in that, the method comprises the steps successively: under the steady illumination environment, the digital camera controlled by computer automatically collects facial images, and the images are automatically stored in the computer, and the The cheek area is segmented from the facial image, and the pictures in HSV color mode are used as test samples and training samples. The improved two-dimensional principal component analysis method is used to extract the features of the test samples and training samples of the facial image, and the test sample features and each The cosine distance between the features of each training sample, and use the nearest neighbor method to analyze the facial gloss of the test sample, where: The aforementioned improved two-dimensional principal component analysis method is used to extract the features of the test samples and training samples of the face image, the method is as follows: ① Use matrix I to represent samples: if I _{a, b} represent the elements at row a and column b of matrix I, then use matrix I with m rows and 3n columns to represent a picture in HSV color mode with a resolution of n×m, I _{i, 3j-2} , I _{i, 3j-1} , I _{i, 3j} respectively store the H, S, and V values (i=1, 2, 3, ..., m, j=1,2,3,...,n); ② Calculate an improved method based on the training samples to calculate the covariance matrix of the two-dimensional analysis:

Where k represents the number of categories of samples, n _i is the number of training samples of the i-th category, N is the total number of training samples,

Is the sample mean value of the i-th type of training sample, and I _i is the overall mean value of the training sample; ③ Find the eigenvalue decomposition of the covariance matrix Cov, and arrange the obtained eigenvalues in descending order: λ ₁ , λ ₂ K λ _3n , and the eigenvectors are sorted in the order of the corresponding eigenvalues, so the eigenvector matrix is obtained: V=[v ₁ v ₂ K v _3n ] ^T ; ④ select the first d rows of V to form a projection matrix: V _project = [v ₁ v ₂ K v _d ] ^T ; ⑤ Perform feature extraction on the sample according to the projection matrix V _project to obtain the sample feature F = V _project ^IT , so the feature set of the training sample is: F _train = {F ₁ , F ₂ K F _N }, and the test sample feature is F _test . 2. the facial gloss analysis method based on traditional Chinese medicine inspection according to claim 1, is characterized in that, described calculation test sample feature and the cosine distance between each training sample feature, promptly seek in F _test and F _train The absolute value of the cosine of the included angle of all samples, and subtract the absolute value from 1. 3. according to claim 1 and 2 described based on the facial gloss analysis method of traditional Chinese medicine inspection, it is characterized in that, the method of described nearest neighbor carries out facial gloss analysis to test sample, selects the minimum training sample with F _test cosine distance The category of the feature is the category of the test sample.

Images