CN109919929B - Tongue crack feature extraction method based on wavelet transformation - Google Patents

Abstract

The invention discloses a tongue crack feature extraction method based on wavelet transformation, which comprises the following steps: applying median filtering to the original tongue image to smooth noise interference factors in the original tongue image; segmenting the smoothed tongue image by using a Canny edge detection operator; wavelet decomposition is carried out on the segmented tongue image to obtain a high-frequency tongue image component diagram and a low-frequency tongue image component diagram; fusing the high-low frequency component graph by using a wavelet fusion technology; and reconstructing clear tongue crack characteristics through wavelet inverse transformation. Decomposing and reconstructing a wavelet function, and applying the wavelet function to processing of tongue crack images, including tongue image denoising; by adding the wavelet transform processing technology, the feature extraction and matching of the tongue crack image can be more accurate.

Description

A tongue crack feature extraction method based on wavelet transform

Technical Field

The invention belongs to the technical field of traditional Chinese medicine, and in particular relates to a tongue crack feature extraction method based on wavelet transform.

Background Art

Traditional Chinese medicine believes that changes in tongue patterns can reflect the pathological conditions of the patient's internal organs. By examining the patient's tongue, doctors can promptly discover abnormal functions of certain parts of the patient, which plays a good auxiliary role in the diagnosis of some diseases. In traditional Chinese medicine, doctors use the "four examinations" to examine patients in order to determine the physiological and pathological conditions of the patient's internal organs' qi and blood, yin and yang. The "four examinations" in traditional Chinese medicine diagnosis are divided into four methods: observation, auscultation, questioning, and palpation. Among them, tongue diagnosis is an important part of traditional Chinese medicine diagnosis and plays an important role in clinical disease diagnosis. Tongue diagnosis is a diagnostic method that discovers pathological and physiological changes in human internal organs by observing the physiological and pathological morphology of the tongue. Doctors can make corresponding diagnoses and evaluations of patients' conditions by observing tongue patterns, which has important value in traditional Chinese medicine. After long-term clinical practice, the results show that tongue diagnosis can accurately identify the site of the lesion and the severity of the disease. Therefore, research on the medical theory of tongue diagnosis is of great significance and can provide an important basis for the diagnosis and treatment of clinical diseases.

Cracks and grooves of various shapes appear on the tongue surface, with varying depths. They can be seen on the entire tongue, the front of the tongue, the tip of the tongue, the sides of the tongue, etc. It mainly indicates deficiency of Yin and blood, and spleen deficiency and dampness. Cracks are a disease that occurs on the mucosa of the back of the tongue. Cracked tongue is distributed on the entire surface of the tongue. It is a pathological tongue, mainly in the front half of the tongue or the sides of the tip of the tongue. The research significance of tongue image has a profound impact on human health and other issues. Cracks are a disease that occurs on the mucosa of the back of the tongue. Cracked tongue is distributed on the entire surface of the tongue. It is a pathological tongue with a low rate of medical treatment, mainly in the front half of the tongue or the sides of the tip of the tongue. Its clinical feature is the appearance of regular or irregular cracks on the mucosa of the back of the tongue. In general, there are four reasons for the occurrence of cracked tongue, namely Yin deficiency theory, blood deficiency theory, Qi deficiency theory and Yang deficiency theory. Common syndromes include liver-kidney yin deficiency, stomach yin deficiency, liver blood deficiency, qi and yin deficiency, liver depression and spleen deficiency, hyperactivity of heart fire, spleen and stomach qi deficiency, and spleen and kidney yang deficiency.

Tongue crack images have the defects of extremely low reflectivity of light at the crack, the grayscale of the crack is much lower than the background grayscale, and the grayscale changes strongly at the crack. At present, the research methods on tongue cracks can be roughly divided into two categories. Among them, one method mainly performs threshold segmentation of tongue cracks based on the grayscale and color information of the tongue image, but does not consider the characteristics of the tongue crack, so it is difficult to accurately and completely segment the tongue crack. The other method is mainly based on line detection methods to segment tongue cracks, which can also be divided into three subcategories, namely contour-based line detection methods, centerline-based line detection methods, and region-based line detection methods. Although the line detection methods take into account the texture features of tongue cracks, they all have certain deficiencies. For example, the contour-based line detection method uses the first-order derivative, which is sensitive to noise and often fails to obtain a closed contour in practical applications. The centerline-based line detection method generally uses the second-order derivative, which is also sensitive to noise and often has large errors in the extraction of the centerline position. The region-based line detection method often segments the coarse texture and pseudo-cracks on the tongue coating, and it is necessary to manually remove redundant texture to segment the cracks.

Summary of the invention

In view of this, the present invention provides a tongue crack feature extraction method based on wavelet transform.

In order to solve the above technical problems, the present invention discloses a tongue crack feature extraction method based on wavelet transform, comprising the following steps:

S1, applying median filtering to the original tongue image to smooth out the noise interference factors in the original tongue image;

S2, segmenting the smoothed tongue image using the Canny edge detection operator;

S3, using wavelet decomposition to obtain high-frequency and low-frequency tongue image component images for the segmented tongue image;

S4, fusion of high and low frequency component images using wavelet fusion technology;

S5. Reconstruct clear tongue crack features through inverse wavelet transform.

Optionally, the median filter is applied to the original tongue image in step S1 as follows:

Use a two-dimensional sliding template to sort the pixels in the board according to the size of the pixel value, and generate a two-dimensional data sequence that is monotonically rising or falling;

The output of the two-dimensional median filter is

g(x, y)=med{f(x-k, y-l), (k, l∈W)} (1)

Where f(x, y), g(x, y) are the original image and the processed image respectively, k and l are the domain pixels of pixels x and y respectively; W is a two-dimensional template, which is a 3×3 area.

Optionally, the step S2 uses the Canny edge detection operator to segment the smoothed tongue image as follows:

S2.1, use Gaussian filter to smooth the image and filter out noise;

S2.2, calculate the gradient strength and direction of each pixel in the image;

S2.3, apply non-maximum suppression to eliminate spurious responses caused by edge detection;

S2.4. Apply dual threshold detection to determine real and potential edges.

Optionally, in step S2.1, a Gaussian filter is used to smooth the image and filter out noise. Specifically, in order to smooth the image, a Gaussian filter is used to convolve the image, and the generation equation of the Gaussian filter kernel with a size of (2k+1)x(2k+1) is given by the following formula:

Among them, σ is the variance, k is the dimension of the kernel matrix, and i and j are the pixel values of the image.

Optionally, the gradient strength and direction of each pixel in the image are calculated in step S2.2 as follows:

The edges in an image can point in various directions, so the Canny algorithm uses the Sobel operator to detect horizontal, vertical, and diagonal edges in an image; the edge detection operator returns the first-order derivative values in the horizontal G _x and vertical G _y directions, from which the gradient G and direction theta of the pixel point can be determined;

The Sobel convolution factor is:

The operator contains two sets of 3×3 matrices, one in the horizontal direction and the other in the vertical direction. Planar convolution is performed with the image to obtain the approximate brightness difference in the horizontal direction and the vertical direction respectively. I represents the original image, G _x and G _y represent the image grayscale values of the horizontal and vertical edge detection respectively. The formula is as follows:

The horizontal and vertical grayscale values of each pixel in the image are combined by the following formula to calculate the grayscale of the point:

The gradient direction is calculated using the following formula:

θ = arc tan (G _y /G _x ) (6)

Among them, G is the gradient strength, θ is the gradient direction, theta is the gradient direction, and arctan is the inverse tangent function;

The Sobel operator detects edges based on the phenomenon that the weighted difference in grayscale between the upper and lower pixels and the left and right pixels reaches an extreme value at the edge.

Optionally, the non-maximum suppression is applied in step S2.3 to eliminate the spurious response caused by edge detection as follows:

In the process of non-maximum suppression, a 3×3 moving window is used to process the image, and the central pixel gradient value is compared with the gradient values of other pixels in the domain. If the central pixel value is not the maximum value of the domain pixel, the pixel point is assigned a value of 0, otherwise the pixel point is regarded as the edge of the image; the steps for non-maximum suppression of each pixel in the gradient image are:

a) approximate its gradient direction to one of the following values (0, 45, 90, 135, 180, 225, 270, 315), i.e. up, down, left, right and 45° directions;

b) Compare the gradient strength of the pixel point with the pixel points in the positive and negative directions of its gradient direction;

c) If the pixel has the largest gradient strength, it is retained; otherwise, it is suppressed and deleted, i.e., set to 0;

Linear interpolation is used between two adjacent pixels that cross the gradient direction to obtain the pixel gradient to be compared; its specific mathematical expression is shown in formula (7):

Among them, i and j represent the pixels of the image, and maxgrade represents the maximum pixel gradient of the image.

Optionally, in step S3, the segmented tongue image is decomposed by wavelet to obtain high-frequency and low-frequency tongue image component images, specifically:

Using Daubechies-4 wavelet, the fingerprint image is decomposed by i-layer wavelet;

An image signal is a square image, divided into two identical left and right areas, where the left area is L and the right area is H; L is low frequency and H is high frequency;

The image signal is subjected to i-time wavelet decomposition to obtain a set of wavelet coefficients, whose size and shape are the same as the original image; among them, the upper left area is decomposed into 4 areas, among which the upper left area is LLi, the lower left area is LHi; the upper right area is HLi, and the lower right area is HHi; the remaining areas are the same as the decomposition of one layer.

Optionally, the step S4 of fusing the high-frequency and low-frequency component images using wavelet fusion technology is specifically as follows:

In low frequency, local variance is used as the basis; assuming that c(X) represents the coefficient matrix of the wavelet low-frequency component of the tongue crack image X; p(m, n) represents the spatial position of the wavelet coefficient; then c(X, p) represents the value of the element with the subscript (m, n) of the low-frequency component coefficient matrix; first, with p as the center, the weighted variance in the area Q represents the regional variance significance; u(X, p) represents the mean of the low-frequency coefficient matrix of the tongue crack image X, and point p is the center of the Q area; if G(X, p) represents the regional variance significance of the low-frequency coefficient matrix in the tongue crack image X, with point p as the center of the Q area, then:

G(X, p)＝∑ _P∈Q ω(p)|c(X, p)-u(X, p)| ² (8)

ω(p) represents the weight, and its value increases as it approaches point p; the regional variance of the low-frequency coefficient matrices of images A and B are expressed as G(A, p) and G(B, p); in addition, the regional variance matching degree of the low-frequency coefficient matrices of images A and B is defined by M ₂ (p) at point p:

The value of M ₂ (p) varies between 0 and 1. The smaller the value, the lower the matching of the low-frequency coefficient matrices of the two images.

Let T ₂ be the threshold of matching degree, which is usually 0.5-1;

When M ₂ (p)＜T ₂ , the fusion strategy is selected as follows:

When M ₂ (p) ≥ T ₂ , the average fusion strategy is as follows:

in,

In the high-frequency part of the wavelet transform, the maximum value of the absolute value of the wavelet coefficient is selected to make up for the real part of the information between the high and low frequencies; since the noise and defects of the crack target are high-frequency information, the median filter is used to filter the high-frequency coefficients of the fused tongue crack image to remove the noise and defects of the tongue crack image:

d(X, p) represents the coefficient matrix of the wavelet high-frequency component at point P.

Compared with the prior art, the present invention can achieve the following technical effects:

1) The tongue crack image has the following characteristics: the reflectivity of light at the crack is extremely low, the crack grayscale is much lower than the background grayscale, and the grayscale at the crack changes strongly. Therefore, the present invention selects median filtering to first smooth the image surface, reduce image noise while retaining the crack edge information, and then uses frequency domain filtering to strengthen the crack area. The method is divided into five parts. The median filter is applied to the original tongue image in order to smooth the noise and other interference factors in the original tongue image; the Canny edge detection operator is used to segment the smoothed tongue image; the segmented tongue image is decomposed by wavelet to obtain high-frequency and low-frequency tongue image component images; the high-frequency and low-frequency component images are fused by wavelet fusion technology, and finally a clear tongue crack feature is reconstructed by inverse wavelet transform. The present invention aims to improve the recognition rate of tongue crack detection. The wavelet function is decomposed and reconstructed, and applied to the processing of tongue crack images, including tongue image denoising. By adding wavelet transform processing technology, the feature extraction and matching of tongue crack images can be made more accurate.

2) Compared with the prior art, the greater the absolute value of the wavelet coefficients in the technical solution proposed by the present invention, the greater the grayscale change of the original image and the higher the importance. Irregular structures such as cracks will be converted into high-amplitude coefficients in the high-frequency sub-band after wavelet transformation. The number of coefficients in the high-frequency sub-band with amplitudes greater than a certain threshold is calculated as a feature of the crack image.

3) The present invention uses 32 tongue crack images provided by Beijing Dongzhimen Hospital, of which 2 tongue cracks failed to be identified, with an accuracy rate of up to 93.75%. The wavelet decomposition and reconstruction technology solves some defects in the traditional segmentation method, such as incomplete segmentation and over-segmentation. The present invention can correctly detect tongue crack images, has a certain anti-interference ability, and solves the problems of information loss and over-segmentation.

Of course, any product implementing the present invention does not necessarily need to achieve all of the above-mentioned technical effects at the same time.

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

FIG1 is a diagram of the tongue crack identification process of the present invention;

FIG2 is a wavelet decomposition of the present invention; wherein a represents the low-frequency and high-frequency components of an image; b represents the first-level wavelet decomposition of an image; and c represents the second-level wavelet decomposition of an image.

DETAILED DESCRIPTION

The following will describe the implementation methods of the present invention in detail with reference to the embodiments, so that the implementation process of how the present invention applies technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

The invention discloses a tongue crack feature extraction method based on wavelet transform, comprising the following steps:

S1, applying median filtering to the original tongue image to smooth out interference factors such as noise in the original tongue image;

The median filtering method is a nonlinear image enhancement technology and a filtering method based on sorting statistics theory. Median filtering can significantly suppress noise. It sets the grayscale value of each pixel in the image to the median of the grayscale values of all pixels in a certain neighborhood window of the point. The basic principle of median filtering is to replace the value of a point in a digital image or digital sequence with the median of the values of each point in a neighborhood of the point, so that the surrounding pixel values are close to the true value, thereby eliminating isolated noise points. The method is to use a two-dimensional sliding template to sort the pixels in the plate according to the size of the pixel value, and generate a monotonically rising (or falling) two-dimensional data sequence.

The output of the two-dimensional median filter is

g(x, y)=med{f(x-k, y-l), (k, l∈W)} (1)

Where f(x, y), g(x, y) are the original image and the processed image respectively, k and 1 are the domain pixels of pixels x and y respectively. W is a two-dimensional template, usually a 3×3, 5×5 area, and can also be different shapes, such as linear, circular, cross, annular, etc. The present invention uses a 3×3 area.

S2, segmenting the smoothed tongue image using the Canny edge detection operator;

In the Canny edge segmentation, the edge is the boundary between the object and the background, and can also represent the boundary between overlapping objects. Compared with other edge detection algorithms, Canny detection can effectively suppress image noise and more accurately determine the location of the image edge. Canny edge detection can be divided into four steps:

S2.1. Use a Gaussian filter to smooth the image and remove noise.

In order to minimize the effect of noise on edge detection results, it is necessary to filter out the noise to prevent false detection caused by noise. To smooth the image, a Gaussian filter is convolved with the image. This step will smooth the image to reduce the obvious effect of noise on the edge detector. The generation equation of the Gaussian filter kernel of size (2k+1)x(2k+1) is given by the following equation:

Among them, σ is the variance, k is the dimension of the kernel matrix, and i and j are the pixel values of the image;

S2.2. Calculate the gradient strength and direction of each pixel in the image.

The edges in an image can point to various directions, so the Canny algorithm uses four operators to detect horizontal, vertical and diagonal edges in an image. Edge detection operators (such as Roberts, Prewitt, Sobel, etc.) return the first-order derivative values in the horizontal G _x and vertical G _y directions, from which the gradient G and direction theta of the pixel point can be determined. The present invention chooses to use the Sobel operator because it is the most common operator.

The Sobel operator is mainly used for edge detection. Technically, it is a discrete difference operator. The Sobel convolution factor is:

The operator contains two sets of 3×3 matrices, one in the horizontal direction and the other in the vertical direction. By performing a plane convolution with the image, the approximate brightness difference in the horizontal direction and the vertical direction can be obtained respectively. Let I represent the original image, G _x and G _y represent the image grayscale values of the horizontal and vertical edge detection respectively, and the formula is as follows:

The horizontal and vertical grayscale values of each pixel in the image can be combined by the following formula to calculate the grayscale of the point:

The gradient direction can be calculated using the following formula:

θ = arc tan (G _y /G _x ) (6)

Where G is the gradient strength, θ is the gradient direction, theta is the gradient direction, and arctan is the inverse tangent function.

The Sobel operator detects edges based on the grayscale weighted difference of the upper and lower pixels and the left and right pixels, which reaches an extreme value at the edge. It has a smoothing effect on noise and provides more accurate edge direction information.

S2.3. Apply non-maximum suppression to eliminate spurious responses caused by edge detection:

After the gradient calculation of the image, the edge extracted based on the gradient value alone is still fuzzy. Non-maximum suppression can help suppress all gradient values outside the local maximum to 0. In the non-maximum suppression process, a 3×3 moving window is used to process the image, and the central pixel gradient value is compared with the gradient values of other pixels in the domain. If the central pixel value is not the maximum value of the domain pixel, the pixel point is assigned a value of 0, otherwise the pixel point is regarded as the edge of the image. The steps for non-maximum suppression of each pixel in the gradient image are:

a) approximate its gradient direction to one of the following values (0, 45, 90, 135, 180, 225, 270, 315) (i.e. up, down, left, right and 45° directions);

b) Compare the gradient strength of the pixel point with the pixel points in the positive and negative directions of its gradient direction;

c) If the pixel has the largest gradient intensity, it is retained; otherwise, it is suppressed (deleted, i.e., set to 0);

Usually, for more accurate calculation, linear interpolation is used between two adjacent pixels across the gradient direction to obtain the pixel gradient to be compared. Its specific mathematical expression is as follows:

Among them, i and j represent the pixels of the image, and maxgrade represents the maximum pixel gradient of the image.

Non-maximum suppression not only effectively preserves the gradient of the image edge, but also achieves the purpose of image refinement.

S2.4. Apply double-threshold detection to determine real and potential edges.

After non-maximum suppression, the image still has many noise points. The Canny algorithm uses a double threshold technology, that is, setting an upper threshold and a lower threshold. If the pixel in the image is greater than the upper threshold, it is considered to be a boundary, and if it is less than the threshold, it is considered not to be a boundary.

For the canny edge segmentation image, the high-frequency image contains the virtual and real edge information of the target defect, and also contains some noise in the tongue crack image. The low-frequency image contains the contour information of the target defect.

S3, using wavelet decomposition to obtain high-frequency and low-frequency tongue image component images for the segmented tongue image;

The wavelet transform is a method of decomposing an image into different frequency domains. Different fusion rules are then applied in different frequency domains, and finally, the image is reconstructed using an inverse wavelet transform. Wavelet transform can decompose the tongue crack image into a combination of an average image and a detail image, which represent different structures of the image. Therefore, it is easy to extract the structural information and detail information of the original image. Wavelet transform is based on Fourier analysis. It uses the characteristics of multi-resolution analysis of wavelet transform to characterize the local characteristics of the signal in both the time domain and the frequency domain. Based on the characteristics of the constant window size and variable shape, a higher frequency resolution is used in the low-frequency part of the image signal, and a method with a higher time resolution and a lower frequency resolution is used in the high-frequency part. It is used in the preprocessing stage of tongue crack recognition and can process tongue cracks with irregular signals.

For the two-dimensional wavelet transform, it can be regarded as two consecutive one-dimensional wavelet transforms. By processing the image through the two-dimensional wavelet transform, it can be decomposed into a series of low-frequency sub-images. The result depends on the type of wavelet basis, that is, the type of filter. The present invention adopts the widely used Daubechies-4 wavelet to perform a two-layer wavelet decomposition on the fingerprint image.

The present invention performs wavelet decomposition on an image signal to obtain a set of wavelet coefficients, whose size and shape are the same as the original image. A 300×300 image is decomposed by two layers of wavelets to obtain 7 blocks of wavelet decomposition results as shown in Figure 2, with a total of 90,000 coefficients.

The image signal is decomposed by wavelet i times to obtain a set of wavelet coefficients, whose size and shape are the same as the original image; the upper left area is decomposed into 4 areas, among which the upper left area is LLi and the lower left area is LHi; the upper right area is HLi and the lower right area is HHi; the remaining areas are the same as the first layer of decomposition; L is low frequency and H is high frequency; subscripts 1 and 2 indicate one or two decompositions. At each decomposition layer, the image is decomposed into four bands: LL, LH, HH, and HL. Only the low-frequency components of the next layer are decomposed. Each of the four sub-images is generated by the inner product of the original image and a wavelet basis function. Then it is sampled twice in the X and Y directions. The inverse transform, that is, image reconstruction, is achieved by increasing the sampling frequency and convolution of the image. Since the processed data may exceed 255 or appear negative, it needs to be normalized to 0-255 to display the image.

S4. Use wavelet fusion technology to fuse the high and low frequency component images:

In the wavelet fusion, the low-frequency wavelet coefficients contain the contour information of the image, so it is necessary to select appropriate wavelet coefficients for the fusion operation. The present invention uses local variance as a basis in low frequency. Assume that c(X) represents the coefficient matrix of the wavelet low-frequency component of the tongue crack image X. p(m, n) represents the spatial position of the wavelet coefficient. Then c(X, p) represents the value of the element with the subscript (m, n) of the low-frequency component coefficient matrix. First, with p as the center, the weighted variance in the area Q is used to represent the regional variance significance. u(X, p) represents the mean of the low-frequency coefficient matrix of the tongue crack image X, and point p is the center of the Q area. If G(X, p) represents the regional variance significance of the low-frequency coefficient matrix in the tongue crack image X, with point p as the center of the Q area, then:

G(X, p)＝∑ _P∈Q ω(p)|c(X, p)-u(X, p)| ² (8)

ω(p) represents a weight, and its value increases as it approaches point p. The regional variance of the low-frequency coefficient matrices of images A and B are represented by G(A, p) and G(B, p). In addition, the regional variance matching degree of the low-frequency coefficient matrices of images A and B is defined by M ₂ (p) at point p:

The value of M ₂ (p) varies between 0 and 1. The smaller the value, the lower the matching degree of the low-frequency coefficient matrices of the two images.

Let T ₂ be the threshold of matching degree, which is usually (0.5-1).

When M ₂ (p)＜T ₂ , the (optimal) fusion strategy is selected as follows:

When M ₂ (p) ≥ T ₂ , the average fusion strategy is as follows:

in

The above strategy is based on the correlation between pixels in the field, which can effectively preserve the details and edges based on canny detection.

The larger the absolute value of the wavelet coefficient, the greater the grayscale change of the original image and the higher its importance. Irregular structures such as cracks will be converted into high-amplitude coefficients in the high-frequency sub-band after wavelet transformation. The number of coefficients in the high-frequency sub-band with amplitudes greater than a certain threshold is calculated as a feature of the crack image.

After accurate canny edge segmentation, the image has a strong contrast. The present invention selects the maximum value of the absolute value of the wavelet coefficient in the high-frequency part of the wavelet transform, which can make up for the exact part of the information between high and low frequencies. Since the noise and defects of the crack target are high-frequency information, the median filter is used to filter the high-frequency coefficients of the fused tongue crack image to remove the noise and defects of the tongue crack image.

d(X, p) represents the coefficient matrix of the wavelet high-frequency component at point P.

S5. Reconstruct clear tongue crack features through inverse wavelet transform.

The inverse wavelet transform acts on the processed high-frequency and low-frequency components to reconstruct a clear tongue crack image.

The present invention uses 32 tongue crack images provided by Beijing Dongzhimen Hospital, of which 2 tongue cracks failed to be identified, with an accuracy rate of up to 93.75%. The wavelet decomposition and reconstruction technical solution solves some defects in the traditional segmentation method, such as incomplete segmentation and over-segmentation. The present invention can correctly detect tongue crack images, has a certain anti-interference ability, and solves the problems of information loss and over-segmentation.

The above description shows and describes several preferred embodiments of the invention, but as mentioned above, it should be understood that the invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the invention concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall be within the scope of protection of the claims attached to the invention without departing from the spirit and scope of the invention.

Claims (5)

1. A tongue crack feature extraction method based on wavelet transform, characterized in that it comprises the following steps: S1. Apply median filtering to the original tongue image to smooth out the noise interference factors in the original tongue image. The specific steps of applying median filtering to the original tongue image are as follows: Use a two-dimensional sliding template to sort the pixels in the board according to the size of the pixel value, and generate a two-dimensional data sequence that is monotonically rising or falling; The output of the two-dimensional median filter is g(x,y)=med{f(x-k,y-l),(k,l∈W)} (1) Where f(x,y), g(x,y) are the original image and the processed image respectively, k and l are the domain pixels of pixels x and y respectively; W is a two-dimensional template, which is a 3×3 area; S2. Segment the smoothed tongue image using the Canny edge detection operator. Segment the smoothed tongue image using the Canny edge detection operator is specifically as follows: S2.1, use Gaussian filter to smooth the image and filter out noise; S2.2, calculate the gradient strength and direction of each pixel in the image; S2.3, apply non-maximum suppression to eliminate spurious responses caused by edge detection; S2.4, apply double threshold detection to determine real and potential edges; S3, using wavelet decomposition to obtain high-frequency and low-frequency tongue image component images for the segmented tongue image; S4, fusion of high and low frequency component images using wavelet fusion technology; S5. Reconstruct clear tongue crack features through inverse wavelet transform. 2. The method according to claim 1 is characterized in that the step S2.1 uses a Gaussian filter to smooth the image and filter out noise. Specifically, in order to smooth the image, a Gaussian filter is used to convolve with the image, and the generation equation of the Gaussian filter kernel with a size of (2k+1)x(2k+1) is given by the following formula:

Among them, σ is the variance, k is the dimension of the kernel matrix, and i and j are the pixel values of the image. 3. The method according to claim 2, characterized in that the step S2.2 of calculating the gradient strength and direction of each pixel in the image is specifically: The edges in an image can point in various directions, so the Canny algorithm uses the Sobel operator to detect horizontal, vertical, and diagonal edges in an image; the edge detection operator returns the first-order derivative values in the horizontal G _x and vertical G _y directions, from which the gradient G and direction theta of the pixel point can be determined; The Sobel convolution factor is: -1 0 +1 -2 0 +2 -1 0 +1 +1 +2 +1 0 0 0 -1 -2 -1 The operator contains two sets of 3×3 matrices, one in the horizontal direction and the other in the vertical direction. Planar convolution is performed with the image to obtain the approximate brightness difference in the horizontal direction and the vertical direction respectively. I represents the original image, G _x and G _y represent the image grayscale values of the horizontal and vertical edge detection respectively. The formula is as follows:

The horizontal and vertical grayscale values of each pixel in the image are combined by the following formula to calculate the grayscale of the point:

The gradient direction is calculated using the following formula: θ＝arctan(G _y /G _x ) (6) Among them, G is the gradient strength, θ is the gradient direction, theta is the gradient direction, and arctan is the inverse tangent function; The Sobel operator detects edges based on the phenomenon that the weighted grayscale difference of the upper and lower pixels and the left and right pixels reaches an extreme value at the edge. 4. The method according to claim 1, characterized in that the non-maximum suppression is applied in step S2.3 to eliminate the spurious response caused by edge detection, specifically: In the process of non-maximum suppression, a 3×3 moving window is used to process the image, and the central pixel gradient value is compared with the gradient values of other pixels in the domain. If the central pixel value is not the maximum value of the domain pixel, the pixel point is assigned a value of 0, otherwise the pixel point is regarded as the edge of the image; the steps for non-maximum suppression of each pixel in the gradient image are: a) Its gradient direction is approximated to one of the following values (0, 45, 90, 135, 180, 225, 270, 315), i.e. up, down, left, right and 45° directions; b) Compare the gradient strength of the pixel point with the pixel points in the positive and negative directions of its gradient direction; c) If the pixel has the largest gradient strength, it is retained; otherwise, it is suppressed and deleted, i.e., set to 0; Linear interpolation is used between two adjacent pixels that cross the gradient direction to obtain the pixel gradient to be compared; its specific mathematical expression is shown in formula (7):

Among them, i and j represent the pixels of the image, and maxgrade represents the maximum pixel gradient of the image. 5. The method according to claim 1, characterized in that in step S3, the segmented tongue image is decomposed by wavelet to obtain high-frequency and low-frequency tongue image component images, specifically: Using Daubechies-4 wavelet, the fingerprint image is decomposed by i-layer wavelet; An image signal is a square image, divided into two identical left and right areas, where the left area is L and the right area is H; L is low frequency and H is high frequency; The image signal is subjected to i-time wavelet decomposition to obtain a set of wavelet coefficients, whose size and shape are the same as the original image; among them, the upper left area is decomposed into 4 areas, among which the upper left area is LLi, the lower left area is LHi; the upper right area is HLi, and the lower right area is HHi; the remaining areas are the same as the decomposition of one layer.

Images