CN112037242B

CN112037242B - Automatic identification method for epidermal layer in skin optical coherence tomography image

Info

Publication number: CN112037242B
Application number: CN202010895236.1A
Authority: CN
Inventors: 杨晨铭; 李中梁; 南楠; 张茜; 欧阳君怡; 刘腾; 步扬; 王向朝
Original assignee: Shanghai Institute of Optics and Fine Mechanics of CAS
Current assignee: Shanghai Institute of Optics and Fine Mechanics of CAS
Priority date: 2020-08-31
Filing date: 2020-08-31
Publication date: 2022-03-08
Anticipated expiration: 2040-08-31
Also published as: CN112037242A

Abstract

A method for automatic identification of epidermal layers in Optical Coherence Tomography (OCT) images of the skin. Firstly, searching the maximum axial gradient position of each column in a skin OCT two-dimensional image, and combining boundary continuity to obtain an air-epidermis boundary line; then, a certain number of pixels of each boundary pixel in the image in the depth direction are taken out to form a new image, the new image is subjected to sliding window processing to obtain an average signal of each row, and an approximate region of an epidermis-dermis boundary line is obtained according to a peak value; obtaining a continuous gradient boundary line through multi-scale topological transmission; the area where the epidermis layer is located is identified in the image by using a seed filling and morphological algorithm, and the boundary position of the epidermis layer area is extracted by a boundary tracking algorithm. The invention can automatically segment the mastoid structure bent at the epidermis-dermis boundary without manual identification, and is suitable for the skin OCT image with reduced signal intensity of the dermis caused by diseases such as dermatitis.

Description

Automatic identification method for epidermal layer in skin optical coherence tomography image

Technical Field

The invention relates to Optical Coherence Tomography (OCT for short), in particular to a method for automatically extracting information in a skin OCT image, and more particularly to a method for automatically identifying an epidermal layer in a skin OCT image.

Background

Optical Coherence Tomography (OCT) is a biomedical imaging technique based on low Coherence interferometry for detecting backscattered light from a sample, and can detect the internal microstructure of biological tissues non-invasively and with high resolution in vivo. This concept was first proposed by j.g. fujimoto and d.huang et al, the american academy of labor in the 20 th century, the 90 s, and structural images of isolated tissues such as retina and coronary artery were obtained. In recent years, the application fields of the medicine are gradually expanded, and the medicine comprises dermatology, cardiovascular, dentistry, neurology and the like.

Human skin is a complex system composed of different layers. Aging affects human skin and is increasingly important in medical, social and aesthetic problems. And the skin is very easy to be damaged when contacting with the outside world. All skin diseases reach more than 6000 species, the prevalence rate of the population is close to 100 percent, and great social burden is caused. Histology is the gold standard for skin morphology studies, but tissue biopsy is an invasive test that causes pain to the patient, may alter the original morphology, and does not allow for repeated studies in one area and follow-up of disease treatment. OCT provides a non-invasive method for in vivo imaging of surface skin tissue. Due to its relatively high resolution and high imaging depth, OCT fills the imaging gap between ultrasound and confocal microscopes and has been used to screen skin diseases and monitor interventions.

Studies have shown that with age, the epidermal layer of the skin becomes thinner and the epidermal-dermal boundary tends to flatten. The thickness of the epidermis layer is an index of skin aging and health, and the local shape of the epidermis-dermis boundary is an important factor for detecting skin health, aging, and photodamage, so identification of the epidermis is of great significance in medical fields such as dermatology, orthopedics, and pharmacology.

Inflammatory dermatoses such as contact dermatitis and psoriasis are common. Psoriasis, also known as psoriasis, is a chronic skin disease affecting 1-3% of the population. The main symptoms are development of skin plaques and increased epidermal thickness. The air-epidermal boundary of the optical coherence tomography image of psoriatic skin is more irregular and the entry signal is stronger than that of normal skin, the dermal-epidermal boundary is jagged, and the signal intensity in the dermis is smaller. Visual scoring by the naked eye is used to quantify and monitor the efficacy of treatment of disease in clinical and experimental studies, but lacks objectivity and reliability, such as PASI scores that combine the Area and Severity of Psoriasis (Psoriasis Area and Severity Index). Epidermal thickness based on image analysis has been suggested as a surrogate indicator of psoriasis severity. OCT image epidermal layer identification is an in-vivo, convenient and non-invasive method, and has great significance in the aspect of researching the treatment effect of diseases and reducing the suffering of patients.

The existing identification method of the epidermal layer in the skin optical coherence tomography image is mainly divided into two types, namely manual segmentation and algorithm segmentation depending on gray information, and the two types are respectively as follows:

1) the researcher manually identified the method. According to the method, a researcher manually marks out the epidermal layer area according to experience by means of measurement software configured by a commercial OCT system. (see prior art [1] gambicher T, Moussa G, Regeniter P, et al. differentiation of optical coherence Biology in vivo using cryostat Biology [ J ]. Physics in Medicine & Biology,2007,52(5):75-85.)

2) And (3) identifying the method according to the gray information algorithm. One way is to identify the epidermal layer from the gray scale curve peaks, identify the first peak in the axial direction as the skin surface, and the minimum between the first peak and the second peak as the epidermal-dermal boundary (see prior art [2] P Zakharov, M S Talar, I Kolm, et al. full-field optical coherence tomography for the radial estimation of epidermal thickness: the term of properties with the geometry type 1[ J ]. physical Measurement,2009,31(2): 193-205.). The other mode is that the epidermal layer is identified according to the gray gradient value, and firstly, a mean filter is used for removing speckle noise; detecting sharp changes in image intensity in the depth direction (from dark to light) by applying an optimized Sobel edge detection filter, recursively calculating confidence for each measurement for candidate edges with sizes exceeding a 30-pixel threshold based on surface smoothness and edge fitting, selecting the surface edge points of each column with the greatest confidence for interpolation to identify the entire air-skin boundary; transitions from low reflectance bands to higher reflectance bands are detected in the same manner to form a complete epidermal-dermal boundary detection (see prior art [3] r. maiti, l.c. gerhardt, z.s. lee, et al. in visual space of skin surface string and sub-surface layer development induced by natural tissue string [ j.j.mech. behav. biomed. mater,2016(62). 556-.

The epidermis layers can be identified by the methods, but the manual method requires a long time and has a large subjective effect on a measuring person; the method of the prior art 2 can only obtain the average thickness of the epidermis, and the method of judging the boundary according to the gray peak value of the image row defaults that each row only has one epidermis-dermis boundary point, and cannot identify the curved papillary structure on the epidermis-dermis boundary; the method of prior art 3 is considered to change the intensity of the image from the low reflection band to the higher reflection band to the epidermal-dermal boundary, but in the case where the intensity of the dermal layer decreases due to a disease such as psoriasis, the result of the determination of this method is greatly different from the result of the histological epidermal layer, as in the image of prior art 1.

Disclosure of Invention

The present invention is directed to overcoming the above-mentioned deficiencies in the prior art and providing a method for automatically identifying an epidermal layer in an optical coherence tomography image of skin.

The technical solution of the invention is as follows:

a method for automatically identifying an epidermal layer in a skin optical coherence tomography image is characterized by comprising the following steps:

firstly, an interference signal of skin tissue is collected by using an OCT system, a two-dimensional intensity image is obtained by image reconstruction, and a two-dimensional original input image I is obtained by carrying out logarithmic transformation on the two-dimensional intensity image. The first pixel defining the first row of the image is the origin, the horizontal right abscissa x is the lateral scan distance, and the vertical down ordinate z is the longitudinal imaging depth. The two-dimensional original input image has n columns, and n is a positive integer.

Secondly, axial derivation based on Gaussian kernel is carried out on the two-dimensional original input image I to obtain an axial gradient image I_zTaking an axial gradient map I_zThe vertical coordinates of the maximum pixels in each column form a one-dimensional array z _ surface containing n elements, the array elements are z _ surface _1, z _ surface _2, … …, z _ surface _ m, … …, and z _ surface _ n, wherein m is an element index and is a positive integer. Making difference between every two adjacent elements in the array z _ surface, and for the index M in the array z _ surface, which is greater than M in difference with the previous element, in I_zThe mth column takes a pixel point with the ordinate as z _ surface _ (M-1) as the center, the maximum gradient position in a certain number of pixel ranges in the depth direction is searched again, a new ordinate z _ surface _ M (new) is obtained, and the value of M can be set to be 1 to 5; and performing the same processing on indexes with the difference value of adjacent elements being larger than M in the z _ surface to obtain a new one-dimensional array z _ surface _ new. The values of n elements in the one-dimensional array z _ surface _ new are considered as the ordinate of n pixels on the air-skin boundary in the two-dimensional original input image I, and the indices of the n elements are considered as the abscissa of n pixels on the air-skin boundary.

Taking out a certain number of continuous pixels of each pixel on the air-epidermis boundary line in the vertical upward direction and the downward direction from the two-dimensional original input image I, storing the pixels in a two-dimensional array, and recording the image represented by the array as I_roiDefining the first pixel of the first row of the image as the origin and the horizontal right as the abscissa x₁Vertically downwards as ordinate z₁. First pair of images I_roiAnd performing Gaussian filtering, and performing sliding window processing on each row respectively to obtain an average signal of each row, and recording the average signal as a one-dimensional array avr. And performing one-dimensional linear filtering on the avr of each column to obtain an array. The maximum index after the first maximum index loc1 in the array is denoted as maxloc. The array avr is inverted in intensity, i.e. the intensity value is subtracted from 255, and the result is recordedIs array avr 2. Marking the index of the first maximum value of the array avr2 after the index loc1 as minloc; if this first maximum is the maximum of array avr2 that is midway between loc1 and maxloc, then the index minloc is shifted up by a certain number of pixels to get a new index minloc. And forming the minloc and maxloc of each column into one-dimensional arrays min _ loc and max _ loc respectively. The difference is made between the elements in the array max _ loc and the average value of all the elements in the array, and the value of the element with the difference value larger than M2 is changed into an image I_roiThe image width of (2). For with I_roiImage ordinate z₁The array min _ loc and max _ loc elements which are used as the reference are respectively found out, and the array min _ loc _ real and max _ loc _ real elements which are used as the reference are respectively found out. Setting the gray value of each row of pixels above the index min _ loc _ real and below the index max _ loc _ real of the two-dimensional original input image I as 0, and setting the gray values of other pixels as 1 to obtain the template image I_m。

Fourthly, using at least two edge detection operators with different scales to obtain gradient maps with different scales of the two-dimensional original input image I; taking a gradient map of a certain scale as a main image I_cUsing a topology transfer method to transfer the primary image I_cComparing with other scale gradient maps to connect the main image I_cThe edge line with discontinuous gradient is obtained to obtain a gradient map I with continuous edge line_t(ii) a Map of gradients I_tMultiplying by the template image I_mObtaining a gradient map I of the approximate location of the epidermal-dermal boundary_tm. Carrying out axial derivation based on Gaussian kernel on an original input image I to obtain an axial gradient image, and multiplying the axial gradient image by a template image I_mObtaining an axial gradient map I of the approximate location of the epidermal-dermal boundary_tzmThe sigma value of the gaussian kernel used may be different from that in step two. In the gradient map I_tmIn (1), will be plotted against the axial gradient I_tzmSetting the pixel with the same maximum value coordinate of each column as 1, and connecting the points by using a straight line; in the gradient map I_tmWherein the pixel of each row corresponding to the air-epidermis boundary z _ surface _ new is set as 1, and the points are connected by straight lines to finally obtain the image I_tms；

Fifthly, theImage I_tmsThe middle coordinate is set as a starting point for the pixel of the index min _ loc _ real, namely the seed filled by the seed, a first closed area is searched according to the four-connection rule, and morphological closed operation is carried out on the area to obtain the area where the epidermis layer is located; and then extracting edges by using a boundary tracking operator to obtain a boundary line of a surface layer region on the two-dimensional original input image.

The certain number of pixels and M2 are adjustable according to different OCT systems and different parts of skin images.

The sliding window takes the data of each row of the image as the center, the data of each window with a certain width at the left and the right of the row of the image is obtained, the average value of each row is averaged to form a one-dimensional array avr, the row number of the window is equal to the row number of the image, the row number is less than the row number of the image, if the data of each row at the left and the right is obtained for each row, if the distance between the row and the edge of the image is less than the row a, the row is the leftmost or the rightmost position of the window.

And fourthly, the edge detection operator is a canny edge operator based on Gaussian derivation.

And fourthly, setting different sigma values in the Gaussian derivation process according to different scales.

Establishing a new two-dimensional image I _ find, setting the gray of the pixel point with the gray value of 1 in the two compared binary images to be 2 at the same coordinate position of the I _ find, setting the gray of the pixel point with the gray value of 0 to be 0 at the same coordinate position of the I _ find, and setting the gray of the pixel point with the different gray value to be 1 at the same coordinate position of the I _ find; in an image I _ find, for a pixel point with a gray value of 1, in a set search range, searching for a pixel point with a gray value not being 0 according to an 8-connection rule for connection, namely, for a pixel point with a gray value of 1, in 8 directions meeting 8-connection in the set search range, a pixel point with a gray value of 2 can be found along the pixel point with a gray value not being 0 according to the 8-connection rule in two directions, and then on a main image, the gray value of the pixel point with the same coordinate as the pixel point with the gray value being 1 on the I _ find is set to be 1.

Compared with the prior art, the invention has the beneficial effects that:

1. compared with the prior art [1], the method does not need manual marking and drawing, improves the identification speed and reduces the influence of subjective judgment of an identifier.

2. In contrast to the prior art [2], the present invention allows multiple epidermal-dermal demarcation points per column, allowing identification of the papillary structures that are curved at the epidermal-dermal demarcation line.

3. In contrast to prior art [3], the present invention identifies the first clear definition of a sharp demarcation of reflectance intensity below the skin air-epidermal demarcation location as the epidermal-dermal demarcation based on the manual identification rule that best fits the biopsy results. The invention has better result for identifying the epidermis layer of a healthy skin image and a skin image with reduced dermis strength caused by dermatitis.

Drawings

FIG. 1 is a flow chart of the method for automatically identifying the epidermis layer in the skin optical coherence tomography image.

Fig. 2 is an original input image.

Fig. 3 is an axial gradient image of the original input image.

Fig. 4 is an image of the original image with the air-epidermis boundary line being cut.

FIG. 5 is I_roiAnd (4) an image.

FIG. 6 shows a template I_mAnd (4) an image.

Fig. 7 is a gradient image with continuous boundaries.

FIG. 8 is I_tmsAnd (4) an image.

Fig. 9 is an image of the area where the epidermis layer is located.

Fig. 10 is a diagram of an original input image with an epidermal layer divided.

Detailed Description

The present invention will be further described with reference to the following examples and drawings, but the scope of the present invention should not be limited by these examples.

Fig. 1 is a flowchart of the method for automatically identifying the epidermis layer in the skin optical coherence tomography image according to the present invention. The method comprises the following steps:

firstly, an optical coherence tomography system is used for imaging the part of the skin of the eczema patient, which does not involve the plaque, and after the processing such as logarithmic transformation, a two-dimensional original input image is obtained and recorded as an image I, as shown in fig. 2. The first pixel defining the first row of the image is the origin, the horizontal right abscissa x is the lateral scan distance, the vertical down ordinate z is the longitudinal imaging depth, and the two-dimensional image has a total of 752 columns.

Secondly, carrying out axial derivation on the original input image I, wherein the sigma of a Gaussian kernel of the original input image I is 4, and obtaining an axial gradient image I_zAs shown in fig. 3. Taking an axial gradient map I_zThe ordinate of the maximum pixel in each column forms a one-dimensional array z _ surface containing 752 elements, which are z _ surface _1, z _ surface _2, … …, z _ surface _ m, … …, and z _ surface _752, where m is the element index and is a positive integer. For the index m of the element in the array z _ surface that differs by more than 5 from the previous element, at I_zThe mth column (m) takes the pixel point with the ordinate as z _ surface _ (m-1) as the center, and the maximum gradient position in the range of 4 pixels above and below the depth direction is searched again to obtain a new ordinate z _ surface (new). And performing the same processing on elements with the difference value larger than 5 in the z _ surface and the previous element to obtain a new one-dimensional array z _ surface _ new. The 752 values of the elements in the one-dimensional array z _ surface _ new are considered to be the ordinate of 752 pixels on the air-skin boundary in the two-dimensional original input image I, and the indices of the elements are considered to be the abscissa of the pixels on the air-skin boundary. The air-skin boundary line is drawn in the original as shown in fig. 4.

Taking the air-epidermis boundary as the initial position of each row of the two-dimensional original input image I, continuously taking out 144 pixels along the vertical downward direction, continuously taking out 5 pixels along the vertical upward direction, storing the pixels in a two-dimensional array, and recording the image represented by the array as I_roiAs shown in fig. 5. The first pixel defining the first line of the image is the origin and the horizontal right is the abscissa x₁Vertically downwards as ordinate z₁. First pair of images I_roiGaussian filtering is carried out, and then the image is obtained by taking each line of data of the image as the centerThe data of 10 windows on the left and right of the row are averaged for each row to obtain 752 one-dimensional arrays avr. If the distance between the row and the image edge is less than 10 rows, the row is the leftmost position or the rightmost position of the window. And performing moving average filtering on the avr of each column to obtain an array, and recording the maximum value index behind the first maximum value index loc1 in the array as maxloc. The intensity values of the elements in the array avr are subtracted from 255, respectively, and the result is recorded as the array avr 2. The index of the first maximum value of the array avr2 after the index loc1 is recorded as minloc. If this first maximum is the maximum of array avr2 that is midway between loc1 and maxloc, then index minloc is shifted up by 3 pixels. And forming the minloc and maxloc of each column into one-dimensional arrays min _ loc and max _ loc respectively. The element in array max _ loc that differs by more than 55 from the average of all elements in the array is changed to 150. For with I_roiImage ordinate z₁And respectively finding corresponding array min _ loc _ real and max _ loc _ real elements which take the vertical coordinate z of the two-dimensional original input image I as the reference, namely respectively adding the air-skin boundary vertical coordinate z _ surface and subtracting 6 to obtain array min _ loc _ real and max _ loc _ real. Setting the gray scale of each row of pixels above the index min _ loc _ real and below the index max _ loc _ real of the two-dimensional original input image I to be 0, and setting the gray scale of other pixels to be 1 to obtain the template image I_mAs shown in fig. 6.

And fourthly, respectively using Gaussian derivative filters with sigma values of 4.5, 5, 5.5, 6, 6.5, 7, 7.5, 8, 8.5 and 9 to obtain different scale gradient maps of the original image I through non-maximum value suppression and dual-threshold processing, namely different scale canny edge operator processing. Taking a gradient map with a sigma value of 6.5 as a main image I_c. Using topological transmission method to transmit main image I_cComparing with other scale gradient maps to connect the main image I_cEdge lines where the gradient of (a) is discontinuous. In the topology transmission process, pixel points with gray values of 1 in two binary images are set as 2, pixel points with gray values of 0 are set as 0, pixel points with different gray values are set as 1, a new two-dimensional array is obtained, and the two-dimensional array is set as an image I _ find. For pixel point with gray value of 1 in I _ find, 10 pixels are arrangedIn the searching range of the pixel points, the pixel connection with the gray value not being 0 is searched according to the 8-connection rule, that is, for a pixel point with the gray value being 1, in two directions in 8 directions satisfying the 8-connection rule in the searching range of 10 pixel points, the pixel point with the gray value being 2 can be found along the pixel with the gray value not being 0 according to the 8-connection rule, and then on the main image, the gray value of the pixel point with the same coordinate as the pixel point with the gray value being 1 on the I _ find is set as 1. Finally obtaining a gradient map I with continuous edges_tAs shown in fig. 7. Map of gradients I_tMultiplying by the template image I_mObtaining a gradient map I of the approximate location of the epidermal-dermal boundary_tm. Carrying out axial derivation with sigma of Gaussian kernel of 6.5 on the original input image I to obtain an axial gradient image, and multiplying the axial gradient image by the template image I_mObtaining an axial gradient map I of the approximate location of the epidermal-dermal boundary_tzm. In the gradient map I_tmIn (1), will be plotted against the axial gradient I_tzmThe pixel with the same maximum value coordinate in each column is set as 1, and the points are connected by a straight line. In the gradient map I_tmThe pixel of each column of (a) corresponding to the air-skin boundary z _ surface is set to 1, and the points are connected by a straight line. Finally obtaining an image I containing the air-epidermis boundary, the canny gradient information and the axial gradient information_tmsAs shown in fig. 8.

Fifthly, image I_tmsThe pixel of the index min _ loc _ real corresponding to the middle coordinate is set as a starting point, i.e., a seed filled by the seed, a first closed region is found according to the four-connectivity rule, and morphological closing operation is performed on the region to obtain a region where the epidermis layer is located, as shown in fig. 9. Then, the boundary tracing operator is used to extract the edge, and the boundary of the cortical layer on the original image is obtained, as shown in fig. 10.

The above description is only one specific embodiment of the present invention, and the embodiment is only used to illustrate the technical solution of the present invention and not to limit the present invention. The technical solutions available to those skilled in the art through logical analysis, reasoning or limited experiments according to the concepts of the present invention are all within the scope of the present invention.

Claims

1. A method for automatically identifying an epidermal layer in an optical coherence tomography image of the skin, the method comprising the steps of:

firstly, acquiring an interference signal of skin tissue by using an OCT system, reconstructing the image to obtain a two-dimensional intensity image, carrying out logarithmic transformation on the two-dimensional intensity image to obtain a two-dimensional original input image I, defining a first pixel of a first row of the image as an origin, defining a horizontal coordinate x on the right and the left as a transverse scanning distance, defining a vertical coordinate z on the lower and the left as a longitudinal imaging depth, wherein the two-dimensional original input image I has n columns in total, and n is a positive integer;

secondly, axial derivation based on Gaussian kernel is carried out on the two-dimensional original input image I to obtain an axial gradient image I_zTaking an axial gradient map I_zThe ordinate of each column of maximum value pixel forms a one-dimensional array z _ surface containing n elements, the array elements are z _ surface _1, z _ surface _2, … …, z _ surface _ m, … … and z _ surface _ n, wherein m is an element index and is a positive integer; for the index M of the element in the array z _ surface that differs from the previous element by more than M, at I_zThe mth column takes a pixel point with the ordinate as z _ surface _ (M-1) as the center, the maximum gradient position in a certain number of pixel ranges in the depth direction is searched again, a new ordinate z _ surface _ M (new) is obtained, and the value of M can be set to be 1 to 5; performing the same processing on indexes of elements, the difference value of which is greater than M, in the z _ surface to obtain a new one-dimensional array z _ surface _ new; the values of n elements in the one-dimensional array z _ surface _ new are regarded as the vertical coordinates of n pixels on the air-epidermis boundary in the two-dimensional original input image I, and the indexes of the n elements are regarded as the horizontal coordinates of the n pixels on the air-epidermis boundary;

taking out a certain number of continuous pixels of each pixel on the air-epidermis boundary line in the vertical upward direction and the downward direction from the two-dimensional original input image I, storing the pixels in a two-dimensional array, and recording the image represented by the array as I_roiDefining the first pixel of the first row of the image as the origin and the horizontal right as the abscissa x₁Vertically downwards as ordinate z₁(ii) a First pair of images I_roiPerforming Gaussian filtering, and performing sliding window on each columnCarrying out mouth processing to obtain an average signal of each row, and recording the average signal as a one-dimensional array avr; performing one-dimensional linear filtering on the avr of each row to obtain an array, and recording a maximum value index behind a first maximum value index loc1 in the array as maxloc; the intensity of the array avr is inverted, namely the intensity value is subtracted from 255, and the result is recorded as an array avr 2; recording the index of the first maximum value of the array avr2 after the index loc1 as minloc, and if the first maximum value is the maximum value of the array avr2 between loc1 and maxloc, moving the index minloc upwards by a certain number of pixels to obtain a new index minloc; respectively forming a one-dimensional array min _ loc and a one-dimensional array max _ loc by minloc and maxloc in each row; the difference is made between the elements in the array max _ loc and the average value of all the elements in the array, and the value of the element with the difference value larger than M2 is changed into an image I_roiThe image width of (d); for with I_roiImage ordinate z₁Respectively finding corresponding array min _ loc _ real and max _ loc _ real elements which take the vertical coordinate z of the two-dimensional original input image I as the reference; setting the gray value of pixels above the index min _ loc _ real and below the index max _ loc _ real corresponding to each column of the two-dimensional original input image I as 0, and setting the gray values of other pixels as 1 to obtain the template image I_m；

Fourthly, using at least two edge detection operators with different scales to obtain gradient maps with different scales of the two-dimensional original input image I, and taking the gradient map with a certain scale as a main image I_cUsing a topology transfer method to transfer the primary image I_cComparing with other scale gradient maps to connect the main image I_cThe edge line with discontinuous gradient is obtained to obtain a gradient map I with continuous edge line_t(ii) a Map of gradients I_tMultiplying by the template image I_mObtaining a gradient map I of the approximate location of the epidermal-dermal boundary_tm(ii) a Carrying out axial derivation based on Gaussian kernel on an original input image I to obtain an axial gradient image, and multiplying the axial gradient image by a template image I_mObtaining an axial gradient map I of the approximate location of the epidermal-dermal boundary_tzm(ii) a In the gradient map I_tmIn (1), will be plotted against the axial gradient I_tzmSetting the pixel with the same maximum value coordinate as 1 in each column, and aligning the pointsWire connection; in the gradient map I_tmWherein the pixel of each row corresponding to the air-epidermis boundary z _ surface _ new is set as 1, and the points are connected by straight lines to finally obtain the image I_tms；

Fifthly, image I_tmsAnd (3) setting the pixel of the index min _ loc _ real corresponding to the middle coordinate as a starting point, namely the seed filled by the seed, searching a first closed region according to a four-connection rule, performing morphological closed operation on the region to obtain a region where the epidermis layer is located, and extracting the edge by using a boundary tracking operator to obtain a boundary line of the epidermis layer on the two-dimensional original input image.

2. The method of claim 1, wherein the number of pixels and M2 is adjustable according to different OCT system and different skin image.

3. The method according to claim 1, wherein the sliding window is a one-dimensional array avr formed by averaging each row of the data obtained by taking the data of a window with a certain width around each column of the image as a center, the number of rows of the window is equal to the number of rows of the image, and the number of columns is smaller than the number of columns of the image, and if the distance between the row and the edge of the image is smaller than a, the column is the leftmost or rightmost position of the window.

4. The method according to claim 1, wherein the edge detection operator is a canny edge operator based on Gaussian derivation.

5. The method according to claim 1, wherein the different scales set different "sigma" values in the Gaussian derivative.

6. The method for automatically identifying the epidermis layer in the skin optical coherence tomography image according to claim 1, wherein the topology transferring method is to establish a new two-dimensional image I _ find, and to set the gray of the pixel with gray value of 1 in the two compared binary images to 2 at the same coordinate position of I _ find, the gray of the pixel with gray value of 0 to 0 at the same coordinate position of I _ find, and the gray of the pixel with different gray value to 1 at the same coordinate position of I _ find; in an image I _ find, for a pixel point with a gray value of 1, in a set search range, searching for a pixel point with a gray value not being 0 according to an 8-connection rule for connection, namely, for a pixel point with a gray value of 1, in 8 directions meeting 8-connection in the set search range, a pixel point with a gray value of 2 can be found along the pixel point with a gray value not being 0 according to the 8-connection rule in two directions, and then on a main image, the gray value of the pixel point with the same coordinate as the pixel point with the gray value being 1 on the I _ find is set to be 1.